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 // If the VM might have been created on the primordial thread, we need to resolve the
156 // primordial thread stack bounds and check if the current thread might be the
157 // primordial thread in places. If we know that the primordial thread is never used,
158 // such as when the VM was created by one of the standard java launchers, we can
159 // avoid this
160 static bool suppress_primordial_thread_resolution = false;
161
162 // For diagnostics to print a message once. see run_periodic_checks
163 static sigset_t check_signal_done;
164 static bool check_signals = true;
165
166 // Signal number used to suspend/resume a thread
167
168 // do not use any signal number less than SIGSEGV, see 4355769
169 static int SR_signum = SIGUSR2;
170 sigset_t SR_sigset;
171
172 // utility functions
173
174 static int SR_initialize();
175
176 julong os::available_memory() {
177 return Linux::available_memory();
178 }
179
180 julong os::Linux::available_memory() {
181 // values in struct sysinfo are "unsigned long"
182 struct sysinfo si;
183 julong avail_mem;
184
185 if (OSContainer::is_containerized()) {
186 jlong mem_limit, mem_usage;
187 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
188 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
189 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
190 }
191 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
192 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
193 }
194 if (mem_limit > 0 && mem_usage > 0 ) {
195 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
196 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
197 return avail_mem;
198 }
199 }
200
201 sysinfo(&si);
202 avail_mem = (julong)si.freeram * si.mem_unit;
203 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
204 return avail_mem;
205 }
206
207 julong os::physical_memory() {
208 jlong phys_mem = 0;
209 if (OSContainer::is_containerized()) {
210 jlong mem_limit;
211 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
212 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
213 return mem_limit;
214 }
215 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
216 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
217 }
218
219 phys_mem = Linux::physical_memory();
220 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
221 return phys_mem;
222 }
223
224 // Return true if user is running as root.
225
226 bool os::have_special_privileges() {
227 static bool init = false;
228 static bool privileges = false;
229 if (!init) {
230 privileges = (getuid() != geteuid()) || (getgid() != getegid());
231 init = true;
232 }
233 return privileges;
234 }
235
236
237 #ifndef SYS_gettid
238 // i386: 224, ia64: 1105, amd64: 186, sparc 143
239 #ifdef __ia64__
240 #define SYS_gettid 1105
241 #else
242 #ifdef __i386__
243 #define SYS_gettid 224
244 #else
245 #ifdef __amd64__
246 #define SYS_gettid 186
247 #else
248 #ifdef __sparc__
249 #define SYS_gettid 143
250 #else
251 #error define gettid for the arch
252 #endif
253 #endif
254 #endif
255 #endif
256 #endif
257
258
259 // pid_t gettid()
260 //
261 // Returns the kernel thread id of the currently running thread. Kernel
262 // thread id is used to access /proc.
263 pid_t os::Linux::gettid() {
264 int rslt = syscall(SYS_gettid);
265 assert(rslt != -1, "must be."); // old linuxthreads implementation?
266 return (pid_t)rslt;
267 }
268
269 // Most versions of linux have a bug where the number of processors are
270 // determined by looking at the /proc file system. In a chroot environment,
271 // the system call returns 1. This causes the VM to act as if it is
272 // a single processor and elide locking (see is_MP() call).
273 static bool unsafe_chroot_detected = false;
274 static const char *unstable_chroot_error = "/proc file system not found.\n"
275 "Java may be unstable running multithreaded in a chroot "
276 "environment on Linux when /proc filesystem is not mounted.";
277
278 void os::Linux::initialize_system_info() {
279 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
280 if (processor_count() == 1) {
281 pid_t pid = os::Linux::gettid();
282 char fname[32];
283 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
284 FILE *fp = fopen(fname, "r");
285 if (fp == NULL) {
286 unsafe_chroot_detected = true;
287 } else {
288 fclose(fp);
289 }
290 }
291 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
292 assert(processor_count() > 0, "linux error");
293 }
294
295 void os::init_system_properties_values() {
296 // The next steps are taken in the product version:
297 //
298 // Obtain the JAVA_HOME value from the location of libjvm.so.
299 // This library should be located at:
300 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
301 //
302 // If "/jre/lib/" appears at the right place in the path, then we
303 // assume libjvm.so is installed in a JDK and we use this path.
304 //
305 // Otherwise exit with message: "Could not create the Java virtual machine."
306 //
307 // The following extra steps are taken in the debugging version:
308 //
309 // If "/jre/lib/" does NOT appear at the right place in the path
310 // instead of exit check for $JAVA_HOME environment variable.
311 //
312 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
313 // then we append a fake suffix "hotspot/libjvm.so" to this path so
314 // it looks like libjvm.so is installed there
315 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
316 //
317 // Otherwise exit.
318 //
319 // Important note: if the location of libjvm.so changes this
320 // code needs to be changed accordingly.
321
322 // See ld(1):
323 // The linker uses the following search paths to locate required
324 // shared libraries:
325 // 1: ...
326 // ...
327 // 7: The default directories, normally /lib and /usr/lib.
328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
329 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
330 #else
331 #define DEFAULT_LIBPATH "/lib:/usr/lib"
332 #endif
333
334 // Base path of extensions installed on the system.
335 #define SYS_EXT_DIR "/usr/java/packages"
336 #define EXTENSIONS_DIR "/lib/ext"
337
338 // Buffer that fits several sprintfs.
339 // Note that the space for the colon and the trailing null are provided
340 // by the nulls included by the sizeof operator.
341 const size_t bufsize =
342 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
343 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
344 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
345
346 // sysclasspath, java_home, dll_dir
347 {
348 char *pslash;
349 os::jvm_path(buf, bufsize);
350
351 // Found the full path to libjvm.so.
352 // Now cut the path to <java_home>/jre if we can.
353 pslash = strrchr(buf, '/');
354 if (pslash != NULL) {
355 *pslash = '\0'; // Get rid of /libjvm.so.
356 }
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
360 }
361 Arguments::set_dll_dir(buf);
362
363 if (pslash != NULL) {
364 pslash = strrchr(buf, '/');
365 if (pslash != NULL) {
366 *pslash = '\0'; // Get rid of /lib.
367 }
368 }
369 Arguments::set_java_home(buf);
370 set_boot_path('/', ':');
371 }
372
373 // Where to look for native libraries.
374 //
375 // Note: Due to a legacy implementation, most of the library path
376 // is set in the launcher. This was to accomodate linking restrictions
377 // on legacy Linux implementations (which are no longer supported).
378 // Eventually, all the library path setting will be done here.
379 //
380 // However, to prevent the proliferation of improperly built native
381 // libraries, the new path component /usr/java/packages is added here.
382 // Eventually, all the library path setting will be done here.
383 {
384 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
385 // should always exist (until the legacy problem cited above is
386 // addressed).
387 const char *v = ::getenv("LD_LIBRARY_PATH");
388 const char *v_colon = ":";
389 if (v == NULL) { v = ""; v_colon = ""; }
390 // That's +1 for the colon and +1 for the trailing '\0'.
391 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
392 strlen(v) + 1 +
393 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
394 mtInternal);
395 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
396 Arguments::set_library_path(ld_library_path);
397 FREE_C_HEAP_ARRAY(char, ld_library_path);
398 }
399
400 // Extensions directories.
401 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
402 Arguments::set_ext_dirs(buf);
403
404 FREE_C_HEAP_ARRAY(char, buf);
405
406 #undef DEFAULT_LIBPATH
407 #undef SYS_EXT_DIR
408 #undef EXTENSIONS_DIR
409 }
410
411 ////////////////////////////////////////////////////////////////////////////////
412 // breakpoint support
413
414 void os::breakpoint() {
415 BREAKPOINT;
416 }
417
418 extern "C" void breakpoint() {
419 // use debugger to set breakpoint here
420 }
421
422 ////////////////////////////////////////////////////////////////////////////////
423 // signal support
424
425 debug_only(static bool signal_sets_initialized = false);
426 static sigset_t unblocked_sigs, vm_sigs;
427
428 bool os::Linux::is_sig_ignored(int sig) {
429 struct sigaction oact;
430 sigaction(sig, (struct sigaction*)NULL, &oact);
431 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
432 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
433 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
434 return true;
435 } else {
436 return false;
437 }
438 }
439
440 void os::Linux::signal_sets_init() {
441 // Should also have an assertion stating we are still single-threaded.
442 assert(!signal_sets_initialized, "Already initialized");
443 // Fill in signals that are necessarily unblocked for all threads in
444 // the VM. Currently, we unblock the following signals:
445 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
446 // by -Xrs (=ReduceSignalUsage));
447 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
448 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
449 // the dispositions or masks wrt these signals.
450 // Programs embedding the VM that want to use the above signals for their
451 // own purposes must, at this time, use the "-Xrs" option to prevent
452 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
453 // (See bug 4345157, and other related bugs).
454 // In reality, though, unblocking these signals is really a nop, since
455 // these signals are not blocked by default.
456 sigemptyset(&unblocked_sigs);
457 sigaddset(&unblocked_sigs, SIGILL);
458 sigaddset(&unblocked_sigs, SIGSEGV);
459 sigaddset(&unblocked_sigs, SIGBUS);
460 sigaddset(&unblocked_sigs, SIGFPE);
461 #if defined(PPC64)
462 sigaddset(&unblocked_sigs, SIGTRAP);
463 #endif
464 sigaddset(&unblocked_sigs, SR_signum);
465
466 if (!ReduceSignalUsage) {
467 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
468 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
469 }
470 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
471 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
472 }
473 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
474 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
475 }
476 }
477 // Fill in signals that are blocked by all but the VM thread.
478 sigemptyset(&vm_sigs);
479 if (!ReduceSignalUsage) {
480 sigaddset(&vm_sigs, BREAK_SIGNAL);
481 }
482 debug_only(signal_sets_initialized = true);
483
484 }
485
486 // These are signals that are unblocked while a thread is running Java.
487 // (For some reason, they get blocked by default.)
488 sigset_t* os::Linux::unblocked_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &unblocked_sigs;
491 }
492
493 // These are the signals that are blocked while a (non-VM) thread is
494 // running Java. Only the VM thread handles these signals.
495 sigset_t* os::Linux::vm_signals() {
496 assert(signal_sets_initialized, "Not initialized");
497 return &vm_sigs;
498 }
499
500 void os::Linux::hotspot_sigmask(Thread* thread) {
501
502 //Save caller's signal mask before setting VM signal mask
503 sigset_t caller_sigmask;
504 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
505
506 OSThread* osthread = thread->osthread();
507 osthread->set_caller_sigmask(caller_sigmask);
508
509 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
510
511 if (!ReduceSignalUsage) {
512 if (thread->is_VM_thread()) {
513 // Only the VM thread handles BREAK_SIGNAL ...
514 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
515 } else {
516 // ... all other threads block BREAK_SIGNAL
517 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
518 }
519 }
520 }
521
522 //////////////////////////////////////////////////////////////////////////////
523 // detecting pthread library
524
525 void os::Linux::libpthread_init() {
526 // Save glibc and pthread version strings.
527 #if !defined(_CS_GNU_LIBC_VERSION) || \
528 !defined(_CS_GNU_LIBPTHREAD_VERSION)
529 #error "glibc too old (< 2.3.2)"
530 #endif
531
532 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
533 assert(n > 0, "cannot retrieve glibc version");
534 char *str = (char *)malloc(n, mtInternal);
535 confstr(_CS_GNU_LIBC_VERSION, str, n);
536 os::Linux::set_glibc_version(str);
537
538 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
539 assert(n > 0, "cannot retrieve pthread version");
540 str = (char *)malloc(n, mtInternal);
541 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
542 os::Linux::set_libpthread_version(str);
543 }
544
545 /////////////////////////////////////////////////////////////////////////////
546 // thread stack expansion
547
548 // os::Linux::manually_expand_stack() takes care of expanding the thread
549 // stack. Note that this is normally not needed: pthread stacks allocate
550 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
551 // committed. Therefore it is not necessary to expand the stack manually.
552 //
553 // Manually expanding the stack was historically needed on LinuxThreads
554 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
555 // it is kept to deal with very rare corner cases:
556 //
557 // For one, user may run the VM on an own implementation of threads
558 // whose stacks are - like the old LinuxThreads - implemented using
559 // mmap(MAP_GROWSDOWN).
560 //
561 // Also, this coding may be needed if the VM is running on the primordial
562 // thread. Normally we avoid running on the primordial thread; however,
563 // user may still invoke the VM on the primordial thread.
564 //
565 // The following historical comment describes the details about running
566 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
567
568
569 // Force Linux kernel to expand current thread stack. If "bottom" is close
570 // to the stack guard, caller should block all signals.
571 //
572 // MAP_GROWSDOWN:
573 // A special mmap() flag that is used to implement thread stacks. It tells
574 // kernel that the memory region should extend downwards when needed. This
575 // allows early versions of LinuxThreads to only mmap the first few pages
576 // when creating a new thread. Linux kernel will automatically expand thread
577 // stack as needed (on page faults).
578 //
579 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
580 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
581 // region, it's hard to tell if the fault is due to a legitimate stack
582 // access or because of reading/writing non-exist memory (e.g. buffer
583 // overrun). As a rule, if the fault happens below current stack pointer,
584 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
585 // application (see Linux kernel fault.c).
586 //
587 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
588 // stack overflow detection.
589 //
590 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
591 // not use MAP_GROWSDOWN.
592 //
593 // To get around the problem and allow stack banging on Linux, we need to
594 // manually expand thread stack after receiving the SIGSEGV.
595 //
596 // There are two ways to expand thread stack to address "bottom", we used
597 // both of them in JVM before 1.5:
598 // 1. adjust stack pointer first so that it is below "bottom", and then
599 // touch "bottom"
600 // 2. mmap() the page in question
601 //
602 // Now alternate signal stack is gone, it's harder to use 2. For instance,
603 // if current sp is already near the lower end of page 101, and we need to
604 // call mmap() to map page 100, it is possible that part of the mmap() frame
605 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
606 // That will destroy the mmap() frame and cause VM to crash.
607 //
608 // The following code works by adjusting sp first, then accessing the "bottom"
609 // page to force a page fault. Linux kernel will then automatically expand the
610 // stack mapping.
611 //
612 // _expand_stack_to() assumes its frame size is less than page size, which
613 // should always be true if the function is not inlined.
614
615 static void NOINLINE _expand_stack_to(address bottom) {
616 address sp;
617 size_t size;
618 volatile char *p;
619
620 // Adjust bottom to point to the largest address within the same page, it
621 // gives us a one-page buffer if alloca() allocates slightly more memory.
622 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
623 bottom += os::Linux::page_size() - 1;
624
625 // sp might be slightly above current stack pointer; if that's the case, we
626 // will alloca() a little more space than necessary, which is OK. Don't use
627 // os::current_stack_pointer(), as its result can be slightly below current
628 // stack pointer, causing us to not alloca enough to reach "bottom".
629 sp = (address)&sp;
630
631 if (sp > bottom) {
632 size = sp - bottom;
633 p = (volatile char *)alloca(size);
634 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
635 p[0] = '\0';
636 }
637 }
638
639 void os::Linux::expand_stack_to(address bottom) {
640 _expand_stack_to(bottom);
641 }
642
643 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
644 assert(t!=NULL, "just checking");
645 assert(t->osthread()->expanding_stack(), "expand should be set");
646 assert(t->stack_base() != NULL, "stack_base was not initialized");
647
648 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
649 sigset_t mask_all, old_sigset;
650 sigfillset(&mask_all);
651 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
652 _expand_stack_to(addr);
653 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
654 return true;
655 }
656 return false;
657 }
658
659 //////////////////////////////////////////////////////////////////////////////
660 // create new thread
661
662 // Thread start routine for all newly created threads
663 static void *thread_native_entry(Thread *thread) {
664 // Try to randomize the cache line index of hot stack frames.
665 // This helps when threads of the same stack traces evict each other's
666 // cache lines. The threads can be either from the same JVM instance, or
667 // from different JVM instances. The benefit is especially true for
668 // processors with hyperthreading technology.
669 static int counter = 0;
670 int pid = os::current_process_id();
671 alloca(((pid ^ counter++) & 7) * 128);
672
673 thread->initialize_thread_current();
674
675 OSThread* osthread = thread->osthread();
676 Monitor* sync = osthread->startThread_lock();
677
678 osthread->set_thread_id(os::current_thread_id());
679
680 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
681 os::current_thread_id(), (uintx) pthread_self());
682
683 if (UseNUMA) {
684 int lgrp_id = os::numa_get_group_id();
685 if (lgrp_id != -1) {
686 thread->set_lgrp_id(lgrp_id);
687 }
688 }
689 // initialize signal mask for this thread
690 os::Linux::hotspot_sigmask(thread);
691
692 // initialize floating point control register
693 os::Linux::init_thread_fpu_state();
694
695 // handshaking with parent thread
696 {
697 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
698
699 // notify parent thread
700 osthread->set_state(INITIALIZED);
701 sync->notify_all();
702
703 // wait until os::start_thread()
704 while (osthread->get_state() == INITIALIZED) {
705 sync->wait(Mutex::_no_safepoint_check_flag);
706 }
707 }
708
709 // call one more level start routine
710 thread->run();
711
712 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
713 os::current_thread_id(), (uintx) pthread_self());
714
715 // If a thread has not deleted itself ("delete this") as part of its
716 // termination sequence, we have to ensure thread-local-storage is
717 // cleared before we actually terminate. No threads should ever be
718 // deleted asynchronously with respect to their termination.
719 if (Thread::current_or_null_safe() != NULL) {
720 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
721 thread->clear_thread_current();
722 }
723
724 return 0;
725 }
726
727 bool os::create_thread(Thread* thread, ThreadType thr_type,
728 size_t req_stack_size) {
729 assert(thread->osthread() == NULL, "caller responsible");
730
731 // Allocate the OSThread object
732 OSThread* osthread = new OSThread(NULL, NULL);
733 if (osthread == NULL) {
734 return false;
735 }
736
737 // set the correct thread state
738 osthread->set_thread_type(thr_type);
739
740 // Initial state is ALLOCATED but not INITIALIZED
741 osthread->set_state(ALLOCATED);
742
743 thread->set_osthread(osthread);
744
745 // init thread attributes
746 pthread_attr_t attr;
747 pthread_attr_init(&attr);
748 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
749
750 // Calculate stack size if it's not specified by caller.
751 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
752 // In the Linux NPTL pthread implementation the guard size mechanism
753 // is not implemented properly. The posix standard requires adding
754 // the size of the guard pages to the stack size, instead Linux
755 // takes the space out of 'stacksize'. Thus we adapt the requested
756 // stack_size by the size of the guard pages to mimick proper
757 // behaviour. However, be careful not to end up with a size
758 // of zero due to overflow. Don't add the guard page in that case.
759 size_t guard_size = os::Linux::default_guard_size(thr_type);
760 if (stack_size <= SIZE_MAX - guard_size) {
761 stack_size += guard_size;
762 }
763 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
764
765 int status = pthread_attr_setstacksize(&attr, stack_size);
766 assert_status(status == 0, status, "pthread_attr_setstacksize");
767
768 // Configure glibc guard page.
769 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
770
771 ThreadState state;
772
773 {
774 pthread_t tid;
775 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
776
777 char buf[64];
778 if (ret == 0) {
779 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
780 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
781 } else {
782 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
783 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
784 }
785
786 pthread_attr_destroy(&attr);
787
788 if (ret != 0) {
789 // Need to clean up stuff we've allocated so far
790 thread->set_osthread(NULL);
791 delete osthread;
792 return false;
793 }
794
795 // Store pthread info into the OSThread
796 osthread->set_pthread_id(tid);
797
798 // Wait until child thread is either initialized or aborted
799 {
800 Monitor* sync_with_child = osthread->startThread_lock();
801 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
802 while ((state = osthread->get_state()) == ALLOCATED) {
803 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
804 }
805 }
806 }
807
808 // Aborted due to thread limit being reached
809 if (state == ZOMBIE) {
810 thread->set_osthread(NULL);
811 delete osthread;
812 return false;
813 }
814
815 // The thread is returned suspended (in state INITIALIZED),
816 // and is started higher up in the call chain
817 assert(state == INITIALIZED, "race condition");
818 return true;
819 }
820
821 /////////////////////////////////////////////////////////////////////////////
822 // attach existing thread
823
824 // bootstrap the main thread
825 bool os::create_main_thread(JavaThread* thread) {
826 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
827 return create_attached_thread(thread);
828 }
829
830 bool os::create_attached_thread(JavaThread* thread) {
831 #ifdef ASSERT
832 thread->verify_not_published();
833 #endif
834
835 // Allocate the OSThread object
836 OSThread* osthread = new OSThread(NULL, NULL);
837
838 if (osthread == NULL) {
839 return false;
840 }
841
842 // Store pthread info into the OSThread
843 osthread->set_thread_id(os::Linux::gettid());
844 osthread->set_pthread_id(::pthread_self());
845
846 // initialize floating point control register
847 os::Linux::init_thread_fpu_state();
848
849 // Initial thread state is RUNNABLE
850 osthread->set_state(RUNNABLE);
851
852 thread->set_osthread(osthread);
853
854 if (UseNUMA) {
855 int lgrp_id = os::numa_get_group_id();
856 if (lgrp_id != -1) {
857 thread->set_lgrp_id(lgrp_id);
858 }
859 }
860
861 if (os::is_primordial_thread()) {
862 // If current thread is primordial thread, its stack is mapped on demand,
863 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
864 // the entire stack region to avoid SEGV in stack banging.
865 // It is also useful to get around the heap-stack-gap problem on SuSE
866 // kernel (see 4821821 for details). We first expand stack to the top
867 // of yellow zone, then enable stack yellow zone (order is significant,
868 // enabling yellow zone first will crash JVM on SuSE Linux), so there
869 // is no gap between the last two virtual memory regions.
870
871 JavaThread *jt = (JavaThread *)thread;
872 address addr = jt->stack_reserved_zone_base();
873 assert(addr != NULL, "initialization problem?");
874 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
875
876 osthread->set_expanding_stack();
877 os::Linux::manually_expand_stack(jt, addr);
878 osthread->clear_expanding_stack();
879 }
880
881 // initialize signal mask for this thread
882 // and save the caller's signal mask
883 os::Linux::hotspot_sigmask(thread);
884
885 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
886 os::current_thread_id(), (uintx) pthread_self());
887
888 return true;
889 }
890
891 void os::pd_start_thread(Thread* thread) {
892 OSThread * osthread = thread->osthread();
893 assert(osthread->get_state() != INITIALIZED, "just checking");
894 Monitor* sync_with_child = osthread->startThread_lock();
895 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
896 sync_with_child->notify();
897 }
898
899 // Free Linux resources related to the OSThread
900 void os::free_thread(OSThread* osthread) {
901 assert(osthread != NULL, "osthread not set");
902
903 // We are told to free resources of the argument thread,
904 // but we can only really operate on the current thread.
905 assert(Thread::current()->osthread() == osthread,
906 "os::free_thread but not current thread");
907
908 #ifdef ASSERT
909 sigset_t current;
910 sigemptyset(¤t);
911 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
912 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
913 #endif
914
915 // Restore caller's signal mask
916 sigset_t sigmask = osthread->caller_sigmask();
917 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
918
919 delete osthread;
920 }
921
922 //////////////////////////////////////////////////////////////////////////////
923 // primordial thread
924
925 // Check if current thread is the primordial thread, similar to Solaris thr_main.
926 bool os::is_primordial_thread(void) {
927 if (suppress_primordial_thread_resolution) {
928 return false;
929 }
930 char dummy;
931 // If called before init complete, thread stack bottom will be null.
932 // Can be called if fatal error occurs before initialization.
933 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
934 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
935 os::Linux::initial_thread_stack_size() != 0,
936 "os::init did not locate primordial thread's stack region");
937 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
938 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
939 os::Linux::initial_thread_stack_size()) {
940 return true;
941 } else {
942 return false;
943 }
944 }
945
946 // Find the virtual memory area that contains addr
947 static bool find_vma(address addr, address* vma_low, address* vma_high) {
948 FILE *fp = fopen("/proc/self/maps", "r");
949 if (fp) {
950 address low, high;
951 while (!feof(fp)) {
952 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
953 if (low <= addr && addr < high) {
954 if (vma_low) *vma_low = low;
955 if (vma_high) *vma_high = high;
956 fclose(fp);
957 return true;
958 }
959 }
960 for (;;) {
961 int ch = fgetc(fp);
962 if (ch == EOF || ch == (int)'\n') break;
963 }
964 }
965 fclose(fp);
966 }
967 return false;
968 }
969
970 // Locate primordial thread stack. This special handling of primordial thread stack
971 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
972 // bogus value for the primordial process thread. While the launcher has created
973 // the VM in a new thread since JDK 6, we still have to allow for the use of the
974 // JNI invocation API from a primordial thread.
975 void os::Linux::capture_initial_stack(size_t max_size) {
976
977 // max_size is either 0 (which means accept OS default for thread stacks) or
978 // a user-specified value known to be at least the minimum needed. If we
979 // are actually on the primordial thread we can make it appear that we have a
980 // smaller max_size stack by inserting the guard pages at that location. But we
981 // cannot do anything to emulate a larger stack than what has been provided by
982 // the OS or threading library. In fact if we try to use a stack greater than
983 // what is set by rlimit then we will crash the hosting process.
984
985 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
986 // If this is "unlimited" then it will be a huge value.
987 struct rlimit rlim;
988 getrlimit(RLIMIT_STACK, &rlim);
989 size_t stack_size = rlim.rlim_cur;
990
991 // 6308388: a bug in ld.so will relocate its own .data section to the
992 // lower end of primordial stack; reduce ulimit -s value a little bit
993 // so we won't install guard page on ld.so's data section.
994 // But ensure we don't underflow the stack size - allow 1 page spare
995 if (stack_size >= (size_t)(3 * page_size())) {
996 stack_size -= 2 * page_size();
997 }
998
999 // Try to figure out where the stack base (top) is. This is harder.
1000 //
1001 // When an application is started, glibc saves the initial stack pointer in
1002 // a global variable "__libc_stack_end", which is then used by system
1003 // libraries. __libc_stack_end should be pretty close to stack top. The
1004 // variable is available since the very early days. However, because it is
1005 // a private interface, it could disappear in the future.
1006 //
1007 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1008 // to __libc_stack_end, it is very close to stack top, but isn't the real
1009 // stack top. Note that /proc may not exist if VM is running as a chroot
1010 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1011 // /proc/<pid>/stat could change in the future (though unlikely).
1012 //
1013 // We try __libc_stack_end first. If that doesn't work, look for
1014 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1015 // as a hint, which should work well in most cases.
1016
1017 uintptr_t stack_start;
1018
1019 // try __libc_stack_end first
1020 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1021 if (p && *p) {
1022 stack_start = *p;
1023 } else {
1024 // see if we can get the start_stack field from /proc/self/stat
1025 FILE *fp;
1026 int pid;
1027 char state;
1028 int ppid;
1029 int pgrp;
1030 int session;
1031 int nr;
1032 int tpgrp;
1033 unsigned long flags;
1034 unsigned long minflt;
1035 unsigned long cminflt;
1036 unsigned long majflt;
1037 unsigned long cmajflt;
1038 unsigned long utime;
1039 unsigned long stime;
1040 long cutime;
1041 long cstime;
1042 long prio;
1043 long nice;
1044 long junk;
1045 long it_real;
1046 uintptr_t start;
1047 uintptr_t vsize;
1048 intptr_t rss;
1049 uintptr_t rsslim;
1050 uintptr_t scodes;
1051 uintptr_t ecode;
1052 int i;
1053
1054 // Figure what the primordial thread stack base is. Code is inspired
1055 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1056 // followed by command name surrounded by parentheses, state, etc.
1057 char stat[2048];
1058 int statlen;
1059
1060 fp = fopen("/proc/self/stat", "r");
1061 if (fp) {
1062 statlen = fread(stat, 1, 2047, fp);
1063 stat[statlen] = '\0';
1064 fclose(fp);
1065
1066 // Skip pid and the command string. Note that we could be dealing with
1067 // weird command names, e.g. user could decide to rename java launcher
1068 // to "java 1.4.2 :)", then the stat file would look like
1069 // 1234 (java 1.4.2 :)) R ... ...
1070 // We don't really need to know the command string, just find the last
1071 // occurrence of ")" and then start parsing from there. See bug 4726580.
1072 char * s = strrchr(stat, ')');
1073
1074 i = 0;
1075 if (s) {
1076 // Skip blank chars
1077 do { s++; } while (s && isspace(*s));
1078
1079 #define _UFM UINTX_FORMAT
1080 #define _DFM INTX_FORMAT
1081
1082 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1083 // 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
1084 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,
1085 &state, // 3 %c
1086 &ppid, // 4 %d
1087 &pgrp, // 5 %d
1088 &session, // 6 %d
1089 &nr, // 7 %d
1090 &tpgrp, // 8 %d
1091 &flags, // 9 %lu
1092 &minflt, // 10 %lu
1093 &cminflt, // 11 %lu
1094 &majflt, // 12 %lu
1095 &cmajflt, // 13 %lu
1096 &utime, // 14 %lu
1097 &stime, // 15 %lu
1098 &cutime, // 16 %ld
1099 &cstime, // 17 %ld
1100 &prio, // 18 %ld
1101 &nice, // 19 %ld
1102 &junk, // 20 %ld
1103 &it_real, // 21 %ld
1104 &start, // 22 UINTX_FORMAT
1105 &vsize, // 23 UINTX_FORMAT
1106 &rss, // 24 INTX_FORMAT
1107 &rsslim, // 25 UINTX_FORMAT
1108 &scodes, // 26 UINTX_FORMAT
1109 &ecode, // 27 UINTX_FORMAT
1110 &stack_start); // 28 UINTX_FORMAT
1111 }
1112
1113 #undef _UFM
1114 #undef _DFM
1115
1116 if (i != 28 - 2) {
1117 assert(false, "Bad conversion from /proc/self/stat");
1118 // product mode - assume we are the primordial thread, good luck in the
1119 // embedded case.
1120 warning("Can't detect primordial thread stack location - bad conversion");
1121 stack_start = (uintptr_t) &rlim;
1122 }
1123 } else {
1124 // For some reason we can't open /proc/self/stat (for example, running on
1125 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1126 // most cases, so don't abort:
1127 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1128 stack_start = (uintptr_t) &rlim;
1129 }
1130 }
1131
1132 // Now we have a pointer (stack_start) very close to the stack top, the
1133 // next thing to do is to figure out the exact location of stack top. We
1134 // can find out the virtual memory area that contains stack_start by
1135 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1136 // and its upper limit is the real stack top. (again, this would fail if
1137 // running inside chroot, because /proc may not exist.)
1138
1139 uintptr_t stack_top;
1140 address low, high;
1141 if (find_vma((address)stack_start, &low, &high)) {
1142 // success, "high" is the true stack top. (ignore "low", because initial
1143 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1144 stack_top = (uintptr_t)high;
1145 } else {
1146 // failed, likely because /proc/self/maps does not exist
1147 warning("Can't detect primordial thread stack location - find_vma failed");
1148 // best effort: stack_start is normally within a few pages below the real
1149 // stack top, use it as stack top, and reduce stack size so we won't put
1150 // guard page outside stack.
1151 stack_top = stack_start;
1152 stack_size -= 16 * page_size();
1153 }
1154
1155 // stack_top could be partially down the page so align it
1156 stack_top = align_up(stack_top, page_size());
1157
1158 // Allowed stack value is minimum of max_size and what we derived from rlimit
1159 if (max_size > 0) {
1160 _initial_thread_stack_size = MIN2(max_size, stack_size);
1161 } else {
1162 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1163 // clamp it at 8MB as we do on Solaris
1164 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1165 }
1166 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1167 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1168
1169 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1170
1171 if (log_is_enabled(Info, os, thread)) {
1172 // See if we seem to be on primordial process thread
1173 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1174 uintptr_t(&rlim) < stack_top;
1175
1176 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1177 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1178 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1179 stack_top, intptr_t(_initial_thread_stack_bottom));
1180 }
1181 }
1182
1183 ////////////////////////////////////////////////////////////////////////////////
1184 // time support
1185
1186 // Time since start-up in seconds to a fine granularity.
1187 // Used by VMSelfDestructTimer and the MemProfiler.
1188 double os::elapsedTime() {
1189
1190 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1191 }
1192
1193 jlong os::elapsed_counter() {
1194 return javaTimeNanos() - initial_time_count;
1195 }
1196
1197 jlong os::elapsed_frequency() {
1198 return NANOSECS_PER_SEC; // nanosecond resolution
1199 }
1200
1201 bool os::supports_vtime() { return true; }
1202 bool os::enable_vtime() { return false; }
1203 bool os::vtime_enabled() { return false; }
1204
1205 double os::elapsedVTime() {
1206 struct rusage usage;
1207 int retval = getrusage(RUSAGE_THREAD, &usage);
1208 if (retval == 0) {
1209 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);
1210 } else {
1211 // better than nothing, but not much
1212 return elapsedTime();
1213 }
1214 }
1215
1216 jlong os::javaTimeMillis() {
1217 timeval time;
1218 int status = gettimeofday(&time, NULL);
1219 assert(status != -1, "linux error");
1220 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1221 }
1222
1223 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1224 timeval time;
1225 int status = gettimeofday(&time, NULL);
1226 assert(status != -1, "linux error");
1227 seconds = jlong(time.tv_sec);
1228 nanos = jlong(time.tv_usec) * 1000;
1229 }
1230
1231
1232 #ifndef CLOCK_MONOTONIC
1233 #define CLOCK_MONOTONIC (1)
1234 #endif
1235
1236 void os::Linux::clock_init() {
1237 // we do dlopen's in this particular order due to bug in linux
1238 // dynamical loader (see 6348968) leading to crash on exit
1239 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1240 if (handle == NULL) {
1241 handle = dlopen("librt.so", RTLD_LAZY);
1242 }
1243
1244 if (handle) {
1245 int (*clock_getres_func)(clockid_t, struct timespec*) =
1246 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1247 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1248 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1249 if (clock_getres_func && clock_gettime_func) {
1250 // See if monotonic clock is supported by the kernel. Note that some
1251 // early implementations simply return kernel jiffies (updated every
1252 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1253 // for nano time (though the monotonic property is still nice to have).
1254 // It's fixed in newer kernels, however clock_getres() still returns
1255 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1256 // resolution for now. Hopefully as people move to new kernels, this
1257 // won't be a problem.
1258 struct timespec res;
1259 struct timespec tp;
1260 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1261 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1262 // yes, monotonic clock is supported
1263 _clock_gettime = clock_gettime_func;
1264 return;
1265 } else {
1266 // close librt if there is no monotonic clock
1267 dlclose(handle);
1268 }
1269 }
1270 }
1271 warning("No monotonic clock was available - timed services may " \
1272 "be adversely affected if the time-of-day clock changes");
1273 }
1274
1275 #ifndef SYS_clock_getres
1276 #if defined(X86) || defined(PPC64) || defined(S390)
1277 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1278 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1279 #else
1280 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1281 #define sys_clock_getres(x,y) -1
1282 #endif
1283 #else
1284 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1285 #endif
1286
1287 void os::Linux::fast_thread_clock_init() {
1288 if (!UseLinuxPosixThreadCPUClocks) {
1289 return;
1290 }
1291 clockid_t clockid;
1292 struct timespec tp;
1293 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1294 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1295
1296 // Switch to using fast clocks for thread cpu time if
1297 // the sys_clock_getres() returns 0 error code.
1298 // Note, that some kernels may support the current thread
1299 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1300 // returned by the pthread_getcpuclockid().
1301 // If the fast Posix clocks are supported then the sys_clock_getres()
1302 // must return at least tp.tv_sec == 0 which means a resolution
1303 // better than 1 sec. This is extra check for reliability.
1304
1305 if (pthread_getcpuclockid_func &&
1306 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1307 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1308 _supports_fast_thread_cpu_time = true;
1309 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1310 }
1311 }
1312
1313 jlong os::javaTimeNanos() {
1314 if (os::supports_monotonic_clock()) {
1315 struct timespec tp;
1316 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1317 assert(status == 0, "gettime error");
1318 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1319 return result;
1320 } else {
1321 timeval time;
1322 int status = gettimeofday(&time, NULL);
1323 assert(status != -1, "linux error");
1324 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1325 return 1000 * usecs;
1326 }
1327 }
1328
1329 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1330 if (os::supports_monotonic_clock()) {
1331 info_ptr->max_value = ALL_64_BITS;
1332
1333 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1334 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1335 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1336 } else {
1337 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1338 info_ptr->max_value = ALL_64_BITS;
1339
1340 // gettimeofday is a real time clock so it skips
1341 info_ptr->may_skip_backward = true;
1342 info_ptr->may_skip_forward = true;
1343 }
1344
1345 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1346 }
1347
1348 // Return the real, user, and system times in seconds from an
1349 // arbitrary fixed point in the past.
1350 bool os::getTimesSecs(double* process_real_time,
1351 double* process_user_time,
1352 double* process_system_time) {
1353 struct tms ticks;
1354 clock_t real_ticks = times(&ticks);
1355
1356 if (real_ticks == (clock_t) (-1)) {
1357 return false;
1358 } else {
1359 double ticks_per_second = (double) clock_tics_per_sec;
1360 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1361 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1362 *process_real_time = ((double) real_ticks) / ticks_per_second;
1363
1364 return true;
1365 }
1366 }
1367
1368
1369 char * os::local_time_string(char *buf, size_t buflen) {
1370 struct tm t;
1371 time_t long_time;
1372 time(&long_time);
1373 localtime_r(&long_time, &t);
1374 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1375 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1376 t.tm_hour, t.tm_min, t.tm_sec);
1377 return buf;
1378 }
1379
1380 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1381 return localtime_r(clock, res);
1382 }
1383
1384 ////////////////////////////////////////////////////////////////////////////////
1385 // runtime exit support
1386
1387 // Note: os::shutdown() might be called very early during initialization, or
1388 // called from signal handler. Before adding something to os::shutdown(), make
1389 // sure it is async-safe and can handle partially initialized VM.
1390 void os::shutdown() {
1391
1392 // allow PerfMemory to attempt cleanup of any persistent resources
1393 perfMemory_exit();
1394
1395 // needs to remove object in file system
1396 AttachListener::abort();
1397
1398 // flush buffered output, finish log files
1399 ostream_abort();
1400
1401 // Check for abort hook
1402 abort_hook_t abort_hook = Arguments::abort_hook();
1403 if (abort_hook != NULL) {
1404 abort_hook();
1405 }
1406
1407 }
1408
1409 // Note: os::abort() might be called very early during initialization, or
1410 // called from signal handler. Before adding something to os::abort(), make
1411 // sure it is async-safe and can handle partially initialized VM.
1412 void os::abort(bool dump_core, void* siginfo, const void* context) {
1413 os::shutdown();
1414 if (dump_core) {
1415 #ifndef PRODUCT
1416 fdStream out(defaultStream::output_fd());
1417 out.print_raw("Current thread is ");
1418 char buf[16];
1419 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1420 out.print_raw_cr(buf);
1421 out.print_raw_cr("Dumping core ...");
1422 #endif
1423 ::abort(); // dump core
1424 }
1425
1426 ::exit(1);
1427 }
1428
1429 // Die immediately, no exit hook, no abort hook, no cleanup.
1430 void os::die() {
1431 ::abort();
1432 }
1433
1434
1435 // This method is a copy of JDK's sysGetLastErrorString
1436 // from src/solaris/hpi/src/system_md.c
1437
1438 size_t os::lasterror(char *buf, size_t len) {
1439 if (errno == 0) return 0;
1440
1441 const char *s = os::strerror(errno);
1442 size_t n = ::strlen(s);
1443 if (n >= len) {
1444 n = len - 1;
1445 }
1446 ::strncpy(buf, s, n);
1447 buf[n] = '\0';
1448 return n;
1449 }
1450
1451 // thread_id is kernel thread id (similar to Solaris LWP id)
1452 intx os::current_thread_id() { return os::Linux::gettid(); }
1453 int os::current_process_id() {
1454 return ::getpid();
1455 }
1456
1457 // DLL functions
1458
1459 const char* os::dll_file_extension() { return ".so"; }
1460
1461 // This must be hard coded because it's the system's temporary
1462 // directory not the java application's temp directory, ala java.io.tmpdir.
1463 const char* os::get_temp_directory() { return "/tmp"; }
1464
1465 static bool file_exists(const char* filename) {
1466 struct stat statbuf;
1467 if (filename == NULL || strlen(filename) == 0) {
1468 return false;
1469 }
1470 return os::stat(filename, &statbuf) == 0;
1471 }
1472
1473 // check if addr is inside libjvm.so
1474 bool os::address_is_in_vm(address addr) {
1475 static address libjvm_base_addr;
1476 Dl_info dlinfo;
1477
1478 if (libjvm_base_addr == NULL) {
1479 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1480 libjvm_base_addr = (address)dlinfo.dli_fbase;
1481 }
1482 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1483 }
1484
1485 if (dladdr((void *)addr, &dlinfo) != 0) {
1486 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1487 }
1488
1489 return false;
1490 }
1491
1492 bool os::dll_address_to_function_name(address addr, char *buf,
1493 int buflen, int *offset,
1494 bool demangle) {
1495 // buf is not optional, but offset is optional
1496 assert(buf != NULL, "sanity check");
1497
1498 Dl_info dlinfo;
1499
1500 if (dladdr((void*)addr, &dlinfo) != 0) {
1501 // see if we have a matching symbol
1502 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1503 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1504 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1505 }
1506 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1507 return true;
1508 }
1509 // no matching symbol so try for just file info
1510 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1511 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1512 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1513 return true;
1514 }
1515 }
1516 }
1517
1518 buf[0] = '\0';
1519 if (offset != NULL) *offset = -1;
1520 return false;
1521 }
1522
1523 struct _address_to_library_name {
1524 address addr; // input : memory address
1525 size_t buflen; // size of fname
1526 char* fname; // output: library name
1527 address base; // library base addr
1528 };
1529
1530 static int address_to_library_name_callback(struct dl_phdr_info *info,
1531 size_t size, void *data) {
1532 int i;
1533 bool found = false;
1534 address libbase = NULL;
1535 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1536
1537 // iterate through all loadable segments
1538 for (i = 0; i < info->dlpi_phnum; i++) {
1539 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1540 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1541 // base address of a library is the lowest address of its loaded
1542 // segments.
1543 if (libbase == NULL || libbase > segbase) {
1544 libbase = segbase;
1545 }
1546 // see if 'addr' is within current segment
1547 if (segbase <= d->addr &&
1548 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1549 found = true;
1550 }
1551 }
1552 }
1553
1554 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1555 // so dll_address_to_library_name() can fall through to use dladdr() which
1556 // can figure out executable name from argv[0].
1557 if (found && info->dlpi_name && info->dlpi_name[0]) {
1558 d->base = libbase;
1559 if (d->fname) {
1560 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1561 }
1562 return 1;
1563 }
1564 return 0;
1565 }
1566
1567 bool os::dll_address_to_library_name(address addr, char* buf,
1568 int buflen, int* offset) {
1569 // buf is not optional, but offset is optional
1570 assert(buf != NULL, "sanity check");
1571
1572 Dl_info dlinfo;
1573 struct _address_to_library_name data;
1574
1575 // There is a bug in old glibc dladdr() implementation that it could resolve
1576 // to wrong library name if the .so file has a base address != NULL. Here
1577 // we iterate through the program headers of all loaded libraries to find
1578 // out which library 'addr' really belongs to. This workaround can be
1579 // removed once the minimum requirement for glibc is moved to 2.3.x.
1580 data.addr = addr;
1581 data.fname = buf;
1582 data.buflen = buflen;
1583 data.base = NULL;
1584 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1585
1586 if (rslt) {
1587 // buf already contains library name
1588 if (offset) *offset = addr - data.base;
1589 return true;
1590 }
1591 if (dladdr((void*)addr, &dlinfo) != 0) {
1592 if (dlinfo.dli_fname != NULL) {
1593 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1594 }
1595 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1596 *offset = addr - (address)dlinfo.dli_fbase;
1597 }
1598 return true;
1599 }
1600
1601 buf[0] = '\0';
1602 if (offset) *offset = -1;
1603 return false;
1604 }
1605
1606 // Loads .dll/.so and
1607 // in case of error it checks if .dll/.so was built for the
1608 // same architecture as Hotspot is running on
1609
1610
1611 // Remember the stack's state. The Linux dynamic linker will change
1612 // the stack to 'executable' at most once, so we must safepoint only once.
1613 bool os::Linux::_stack_is_executable = false;
1614
1615 // VM operation that loads a library. This is necessary if stack protection
1616 // of the Java stacks can be lost during loading the library. If we
1617 // do not stop the Java threads, they can stack overflow before the stacks
1618 // are protected again.
1619 class VM_LinuxDllLoad: public VM_Operation {
1620 private:
1621 const char *_filename;
1622 char *_ebuf;
1623 int _ebuflen;
1624 void *_lib;
1625 public:
1626 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1627 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1628 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1629 void doit() {
1630 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1631 os::Linux::_stack_is_executable = true;
1632 }
1633 void* loaded_library() { return _lib; }
1634 };
1635
1636 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1637 void * result = NULL;
1638 bool load_attempted = false;
1639
1640 // Check whether the library to load might change execution rights
1641 // of the stack. If they are changed, the protection of the stack
1642 // guard pages will be lost. We need a safepoint to fix this.
1643 //
1644 // See Linux man page execstack(8) for more info.
1645 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1646 if (!ElfFile::specifies_noexecstack(filename)) {
1647 if (!is_init_completed()) {
1648 os::Linux::_stack_is_executable = true;
1649 // This is OK - No Java threads have been created yet, and hence no
1650 // stack guard pages to fix.
1651 //
1652 // This should happen only when you are building JDK7 using a very
1653 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1654 //
1655 // Dynamic loader will make all stacks executable after
1656 // this function returns, and will not do that again.
1657 #ifdef ASSERT
1658 ThreadsListHandle tlh;
1659 assert(tlh.length() == 0, "no Java threads should exist yet.");
1660 #endif
1661 } else {
1662 warning("You have loaded library %s which might have disabled stack guard. "
1663 "The VM will try to fix the stack guard now.\n"
1664 "It's highly recommended that you fix the library with "
1665 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1666 filename);
1667
1668 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1669 JavaThread *jt = JavaThread::current();
1670 if (jt->thread_state() != _thread_in_native) {
1671 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1672 // that requires ExecStack. Cannot enter safe point. Let's give up.
1673 warning("Unable to fix stack guard. Giving up.");
1674 } else {
1675 if (!LoadExecStackDllInVMThread) {
1676 // This is for the case where the DLL has an static
1677 // constructor function that executes JNI code. We cannot
1678 // load such DLLs in the VMThread.
1679 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1680 }
1681
1682 ThreadInVMfromNative tiv(jt);
1683 debug_only(VMNativeEntryWrapper vew;)
1684
1685 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1686 VMThread::execute(&op);
1687 if (LoadExecStackDllInVMThread) {
1688 result = op.loaded_library();
1689 }
1690 load_attempted = true;
1691 }
1692 }
1693 }
1694 }
1695
1696 if (!load_attempted) {
1697 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1698 }
1699
1700 if (result != NULL) {
1701 // Successful loading
1702 return result;
1703 }
1704
1705 Elf32_Ehdr elf_head;
1706 int diag_msg_max_length=ebuflen-strlen(ebuf);
1707 char* diag_msg_buf=ebuf+strlen(ebuf);
1708
1709 if (diag_msg_max_length==0) {
1710 // No more space in ebuf for additional diagnostics message
1711 return NULL;
1712 }
1713
1714
1715 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1716
1717 if (file_descriptor < 0) {
1718 // Can't open library, report dlerror() message
1719 return NULL;
1720 }
1721
1722 bool failed_to_read_elf_head=
1723 (sizeof(elf_head)!=
1724 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1725
1726 ::close(file_descriptor);
1727 if (failed_to_read_elf_head) {
1728 // file i/o error - report dlerror() msg
1729 return NULL;
1730 }
1731
1732 typedef struct {
1733 Elf32_Half code; // Actual value as defined in elf.h
1734 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1735 unsigned char elf_class; // 32 or 64 bit
1736 unsigned char endianess; // MSB or LSB
1737 char* name; // String representation
1738 } arch_t;
1739
1740 #ifndef EM_486
1741 #define EM_486 6 /* Intel 80486 */
1742 #endif
1743 #ifndef EM_AARCH64
1744 #define EM_AARCH64 183 /* ARM AARCH64 */
1745 #endif
1746
1747 static const arch_t arch_array[]={
1748 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1749 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1750 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1751 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1752 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1753 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1754 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1755 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1756 #if defined(VM_LITTLE_ENDIAN)
1757 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1758 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1759 #else
1760 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1761 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1762 #endif
1763 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1764 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1765 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1766 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1767 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1768 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1769 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1770 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1771 };
1772
1773 #if (defined IA32)
1774 static Elf32_Half running_arch_code=EM_386;
1775 #elif (defined AMD64)
1776 static Elf32_Half running_arch_code=EM_X86_64;
1777 #elif (defined IA64)
1778 static Elf32_Half running_arch_code=EM_IA_64;
1779 #elif (defined __sparc) && (defined _LP64)
1780 static Elf32_Half running_arch_code=EM_SPARCV9;
1781 #elif (defined __sparc) && (!defined _LP64)
1782 static Elf32_Half running_arch_code=EM_SPARC;
1783 #elif (defined __powerpc64__)
1784 static Elf32_Half running_arch_code=EM_PPC64;
1785 #elif (defined __powerpc__)
1786 static Elf32_Half running_arch_code=EM_PPC;
1787 #elif (defined AARCH64)
1788 static Elf32_Half running_arch_code=EM_AARCH64;
1789 #elif (defined ARM)
1790 static Elf32_Half running_arch_code=EM_ARM;
1791 #elif (defined S390)
1792 static Elf32_Half running_arch_code=EM_S390;
1793 #elif (defined ALPHA)
1794 static Elf32_Half running_arch_code=EM_ALPHA;
1795 #elif (defined MIPSEL)
1796 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1797 #elif (defined PARISC)
1798 static Elf32_Half running_arch_code=EM_PARISC;
1799 #elif (defined MIPS)
1800 static Elf32_Half running_arch_code=EM_MIPS;
1801 #elif (defined M68K)
1802 static Elf32_Half running_arch_code=EM_68K;
1803 #elif (defined SH)
1804 static Elf32_Half running_arch_code=EM_SH;
1805 #else
1806 #error Method os::dll_load requires that one of following is defined:\
1807 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1808 #endif
1809
1810 // Identify compatability class for VM's architecture and library's architecture
1811 // Obtain string descriptions for architectures
1812
1813 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1814 int running_arch_index=-1;
1815
1816 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1817 if (running_arch_code == arch_array[i].code) {
1818 running_arch_index = i;
1819 }
1820 if (lib_arch.code == arch_array[i].code) {
1821 lib_arch.compat_class = arch_array[i].compat_class;
1822 lib_arch.name = arch_array[i].name;
1823 }
1824 }
1825
1826 assert(running_arch_index != -1,
1827 "Didn't find running architecture code (running_arch_code) in arch_array");
1828 if (running_arch_index == -1) {
1829 // Even though running architecture detection failed
1830 // we may still continue with reporting dlerror() message
1831 return NULL;
1832 }
1833
1834 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1835 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1836 return NULL;
1837 }
1838
1839 #ifndef S390
1840 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1841 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1842 return NULL;
1843 }
1844 #endif // !S390
1845
1846 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1847 if (lib_arch.name!=NULL) {
1848 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1849 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1850 lib_arch.name, arch_array[running_arch_index].name);
1851 } else {
1852 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1853 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1854 lib_arch.code,
1855 arch_array[running_arch_index].name);
1856 }
1857 }
1858
1859 return NULL;
1860 }
1861
1862 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1863 int ebuflen) {
1864 void * result = ::dlopen(filename, RTLD_LAZY);
1865 if (result == NULL) {
1866 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1867 ebuf[ebuflen-1] = '\0';
1868 }
1869 return result;
1870 }
1871
1872 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1873 int ebuflen) {
1874 void * result = NULL;
1875 if (LoadExecStackDllInVMThread) {
1876 result = dlopen_helper(filename, ebuf, ebuflen);
1877 }
1878
1879 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1880 // library that requires an executable stack, or which does not have this
1881 // stack attribute set, dlopen changes the stack attribute to executable. The
1882 // read protection of the guard pages gets lost.
1883 //
1884 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1885 // may have been queued at the same time.
1886
1887 if (!_stack_is_executable) {
1888 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1889 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1890 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1891 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1892 warning("Attempt to reguard stack yellow zone failed.");
1893 }
1894 }
1895 }
1896 }
1897
1898 return result;
1899 }
1900
1901 void* os::dll_lookup(void* handle, const char* name) {
1902 void* res = dlsym(handle, name);
1903 return res;
1904 }
1905
1906 void* os::get_default_process_handle() {
1907 return (void*)::dlopen(NULL, RTLD_LAZY);
1908 }
1909
1910 static bool _print_ascii_file(const char* filename, outputStream* st) {
1911 int fd = ::open(filename, O_RDONLY);
1912 if (fd == -1) {
1913 return false;
1914 }
1915
1916 char buf[33];
1917 int bytes;
1918 buf[32] = '\0';
1919 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1920 st->print_raw(buf, bytes);
1921 }
1922
1923 ::close(fd);
1924
1925 return true;
1926 }
1927
1928 void os::print_dll_info(outputStream *st) {
1929 st->print_cr("Dynamic libraries:");
1930
1931 char fname[32];
1932 pid_t pid = os::Linux::gettid();
1933
1934 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1935
1936 if (!_print_ascii_file(fname, st)) {
1937 st->print("Can not get library information for pid = %d\n", pid);
1938 }
1939 }
1940
1941 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1942 FILE *procmapsFile = NULL;
1943
1944 // Open the procfs maps file for the current process
1945 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1946 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1947 char line[PATH_MAX + 100];
1948
1949 // Read line by line from 'file'
1950 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1951 u8 base, top, offset, inode;
1952 char permissions[5];
1953 char device[6];
1954 char name[PATH_MAX + 1];
1955
1956 // Parse fields from line
1957 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1958 &base, &top, permissions, &offset, device, &inode, name);
1959
1960 // Filter by device id '00:00' so that we only get file system mapped files.
1961 if (strcmp(device, "00:00") != 0) {
1962
1963 // Call callback with the fields of interest
1964 if(callback(name, (address)base, (address)top, param)) {
1965 // Oops abort, callback aborted
1966 fclose(procmapsFile);
1967 return 1;
1968 }
1969 }
1970 }
1971 fclose(procmapsFile);
1972 }
1973 return 0;
1974 }
1975
1976 void os::print_os_info_brief(outputStream* st) {
1977 os::Linux::print_distro_info(st);
1978
1979 os::Posix::print_uname_info(st);
1980
1981 os::Linux::print_libversion_info(st);
1982
1983 }
1984
1985 void os::print_os_info(outputStream* st) {
1986 st->print("OS:");
1987
1988 os::Linux::print_distro_info(st);
1989
1990 os::Posix::print_uname_info(st);
1991
1992 // Print warning if unsafe chroot environment detected
1993 if (unsafe_chroot_detected) {
1994 st->print("WARNING!! ");
1995 st->print_cr("%s", unstable_chroot_error);
1996 }
1997
1998 os::Linux::print_libversion_info(st);
1999
2000 os::Posix::print_rlimit_info(st);
2001
2002 os::Posix::print_load_average(st);
2003
2004 os::Linux::print_full_memory_info(st);
2005
2006 os::Linux::print_container_info(st);
2007 }
2008
2009 // Try to identify popular distros.
2010 // Most Linux distributions have a /etc/XXX-release file, which contains
2011 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2012 // file that also contains the OS version string. Some have more than one
2013 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2014 // /etc/redhat-release.), so the order is important.
2015 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2016 // their own specific XXX-release file as well as a redhat-release file.
2017 // Because of this the XXX-release file needs to be searched for before the
2018 // redhat-release file.
2019 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2020 // search for redhat-release / SuSE-release needs to be before lsb-release.
2021 // Since the lsb-release file is the new standard it needs to be searched
2022 // before the older style release files.
2023 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2024 // next to last resort. The os-release file is a new standard that contains
2025 // distribution information and the system-release file seems to be an old
2026 // standard that has been replaced by the lsb-release and os-release files.
2027 // Searching for the debian_version file is the last resort. It contains
2028 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2029 // "Debian " is printed before the contents of the debian_version file.
2030
2031 const char* distro_files[] = {
2032 "/etc/oracle-release",
2033 "/etc/mandriva-release",
2034 "/etc/mandrake-release",
2035 "/etc/sun-release",
2036 "/etc/redhat-release",
2037 "/etc/SuSE-release",
2038 "/etc/lsb-release",
2039 "/etc/turbolinux-release",
2040 "/etc/gentoo-release",
2041 "/etc/ltib-release",
2042 "/etc/angstrom-version",
2043 "/etc/system-release",
2044 "/etc/os-release",
2045 NULL };
2046
2047 void os::Linux::print_distro_info(outputStream* st) {
2048 for (int i = 0;; i++) {
2049 const char* file = distro_files[i];
2050 if (file == NULL) {
2051 break; // done
2052 }
2053 // If file prints, we found it.
2054 if (_print_ascii_file(file, st)) {
2055 return;
2056 }
2057 }
2058
2059 if (file_exists("/etc/debian_version")) {
2060 st->print("Debian ");
2061 _print_ascii_file("/etc/debian_version", st);
2062 } else {
2063 st->print("Linux");
2064 }
2065 st->cr();
2066 }
2067
2068 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2069 char buf[256];
2070 while (fgets(buf, sizeof(buf), fp)) {
2071 // Edit out extra stuff in expected format
2072 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2073 char* ptr = strstr(buf, "\""); // the name is in quotes
2074 if (ptr != NULL) {
2075 ptr++; // go beyond first quote
2076 char* nl = strchr(ptr, '\"');
2077 if (nl != NULL) *nl = '\0';
2078 strncpy(distro, ptr, length);
2079 } else {
2080 ptr = strstr(buf, "=");
2081 ptr++; // go beyond equals then
2082 char* nl = strchr(ptr, '\n');
2083 if (nl != NULL) *nl = '\0';
2084 strncpy(distro, ptr, length);
2085 }
2086 return;
2087 } else if (get_first_line) {
2088 char* nl = strchr(buf, '\n');
2089 if (nl != NULL) *nl = '\0';
2090 strncpy(distro, buf, length);
2091 return;
2092 }
2093 }
2094 // print last line and close
2095 char* nl = strchr(buf, '\n');
2096 if (nl != NULL) *nl = '\0';
2097 strncpy(distro, buf, length);
2098 }
2099
2100 static void parse_os_info(char* distro, size_t length, const char* file) {
2101 FILE* fp = fopen(file, "r");
2102 if (fp != NULL) {
2103 // if suse format, print out first line
2104 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2105 parse_os_info_helper(fp, distro, length, get_first_line);
2106 fclose(fp);
2107 }
2108 }
2109
2110 void os::get_summary_os_info(char* buf, size_t buflen) {
2111 for (int i = 0;; i++) {
2112 const char* file = distro_files[i];
2113 if (file == NULL) {
2114 break; // ran out of distro_files
2115 }
2116 if (file_exists(file)) {
2117 parse_os_info(buf, buflen, file);
2118 return;
2119 }
2120 }
2121 // special case for debian
2122 if (file_exists("/etc/debian_version")) {
2123 strncpy(buf, "Debian ", buflen);
2124 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2125 } else {
2126 strncpy(buf, "Linux", buflen);
2127 }
2128 }
2129
2130 void os::Linux::print_libversion_info(outputStream* st) {
2131 // libc, pthread
2132 st->print("libc:");
2133 st->print("%s ", os::Linux::glibc_version());
2134 st->print("%s ", os::Linux::libpthread_version());
2135 st->cr();
2136 }
2137
2138 void os::Linux::print_full_memory_info(outputStream* st) {
2139 st->print("\n/proc/meminfo:\n");
2140 _print_ascii_file("/proc/meminfo", st);
2141 st->cr();
2142 }
2143
2144 void os::Linux::print_container_info(outputStream* st) {
2145 if (!OSContainer::is_containerized()) {
2146 return;
2147 }
2148
2149 st->print("container (cgroup) information:\n");
2150
2151 const char *p_ct = OSContainer::container_type();
2152 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2153
2154 char *p = OSContainer::cpu_cpuset_cpus();
2155 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2156 free(p);
2157
2158 p = OSContainer::cpu_cpuset_memory_nodes();
2159 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2160 free(p);
2161
2162 int i = OSContainer::active_processor_count();
2163 if (i > 0) {
2164 st->print("active_processor_count: %d\n", i);
2165 } else {
2166 st->print("active_processor_count: failed\n");
2167 }
2168
2169 i = OSContainer::cpu_quota();
2170 st->print("cpu_quota: %d\n", i);
2171
2172 i = OSContainer::cpu_period();
2173 st->print("cpu_period: %d\n", i);
2174
2175 i = OSContainer::cpu_shares();
2176 st->print("cpu_shares: %d\n", i);
2177
2178 jlong j = OSContainer::memory_limit_in_bytes();
2179 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2180
2181 j = OSContainer::memory_and_swap_limit_in_bytes();
2182 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2183
2184 j = OSContainer::memory_soft_limit_in_bytes();
2185 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2186
2187 j = OSContainer::OSContainer::memory_usage_in_bytes();
2188 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2189
2190 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2191 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2192 st->cr();
2193 }
2194
2195 void os::print_memory_info(outputStream* st) {
2196
2197 st->print("Memory:");
2198 st->print(" %dk page", os::vm_page_size()>>10);
2199
2200 // values in struct sysinfo are "unsigned long"
2201 struct sysinfo si;
2202 sysinfo(&si);
2203
2204 st->print(", physical " UINT64_FORMAT "k",
2205 os::physical_memory() >> 10);
2206 st->print("(" UINT64_FORMAT "k free)",
2207 os::available_memory() >> 10);
2208 st->print(", swap " UINT64_FORMAT "k",
2209 ((jlong)si.totalswap * si.mem_unit) >> 10);
2210 st->print("(" UINT64_FORMAT "k free)",
2211 ((jlong)si.freeswap * si.mem_unit) >> 10);
2212 st->cr();
2213 }
2214
2215 // Print the first "model name" line and the first "flags" line
2216 // that we find and nothing more. We assume "model name" comes
2217 // before "flags" so if we find a second "model name", then the
2218 // "flags" field is considered missing.
2219 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2220 #if defined(IA32) || defined(AMD64)
2221 // Other platforms have less repetitive cpuinfo files
2222 FILE *fp = fopen("/proc/cpuinfo", "r");
2223 if (fp) {
2224 while (!feof(fp)) {
2225 if (fgets(buf, buflen, fp)) {
2226 // Assume model name comes before flags
2227 bool model_name_printed = false;
2228 if (strstr(buf, "model name") != NULL) {
2229 if (!model_name_printed) {
2230 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2231 st->print_raw(buf);
2232 model_name_printed = true;
2233 } else {
2234 // model name printed but not flags? Odd, just return
2235 fclose(fp);
2236 return true;
2237 }
2238 }
2239 // print the flags line too
2240 if (strstr(buf, "flags") != NULL) {
2241 st->print_raw(buf);
2242 fclose(fp);
2243 return true;
2244 }
2245 }
2246 }
2247 fclose(fp);
2248 }
2249 #endif // x86 platforms
2250 return false;
2251 }
2252
2253 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2254 // Only print the model name if the platform provides this as a summary
2255 if (!print_model_name_and_flags(st, buf, buflen)) {
2256 st->print("\n/proc/cpuinfo:\n");
2257 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2258 st->print_cr(" <Not Available>");
2259 }
2260 }
2261 }
2262
2263 #if defined(AMD64) || defined(IA32) || defined(X32)
2264 const char* search_string = "model name";
2265 #elif defined(M68K)
2266 const char* search_string = "CPU";
2267 #elif defined(PPC64)
2268 const char* search_string = "cpu";
2269 #elif defined(S390)
2270 const char* search_string = "processor";
2271 #elif defined(SPARC)
2272 const char* search_string = "cpu";
2273 #else
2274 const char* search_string = "Processor";
2275 #endif
2276
2277 // Parses the cpuinfo file for string representing the model name.
2278 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2279 FILE* fp = fopen("/proc/cpuinfo", "r");
2280 if (fp != NULL) {
2281 while (!feof(fp)) {
2282 char buf[256];
2283 if (fgets(buf, sizeof(buf), fp)) {
2284 char* start = strstr(buf, search_string);
2285 if (start != NULL) {
2286 char *ptr = start + strlen(search_string);
2287 char *end = buf + strlen(buf);
2288 while (ptr != end) {
2289 // skip whitespace and colon for the rest of the name.
2290 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2291 break;
2292 }
2293 ptr++;
2294 }
2295 if (ptr != end) {
2296 // reasonable string, get rid of newline and keep the rest
2297 char* nl = strchr(buf, '\n');
2298 if (nl != NULL) *nl = '\0';
2299 strncpy(cpuinfo, ptr, length);
2300 fclose(fp);
2301 return;
2302 }
2303 }
2304 }
2305 }
2306 fclose(fp);
2307 }
2308 // cpuinfo not found or parsing failed, just print generic string. The entire
2309 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2310 #if defined(AARCH64)
2311 strncpy(cpuinfo, "AArch64", length);
2312 #elif defined(AMD64)
2313 strncpy(cpuinfo, "x86_64", length);
2314 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2315 strncpy(cpuinfo, "ARM", length);
2316 #elif defined(IA32)
2317 strncpy(cpuinfo, "x86_32", length);
2318 #elif defined(IA64)
2319 strncpy(cpuinfo, "IA64", length);
2320 #elif defined(PPC)
2321 strncpy(cpuinfo, "PPC64", length);
2322 #elif defined(S390)
2323 strncpy(cpuinfo, "S390", length);
2324 #elif defined(SPARC)
2325 strncpy(cpuinfo, "sparcv9", length);
2326 #elif defined(ZERO_LIBARCH)
2327 strncpy(cpuinfo, ZERO_LIBARCH, length);
2328 #else
2329 strncpy(cpuinfo, "unknown", length);
2330 #endif
2331 }
2332
2333 static void print_signal_handler(outputStream* st, int sig,
2334 char* buf, size_t buflen);
2335
2336 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2337 st->print_cr("Signal Handlers:");
2338 print_signal_handler(st, SIGSEGV, buf, buflen);
2339 print_signal_handler(st, SIGBUS , buf, buflen);
2340 print_signal_handler(st, SIGFPE , buf, buflen);
2341 print_signal_handler(st, SIGPIPE, buf, buflen);
2342 print_signal_handler(st, SIGXFSZ, buf, buflen);
2343 print_signal_handler(st, SIGILL , buf, buflen);
2344 print_signal_handler(st, SR_signum, buf, buflen);
2345 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2346 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2347 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2348 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2349 #if defined(PPC64)
2350 print_signal_handler(st, SIGTRAP, buf, buflen);
2351 #endif
2352 }
2353
2354 static char saved_jvm_path[MAXPATHLEN] = {0};
2355
2356 // Find the full path to the current module, libjvm.so
2357 void os::jvm_path(char *buf, jint buflen) {
2358 // Error checking.
2359 if (buflen < MAXPATHLEN) {
2360 assert(false, "must use a large-enough buffer");
2361 buf[0] = '\0';
2362 return;
2363 }
2364 // Lazy resolve the path to current module.
2365 if (saved_jvm_path[0] != 0) {
2366 strcpy(buf, saved_jvm_path);
2367 return;
2368 }
2369
2370 char dli_fname[MAXPATHLEN];
2371 bool ret = dll_address_to_library_name(
2372 CAST_FROM_FN_PTR(address, os::jvm_path),
2373 dli_fname, sizeof(dli_fname), NULL);
2374 assert(ret, "cannot locate libjvm");
2375 char *rp = NULL;
2376 if (ret && dli_fname[0] != '\0') {
2377 rp = os::Posix::realpath(dli_fname, buf, buflen);
2378 }
2379 if (rp == NULL) {
2380 return;
2381 }
2382
2383 if (Arguments::sun_java_launcher_is_altjvm()) {
2384 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2385 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2386 // If "/jre/lib/" appears at the right place in the string, then
2387 // assume we are installed in a JDK and we're done. Otherwise, check
2388 // for a JAVA_HOME environment variable and fix up the path so it
2389 // looks like libjvm.so is installed there (append a fake suffix
2390 // hotspot/libjvm.so).
2391 const char *p = buf + strlen(buf) - 1;
2392 for (int count = 0; p > buf && count < 5; ++count) {
2393 for (--p; p > buf && *p != '/'; --p)
2394 /* empty */ ;
2395 }
2396
2397 if (strncmp(p, "/jre/lib/", 9) != 0) {
2398 // Look for JAVA_HOME in the environment.
2399 char* java_home_var = ::getenv("JAVA_HOME");
2400 if (java_home_var != NULL && java_home_var[0] != 0) {
2401 char* jrelib_p;
2402 int len;
2403
2404 // Check the current module name "libjvm.so".
2405 p = strrchr(buf, '/');
2406 if (p == NULL) {
2407 return;
2408 }
2409 assert(strstr(p, "/libjvm") == p, "invalid library name");
2410
2411 rp = os::Posix::realpath(java_home_var, buf, buflen);
2412 if (rp == NULL) {
2413 return;
2414 }
2415
2416 // determine if this is a legacy image or modules image
2417 // modules image doesn't have "jre" subdirectory
2418 len = strlen(buf);
2419 assert(len < buflen, "Ran out of buffer room");
2420 jrelib_p = buf + len;
2421 snprintf(jrelib_p, buflen-len, "/jre/lib");
2422 if (0 != access(buf, F_OK)) {
2423 snprintf(jrelib_p, buflen-len, "/lib");
2424 }
2425
2426 if (0 == access(buf, F_OK)) {
2427 // Use current module name "libjvm.so"
2428 len = strlen(buf);
2429 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2430 } else {
2431 // Go back to path of .so
2432 rp = os::Posix::realpath(dli_fname, buf, buflen);
2433 if (rp == NULL) {
2434 return;
2435 }
2436 }
2437 }
2438 }
2439 }
2440
2441 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2442 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2443 }
2444
2445 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2446 // no prefix required, not even "_"
2447 }
2448
2449 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2450 // no suffix required
2451 }
2452
2453 ////////////////////////////////////////////////////////////////////////////////
2454 // sun.misc.Signal support
2455
2456 static volatile jint sigint_count = 0;
2457
2458 static void UserHandler(int sig, void *siginfo, void *context) {
2459 // 4511530 - sem_post is serialized and handled by the manager thread. When
2460 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2461 // don't want to flood the manager thread with sem_post requests.
2462 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2463 return;
2464 }
2465
2466 // Ctrl-C is pressed during error reporting, likely because the error
2467 // handler fails to abort. Let VM die immediately.
2468 if (sig == SIGINT && VMError::is_error_reported()) {
2469 os::die();
2470 }
2471
2472 os::signal_notify(sig);
2473 }
2474
2475 void* os::user_handler() {
2476 return CAST_FROM_FN_PTR(void*, UserHandler);
2477 }
2478
2479 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2480 struct timespec ts;
2481 // Semaphore's are always associated with CLOCK_REALTIME
2482 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2483 // see os_posix.cpp for discussion on overflow checking
2484 if (sec >= MAX_SECS) {
2485 ts.tv_sec += MAX_SECS;
2486 ts.tv_nsec = 0;
2487 } else {
2488 ts.tv_sec += sec;
2489 ts.tv_nsec += nsec;
2490 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2491 ts.tv_nsec -= NANOSECS_PER_SEC;
2492 ++ts.tv_sec; // note: this must be <= max_secs
2493 }
2494 }
2495
2496 return ts;
2497 }
2498
2499 extern "C" {
2500 typedef void (*sa_handler_t)(int);
2501 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2502 }
2503
2504 void* os::signal(int signal_number, void* handler) {
2505 struct sigaction sigAct, oldSigAct;
2506
2507 sigfillset(&(sigAct.sa_mask));
2508 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2509 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2510
2511 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2512 // -1 means registration failed
2513 return (void *)-1;
2514 }
2515
2516 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2517 }
2518
2519 void os::signal_raise(int signal_number) {
2520 ::raise(signal_number);
2521 }
2522
2523 // The following code is moved from os.cpp for making this
2524 // code platform specific, which it is by its very nature.
2525
2526 // Will be modified when max signal is changed to be dynamic
2527 int os::sigexitnum_pd() {
2528 return NSIG;
2529 }
2530
2531 // a counter for each possible signal value
2532 static volatile jint pending_signals[NSIG+1] = { 0 };
2533
2534 // Linux(POSIX) specific hand shaking semaphore.
2535 static Semaphore* sig_sem = NULL;
2536 static PosixSemaphore sr_semaphore;
2537
2538 void os::signal_init_pd() {
2539 // Initialize signal structures
2540 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2541
2542 // Initialize signal semaphore
2543 sig_sem = new Semaphore();
2544 }
2545
2546 void os::signal_notify(int sig) {
2547 if (sig_sem != NULL) {
2548 Atomic::inc(&pending_signals[sig]);
2549 sig_sem->signal();
2550 } else {
2551 // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2552 // initialization isn't called.
2553 assert(ReduceSignalUsage, "signal semaphore should be created");
2554 }
2555 }
2556
2557 static int check_pending_signals() {
2558 Atomic::store(0, &sigint_count);
2559 for (;;) {
2560 for (int i = 0; i < NSIG + 1; i++) {
2561 jint n = pending_signals[i];
2562 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2563 return i;
2564 }
2565 }
2566 JavaThread *thread = JavaThread::current();
2567 ThreadBlockInVM tbivm(thread);
2568
2569 bool threadIsSuspended;
2570 do {
2571 thread->set_suspend_equivalent();
2572 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2573 sig_sem->wait();
2574
2575 // were we externally suspended while we were waiting?
2576 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2577 if (threadIsSuspended) {
2578 // The semaphore has been incremented, but while we were waiting
2579 // another thread suspended us. We don't want to continue running
2580 // while suspended because that would surprise the thread that
2581 // suspended us.
2582 sig_sem->signal();
2583
2584 thread->java_suspend_self();
2585 }
2586 } while (threadIsSuspended);
2587 }
2588 }
2589
2590 int os::signal_wait() {
2591 return check_pending_signals();
2592 }
2593
2594 ////////////////////////////////////////////////////////////////////////////////
2595 // Virtual Memory
2596
2597 int os::vm_page_size() {
2598 // Seems redundant as all get out
2599 assert(os::Linux::page_size() != -1, "must call os::init");
2600 return os::Linux::page_size();
2601 }
2602
2603 // Solaris allocates memory by pages.
2604 int os::vm_allocation_granularity() {
2605 assert(os::Linux::page_size() != -1, "must call os::init");
2606 return os::Linux::page_size();
2607 }
2608
2609 // Rationale behind this function:
2610 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2611 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2612 // samples for JITted code. Here we create private executable mapping over the code cache
2613 // and then we can use standard (well, almost, as mapping can change) way to provide
2614 // info for the reporting script by storing timestamp and location of symbol
2615 void linux_wrap_code(char* base, size_t size) {
2616 static volatile jint cnt = 0;
2617
2618 if (!UseOprofile) {
2619 return;
2620 }
2621
2622 char buf[PATH_MAX+1];
2623 int num = Atomic::add(1, &cnt);
2624
2625 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2626 os::get_temp_directory(), os::current_process_id(), num);
2627 unlink(buf);
2628
2629 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2630
2631 if (fd != -1) {
2632 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2633 if (rv != (off_t)-1) {
2634 if (::write(fd, "", 1) == 1) {
2635 mmap(base, size,
2636 PROT_READ|PROT_WRITE|PROT_EXEC,
2637 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2638 }
2639 }
2640 ::close(fd);
2641 unlink(buf);
2642 }
2643 }
2644
2645 static bool recoverable_mmap_error(int err) {
2646 // See if the error is one we can let the caller handle. This
2647 // list of errno values comes from JBS-6843484. I can't find a
2648 // Linux man page that documents this specific set of errno
2649 // values so while this list currently matches Solaris, it may
2650 // change as we gain experience with this failure mode.
2651 switch (err) {
2652 case EBADF:
2653 case EINVAL:
2654 case ENOTSUP:
2655 // let the caller deal with these errors
2656 return true;
2657
2658 default:
2659 // Any remaining errors on this OS can cause our reserved mapping
2660 // to be lost. That can cause confusion where different data
2661 // structures think they have the same memory mapped. The worst
2662 // scenario is if both the VM and a library think they have the
2663 // same memory mapped.
2664 return false;
2665 }
2666 }
2667
2668 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2669 int err) {
2670 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2671 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2672 os::strerror(err), err);
2673 }
2674
2675 static void warn_fail_commit_memory(char* addr, size_t size,
2676 size_t alignment_hint, bool exec,
2677 int err) {
2678 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2679 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2680 alignment_hint, exec, os::strerror(err), err);
2681 }
2682
2683 // NOTE: Linux kernel does not really reserve the pages for us.
2684 // All it does is to check if there are enough free pages
2685 // left at the time of mmap(). This could be a potential
2686 // problem.
2687 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2688 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2689 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2690 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2691 if (res != (uintptr_t) MAP_FAILED) {
2692 if (UseNUMAInterleaving) {
2693 numa_make_global(addr, size);
2694 }
2695 return 0;
2696 }
2697
2698 int err = errno; // save errno from mmap() call above
2699
2700 if (!recoverable_mmap_error(err)) {
2701 warn_fail_commit_memory(addr, size, exec, err);
2702 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2703 }
2704
2705 return err;
2706 }
2707
2708 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2709 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2710 }
2711
2712 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2713 const char* mesg) {
2714 assert(mesg != NULL, "mesg must be specified");
2715 int err = os::Linux::commit_memory_impl(addr, size, exec);
2716 if (err != 0) {
2717 // the caller wants all commit errors to exit with the specified mesg:
2718 warn_fail_commit_memory(addr, size, exec, err);
2719 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2720 }
2721 }
2722
2723 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2724 #ifndef MAP_HUGETLB
2725 #define MAP_HUGETLB 0x40000
2726 #endif
2727
2728 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2729 #ifndef MADV_HUGEPAGE
2730 #define MADV_HUGEPAGE 14
2731 #endif
2732
2733 int os::Linux::commit_memory_impl(char* addr, size_t size,
2734 size_t alignment_hint, bool exec) {
2735 int err = os::Linux::commit_memory_impl(addr, size, exec);
2736 if (err == 0) {
2737 realign_memory(addr, size, alignment_hint);
2738 }
2739 return err;
2740 }
2741
2742 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2743 bool exec) {
2744 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2745 }
2746
2747 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2748 size_t alignment_hint, bool exec,
2749 const char* mesg) {
2750 assert(mesg != NULL, "mesg must be specified");
2751 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2752 if (err != 0) {
2753 // the caller wants all commit errors to exit with the specified mesg:
2754 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2755 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2756 }
2757 }
2758
2759 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2760 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2761 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2762 // be supported or the memory may already be backed by huge pages.
2763 ::madvise(addr, bytes, MADV_HUGEPAGE);
2764 }
2765 }
2766
2767 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2768 // This method works by doing an mmap over an existing mmaping and effectively discarding
2769 // the existing pages. However it won't work for SHM-based large pages that cannot be
2770 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2771 // small pages on top of the SHM segment. This method always works for small pages, so we
2772 // allow that in any case.
2773 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2774 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2775 }
2776 }
2777
2778 void os::numa_make_global(char *addr, size_t bytes) {
2779 Linux::numa_interleave_memory(addr, bytes);
2780 }
2781
2782 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2783 // bind policy to MPOL_PREFERRED for the current thread.
2784 #define USE_MPOL_PREFERRED 0
2785
2786 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2787 // To make NUMA and large pages more robust when both enabled, we need to ease
2788 // the requirements on where the memory should be allocated. MPOL_BIND is the
2789 // default policy and it will force memory to be allocated on the specified
2790 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2791 // the specified node, but will not force it. Using this policy will prevent
2792 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2793 // free large pages.
2794 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2795 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2796 }
2797
2798 bool os::numa_topology_changed() { return false; }
2799
2800 size_t os::numa_get_groups_num() {
2801 // Return just the number of nodes in which it's possible to allocate memory
2802 // (in numa terminology, configured nodes).
2803 return Linux::numa_num_configured_nodes();
2804 }
2805
2806 int os::numa_get_group_id() {
2807 int cpu_id = Linux::sched_getcpu();
2808 if (cpu_id != -1) {
2809 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2810 if (lgrp_id != -1) {
2811 return lgrp_id;
2812 }
2813 }
2814 return 0;
2815 }
2816
2817 int os::Linux::get_existing_num_nodes() {
2818 size_t node;
2819 size_t highest_node_number = Linux::numa_max_node();
2820 int num_nodes = 0;
2821
2822 // Get the total number of nodes in the system including nodes without memory.
2823 for (node = 0; node <= highest_node_number; node++) {
2824 if (isnode_in_existing_nodes(node)) {
2825 num_nodes++;
2826 }
2827 }
2828 return num_nodes;
2829 }
2830
2831 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2832 size_t highest_node_number = Linux::numa_max_node();
2833 size_t i = 0;
2834
2835 // Map all node ids in which is possible to allocate memory. Also nodes are
2836 // not always consecutively available, i.e. available from 0 to the highest
2837 // node number.
2838 for (size_t node = 0; node <= highest_node_number; node++) {
2839 if (Linux::isnode_in_configured_nodes(node)) {
2840 ids[i++] = node;
2841 }
2842 }
2843 return i;
2844 }
2845
2846 bool os::get_page_info(char *start, page_info* info) {
2847 return false;
2848 }
2849
2850 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2851 page_info* page_found) {
2852 return end;
2853 }
2854
2855
2856 int os::Linux::sched_getcpu_syscall(void) {
2857 unsigned int cpu = 0;
2858 int retval = -1;
2859
2860 #if defined(IA32)
2861 #ifndef SYS_getcpu
2862 #define SYS_getcpu 318
2863 #endif
2864 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2865 #elif defined(AMD64)
2866 // Unfortunately we have to bring all these macros here from vsyscall.h
2867 // to be able to compile on old linuxes.
2868 #define __NR_vgetcpu 2
2869 #define VSYSCALL_START (-10UL << 20)
2870 #define VSYSCALL_SIZE 1024
2871 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2872 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2873 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2874 retval = vgetcpu(&cpu, NULL, NULL);
2875 #endif
2876
2877 return (retval == -1) ? retval : cpu;
2878 }
2879
2880 void os::Linux::sched_getcpu_init() {
2881 // sched_getcpu() should be in libc.
2882 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2883 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2884
2885 // If it's not, try a direct syscall.
2886 if (sched_getcpu() == -1) {
2887 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2888 (void*)&sched_getcpu_syscall));
2889 }
2890 }
2891
2892 // Something to do with the numa-aware allocator needs these symbols
2893 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2894 extern "C" JNIEXPORT void numa_error(char *where) { }
2895
2896 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2897 // load symbol from base version instead.
2898 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2899 void *f = dlvsym(handle, name, "libnuma_1.1");
2900 if (f == NULL) {
2901 f = dlsym(handle, name);
2902 }
2903 return f;
2904 }
2905
2906 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2907 // Return NULL if the symbol is not defined in this particular version.
2908 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2909 return dlvsym(handle, name, "libnuma_1.2");
2910 }
2911
2912 bool os::Linux::libnuma_init() {
2913 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2914 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2915 if (handle != NULL) {
2916 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2917 libnuma_dlsym(handle, "numa_node_to_cpus")));
2918 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2919 libnuma_dlsym(handle, "numa_max_node")));
2920 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2921 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2922 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2923 libnuma_dlsym(handle, "numa_available")));
2924 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2925 libnuma_dlsym(handle, "numa_tonode_memory")));
2926 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2927 libnuma_dlsym(handle, "numa_interleave_memory")));
2928 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2929 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2930 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2931 libnuma_dlsym(handle, "numa_set_bind_policy")));
2932 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2933 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2934 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2935 libnuma_dlsym(handle, "numa_distance")));
2936
2937 if (numa_available() != -1) {
2938 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2939 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2940 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2941 // Create an index -> node mapping, since nodes are not always consecutive
2942 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2943 rebuild_nindex_to_node_map();
2944 // Create a cpu -> node mapping
2945 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2946 rebuild_cpu_to_node_map();
2947 return true;
2948 }
2949 }
2950 }
2951 return false;
2952 }
2953
2954 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2955 // Creating guard page is very expensive. Java thread has HotSpot
2956 // guard pages, only enable glibc guard page for non-Java threads.
2957 // (Remember: compiler thread is a Java thread, too!)
2958 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2959 }
2960
2961 void os::Linux::rebuild_nindex_to_node_map() {
2962 int highest_node_number = Linux::numa_max_node();
2963
2964 nindex_to_node()->clear();
2965 for (int node = 0; node <= highest_node_number; node++) {
2966 if (Linux::isnode_in_existing_nodes(node)) {
2967 nindex_to_node()->append(node);
2968 }
2969 }
2970 }
2971
2972 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2973 // The table is later used in get_node_by_cpu().
2974 void os::Linux::rebuild_cpu_to_node_map() {
2975 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2976 // in libnuma (possible values are starting from 16,
2977 // and continuing up with every other power of 2, but less
2978 // than the maximum number of CPUs supported by kernel), and
2979 // is a subject to change (in libnuma version 2 the requirements
2980 // are more reasonable) we'll just hardcode the number they use
2981 // in the library.
2982 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2983
2984 size_t cpu_num = processor_count();
2985 size_t cpu_map_size = NCPUS / BitsPerCLong;
2986 size_t cpu_map_valid_size =
2987 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2988
2989 cpu_to_node()->clear();
2990 cpu_to_node()->at_grow(cpu_num - 1);
2991
2992 size_t node_num = get_existing_num_nodes();
2993
2994 int distance = 0;
2995 int closest_distance = INT_MAX;
2996 int closest_node = 0;
2997 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2998 for (size_t i = 0; i < node_num; i++) {
2999 // Check if node is configured (not a memory-less node). If it is not, find
3000 // the closest configured node.
3001 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3002 closest_distance = INT_MAX;
3003 // Check distance from all remaining nodes in the system. Ignore distance
3004 // from itself and from another non-configured node.
3005 for (size_t m = 0; m < node_num; m++) {
3006 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3007 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3008 // If a closest node is found, update. There is always at least one
3009 // configured node in the system so there is always at least one node
3010 // close.
3011 if (distance != 0 && distance < closest_distance) {
3012 closest_distance = distance;
3013 closest_node = nindex_to_node()->at(m);
3014 }
3015 }
3016 }
3017 } else {
3018 // Current node is already a configured node.
3019 closest_node = nindex_to_node()->at(i);
3020 }
3021
3022 // Get cpus from the original node and map them to the closest node. If node
3023 // is a configured node (not a memory-less node), then original node and
3024 // closest node are the same.
3025 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3026 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3027 if (cpu_map[j] != 0) {
3028 for (size_t k = 0; k < BitsPerCLong; k++) {
3029 if (cpu_map[j] & (1UL << k)) {
3030 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3031 }
3032 }
3033 }
3034 }
3035 }
3036 }
3037 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3038 }
3039
3040 int os::Linux::get_node_by_cpu(int cpu_id) {
3041 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3042 return cpu_to_node()->at(cpu_id);
3043 }
3044 return -1;
3045 }
3046
3047 GrowableArray<int>* os::Linux::_cpu_to_node;
3048 GrowableArray<int>* os::Linux::_nindex_to_node;
3049 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3050 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3051 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3052 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3053 os::Linux::numa_available_func_t os::Linux::_numa_available;
3054 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3055 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3056 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3057 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3058 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3059 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3060 unsigned long* os::Linux::_numa_all_nodes;
3061 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3062 struct bitmask* os::Linux::_numa_nodes_ptr;
3063
3064 bool os::pd_uncommit_memory(char* addr, size_t size) {
3065 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3066 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3067 return res != (uintptr_t) MAP_FAILED;
3068 }
3069
3070 static address get_stack_commited_bottom(address bottom, size_t size) {
3071 address nbot = bottom;
3072 address ntop = bottom + size;
3073
3074 size_t page_sz = os::vm_page_size();
3075 unsigned pages = size / page_sz;
3076
3077 unsigned char vec[1];
3078 unsigned imin = 1, imax = pages + 1, imid;
3079 int mincore_return_value = 0;
3080
3081 assert(imin <= imax, "Unexpected page size");
3082
3083 while (imin < imax) {
3084 imid = (imax + imin) / 2;
3085 nbot = ntop - (imid * page_sz);
3086
3087 // Use a trick with mincore to check whether the page is mapped or not.
3088 // mincore sets vec to 1 if page resides in memory and to 0 if page
3089 // is swapped output but if page we are asking for is unmapped
3090 // it returns -1,ENOMEM
3091 mincore_return_value = mincore(nbot, page_sz, vec);
3092
3093 if (mincore_return_value == -1) {
3094 // Page is not mapped go up
3095 // to find first mapped page
3096 if (errno != EAGAIN) {
3097 assert(errno == ENOMEM, "Unexpected mincore errno");
3098 imax = imid;
3099 }
3100 } else {
3101 // Page is mapped go down
3102 // to find first not mapped page
3103 imin = imid + 1;
3104 }
3105 }
3106
3107 nbot = nbot + page_sz;
3108
3109 // Adjust stack bottom one page up if last checked page is not mapped
3110 if (mincore_return_value == -1) {
3111 nbot = nbot + page_sz;
3112 }
3113
3114 return nbot;
3115 }
3116
3117
3118 // Linux uses a growable mapping for the stack, and if the mapping for
3119 // the stack guard pages is not removed when we detach a thread the
3120 // stack cannot grow beyond the pages where the stack guard was
3121 // mapped. If at some point later in the process the stack expands to
3122 // that point, the Linux kernel cannot expand the stack any further
3123 // because the guard pages are in the way, and a segfault occurs.
3124 //
3125 // However, it's essential not to split the stack region by unmapping
3126 // a region (leaving a hole) that's already part of the stack mapping,
3127 // so if the stack mapping has already grown beyond the guard pages at
3128 // the time we create them, we have to truncate the stack mapping.
3129 // So, we need to know the extent of the stack mapping when
3130 // create_stack_guard_pages() is called.
3131
3132 // We only need this for stacks that are growable: at the time of
3133 // writing thread stacks don't use growable mappings (i.e. those
3134 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3135 // only applies to the main thread.
3136
3137 // If the (growable) stack mapping already extends beyond the point
3138 // where we're going to put our guard pages, truncate the mapping at
3139 // that point by munmap()ping it. This ensures that when we later
3140 // munmap() the guard pages we don't leave a hole in the stack
3141 // mapping. This only affects the main/primordial thread
3142
3143 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3144 if (os::is_primordial_thread()) {
3145 // As we manually grow stack up to bottom inside create_attached_thread(),
3146 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3147 // we don't need to do anything special.
3148 // Check it first, before calling heavy function.
3149 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3150 unsigned char vec[1];
3151
3152 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3153 // Fallback to slow path on all errors, including EAGAIN
3154 stack_extent = (uintptr_t) get_stack_commited_bottom(
3155 os::Linux::initial_thread_stack_bottom(),
3156 (size_t)addr - stack_extent);
3157 }
3158
3159 if (stack_extent < (uintptr_t)addr) {
3160 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3161 }
3162 }
3163
3164 return os::commit_memory(addr, size, !ExecMem);
3165 }
3166
3167 // If this is a growable mapping, remove the guard pages entirely by
3168 // munmap()ping them. If not, just call uncommit_memory(). This only
3169 // affects the main/primordial thread, but guard against future OS changes.
3170 // It's safe to always unmap guard pages for primordial thread because we
3171 // always place it right after end of the mapped region.
3172
3173 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3174 uintptr_t stack_extent, stack_base;
3175
3176 if (os::is_primordial_thread()) {
3177 return ::munmap(addr, size) == 0;
3178 }
3179
3180 return os::uncommit_memory(addr, size);
3181 }
3182
3183 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3184 // at 'requested_addr'. If there are existing memory mappings at the same
3185 // location, however, they will be overwritten. If 'fixed' is false,
3186 // 'requested_addr' is only treated as a hint, the return value may or
3187 // may not start from the requested address. Unlike Linux mmap(), this
3188 // function returns NULL to indicate failure.
3189 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3190 char * addr;
3191 int flags;
3192
3193 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3194 if (fixed) {
3195 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3196 flags |= MAP_FIXED;
3197 }
3198
3199 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3200 // touch an uncommitted page. Otherwise, the read/write might
3201 // succeed if we have enough swap space to back the physical page.
3202 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3203 flags, -1, 0);
3204
3205 return addr == MAP_FAILED ? NULL : addr;
3206 }
3207
3208 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3209 // (req_addr != NULL) or with a given alignment.
3210 // - bytes shall be a multiple of alignment.
3211 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3212 // - alignment sets the alignment at which memory shall be allocated.
3213 // It must be a multiple of allocation granularity.
3214 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3215 // req_addr or NULL.
3216 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3217
3218 size_t extra_size = bytes;
3219 if (req_addr == NULL && alignment > 0) {
3220 extra_size += alignment;
3221 }
3222
3223 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3224 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3225 -1, 0);
3226 if (start == MAP_FAILED) {
3227 start = NULL;
3228 } else {
3229 if (req_addr != NULL) {
3230 if (start != req_addr) {
3231 ::munmap(start, extra_size);
3232 start = NULL;
3233 }
3234 } else {
3235 char* const start_aligned = align_up(start, alignment);
3236 char* const end_aligned = start_aligned + bytes;
3237 char* const end = start + extra_size;
3238 if (start_aligned > start) {
3239 ::munmap(start, start_aligned - start);
3240 }
3241 if (end_aligned < end) {
3242 ::munmap(end_aligned, end - end_aligned);
3243 }
3244 start = start_aligned;
3245 }
3246 }
3247 return start;
3248 }
3249
3250 static int anon_munmap(char * addr, size_t size) {
3251 return ::munmap(addr, size) == 0;
3252 }
3253
3254 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3255 size_t alignment_hint) {
3256 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3257 }
3258
3259 bool os::pd_release_memory(char* addr, size_t size) {
3260 return anon_munmap(addr, size);
3261 }
3262
3263 static bool linux_mprotect(char* addr, size_t size, int prot) {
3264 // Linux wants the mprotect address argument to be page aligned.
3265 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3266
3267 // According to SUSv3, mprotect() should only be used with mappings
3268 // established by mmap(), and mmap() always maps whole pages. Unaligned
3269 // 'addr' likely indicates problem in the VM (e.g. trying to change
3270 // protection of malloc'ed or statically allocated memory). Check the
3271 // caller if you hit this assert.
3272 assert(addr == bottom, "sanity check");
3273
3274 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3275 return ::mprotect(bottom, size, prot) == 0;
3276 }
3277
3278 // Set protections specified
3279 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3280 bool is_committed) {
3281 unsigned int p = 0;
3282 switch (prot) {
3283 case MEM_PROT_NONE: p = PROT_NONE; break;
3284 case MEM_PROT_READ: p = PROT_READ; break;
3285 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3286 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3287 default:
3288 ShouldNotReachHere();
3289 }
3290 // is_committed is unused.
3291 return linux_mprotect(addr, bytes, p);
3292 }
3293
3294 bool os::guard_memory(char* addr, size_t size) {
3295 return linux_mprotect(addr, size, PROT_NONE);
3296 }
3297
3298 bool os::unguard_memory(char* addr, size_t size) {
3299 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3300 }
3301
3302 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3303 size_t page_size) {
3304 bool result = false;
3305 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3306 MAP_ANONYMOUS|MAP_PRIVATE,
3307 -1, 0);
3308 if (p != MAP_FAILED) {
3309 void *aligned_p = align_up(p, page_size);
3310
3311 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3312
3313 munmap(p, page_size * 2);
3314 }
3315
3316 if (warn && !result) {
3317 warning("TransparentHugePages is not supported by the operating system.");
3318 }
3319
3320 return result;
3321 }
3322
3323 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3324 bool result = false;
3325 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3326 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3327 -1, 0);
3328
3329 if (p != MAP_FAILED) {
3330 // We don't know if this really is a huge page or not.
3331 FILE *fp = fopen("/proc/self/maps", "r");
3332 if (fp) {
3333 while (!feof(fp)) {
3334 char chars[257];
3335 long x = 0;
3336 if (fgets(chars, sizeof(chars), fp)) {
3337 if (sscanf(chars, "%lx-%*x", &x) == 1
3338 && x == (long)p) {
3339 if (strstr (chars, "hugepage")) {
3340 result = true;
3341 break;
3342 }
3343 }
3344 }
3345 }
3346 fclose(fp);
3347 }
3348 munmap(p, page_size);
3349 }
3350
3351 if (warn && !result) {
3352 warning("HugeTLBFS is not supported by the operating system.");
3353 }
3354
3355 return result;
3356 }
3357
3358 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3359 //
3360 // From the coredump_filter documentation:
3361 //
3362 // - (bit 0) anonymous private memory
3363 // - (bit 1) anonymous shared memory
3364 // - (bit 2) file-backed private memory
3365 // - (bit 3) file-backed shared memory
3366 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3367 // effective only if the bit 2 is cleared)
3368 // - (bit 5) hugetlb private memory
3369 // - (bit 6) hugetlb shared memory
3370 // - (bit 7) dax private memory
3371 // - (bit 8) dax shared memory
3372 //
3373 static void set_coredump_filter(bool largepages, bool dax_shared) {
3374 FILE *f;
3375 long cdm;
3376 bool filter_changed = false;
3377
3378 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3379 return;
3380 }
3381
3382 if (fscanf(f, "%lx", &cdm) != 1) {
3383 fclose(f);
3384 return;
3385 }
3386
3387 rewind(f);
3388
3389 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3390 cdm |= LARGEPAGES_BIT;
3391 filter_changed = true;
3392 }
3393 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3394 cdm |= DAX_SHARED_BIT;
3395 filter_changed = true;
3396 }
3397 if (filter_changed) {
3398 fprintf(f, "%#lx", cdm);
3399 }
3400
3401 fclose(f);
3402 }
3403
3404 // Large page support
3405
3406 static size_t _large_page_size = 0;
3407
3408 size_t os::Linux::find_large_page_size() {
3409 size_t large_page_size = 0;
3410
3411 // large_page_size on Linux is used to round up heap size. x86 uses either
3412 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3413 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3414 // page as large as 256M.
3415 //
3416 // Here we try to figure out page size by parsing /proc/meminfo and looking
3417 // for a line with the following format:
3418 // Hugepagesize: 2048 kB
3419 //
3420 // If we can't determine the value (e.g. /proc is not mounted, or the text
3421 // format has been changed), we'll use the largest page size supported by
3422 // the processor.
3423
3424 #ifndef ZERO
3425 large_page_size =
3426 AARCH64_ONLY(2 * M)
3427 AMD64_ONLY(2 * M)
3428 ARM32_ONLY(2 * M)
3429 IA32_ONLY(4 * M)
3430 IA64_ONLY(256 * M)
3431 PPC_ONLY(4 * M)
3432 S390_ONLY(1 * M)
3433 SPARC_ONLY(4 * M);
3434 #endif // ZERO
3435
3436 FILE *fp = fopen("/proc/meminfo", "r");
3437 if (fp) {
3438 while (!feof(fp)) {
3439 int x = 0;
3440 char buf[16];
3441 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3442 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3443 large_page_size = x * K;
3444 break;
3445 }
3446 } else {
3447 // skip to next line
3448 for (;;) {
3449 int ch = fgetc(fp);
3450 if (ch == EOF || ch == (int)'\n') break;
3451 }
3452 }
3453 }
3454 fclose(fp);
3455 }
3456
3457 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3458 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3459 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3460 proper_unit_for_byte_size(large_page_size));
3461 }
3462
3463 return large_page_size;
3464 }
3465
3466 size_t os::Linux::setup_large_page_size() {
3467 _large_page_size = Linux::find_large_page_size();
3468 const size_t default_page_size = (size_t)Linux::page_size();
3469 if (_large_page_size > default_page_size) {
3470 _page_sizes[0] = _large_page_size;
3471 _page_sizes[1] = default_page_size;
3472 _page_sizes[2] = 0;
3473 }
3474
3475 return _large_page_size;
3476 }
3477
3478 bool os::Linux::setup_large_page_type(size_t page_size) {
3479 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3480 FLAG_IS_DEFAULT(UseSHM) &&
3481 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3482
3483 // The type of large pages has not been specified by the user.
3484
3485 // Try UseHugeTLBFS and then UseSHM.
3486 UseHugeTLBFS = UseSHM = true;
3487
3488 // Don't try UseTransparentHugePages since there are known
3489 // performance issues with it turned on. This might change in the future.
3490 UseTransparentHugePages = false;
3491 }
3492
3493 if (UseTransparentHugePages) {
3494 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3495 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3496 UseHugeTLBFS = false;
3497 UseSHM = false;
3498 return true;
3499 }
3500 UseTransparentHugePages = false;
3501 }
3502
3503 if (UseHugeTLBFS) {
3504 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3505 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3506 UseSHM = false;
3507 return true;
3508 }
3509 UseHugeTLBFS = false;
3510 }
3511
3512 return UseSHM;
3513 }
3514
3515 void os::large_page_init() {
3516 if (!UseLargePages &&
3517 !UseTransparentHugePages &&
3518 !UseHugeTLBFS &&
3519 !UseSHM) {
3520 // Not using large pages.
3521 return;
3522 }
3523
3524 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3525 // The user explicitly turned off large pages.
3526 // Ignore the rest of the large pages flags.
3527 UseTransparentHugePages = false;
3528 UseHugeTLBFS = false;
3529 UseSHM = false;
3530 return;
3531 }
3532
3533 size_t large_page_size = Linux::setup_large_page_size();
3534 UseLargePages = Linux::setup_large_page_type(large_page_size);
3535
3536 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3537 }
3538
3539 #ifndef SHM_HUGETLB
3540 #define SHM_HUGETLB 04000
3541 #endif
3542
3543 #define shm_warning_format(format, ...) \
3544 do { \
3545 if (UseLargePages && \
3546 (!FLAG_IS_DEFAULT(UseLargePages) || \
3547 !FLAG_IS_DEFAULT(UseSHM) || \
3548 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3549 warning(format, __VA_ARGS__); \
3550 } \
3551 } while (0)
3552
3553 #define shm_warning(str) shm_warning_format("%s", str)
3554
3555 #define shm_warning_with_errno(str) \
3556 do { \
3557 int err = errno; \
3558 shm_warning_format(str " (error = %d)", err); \
3559 } while (0)
3560
3561 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3562 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3563
3564 if (!is_aligned(alignment, SHMLBA)) {
3565 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3566 return NULL;
3567 }
3568
3569 // To ensure that we get 'alignment' aligned memory from shmat,
3570 // we pre-reserve aligned virtual memory and then attach to that.
3571
3572 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3573 if (pre_reserved_addr == NULL) {
3574 // Couldn't pre-reserve aligned memory.
3575 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3576 return NULL;
3577 }
3578
3579 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3580 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3581
3582 if ((intptr_t)addr == -1) {
3583 int err = errno;
3584 shm_warning_with_errno("Failed to attach shared memory.");
3585
3586 assert(err != EACCES, "Unexpected error");
3587 assert(err != EIDRM, "Unexpected error");
3588 assert(err != EINVAL, "Unexpected error");
3589
3590 // Since we don't know if the kernel unmapped the pre-reserved memory area
3591 // we can't unmap it, since that would potentially unmap memory that was
3592 // mapped from other threads.
3593 return NULL;
3594 }
3595
3596 return addr;
3597 }
3598
3599 static char* shmat_at_address(int shmid, char* req_addr) {
3600 if (!is_aligned(req_addr, SHMLBA)) {
3601 assert(false, "Requested address needs to be SHMLBA aligned");
3602 return NULL;
3603 }
3604
3605 char* addr = (char*)shmat(shmid, req_addr, 0);
3606
3607 if ((intptr_t)addr == -1) {
3608 shm_warning_with_errno("Failed to attach shared memory.");
3609 return NULL;
3610 }
3611
3612 return addr;
3613 }
3614
3615 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3616 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3617 if (req_addr != NULL) {
3618 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3619 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3620 return shmat_at_address(shmid, req_addr);
3621 }
3622
3623 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3624 // return large page size aligned memory addresses when req_addr == NULL.
3625 // However, if the alignment is larger than the large page size, we have
3626 // to manually ensure that the memory returned is 'alignment' aligned.
3627 if (alignment > os::large_page_size()) {
3628 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3629 return shmat_with_alignment(shmid, bytes, alignment);
3630 } else {
3631 return shmat_at_address(shmid, NULL);
3632 }
3633 }
3634
3635 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3636 char* req_addr, bool exec) {
3637 // "exec" is passed in but not used. Creating the shared image for
3638 // the code cache doesn't have an SHM_X executable permission to check.
3639 assert(UseLargePages && UseSHM, "only for SHM large pages");
3640 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3641 assert(is_aligned(req_addr, alignment), "Unaligned address");
3642
3643 if (!is_aligned(bytes, os::large_page_size())) {
3644 return NULL; // Fallback to small pages.
3645 }
3646
3647 // Create a large shared memory region to attach to based on size.
3648 // Currently, size is the total size of the heap.
3649 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3650 if (shmid == -1) {
3651 // Possible reasons for shmget failure:
3652 // 1. shmmax is too small for Java heap.
3653 // > check shmmax value: cat /proc/sys/kernel/shmmax
3654 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3655 // 2. not enough large page memory.
3656 // > check available large pages: cat /proc/meminfo
3657 // > increase amount of large pages:
3658 // echo new_value > /proc/sys/vm/nr_hugepages
3659 // Note 1: different Linux may use different name for this property,
3660 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3661 // Note 2: it's possible there's enough physical memory available but
3662 // they are so fragmented after a long run that they can't
3663 // coalesce into large pages. Try to reserve large pages when
3664 // the system is still "fresh".
3665 shm_warning_with_errno("Failed to reserve shared memory.");
3666 return NULL;
3667 }
3668
3669 // Attach to the region.
3670 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3671
3672 // Remove shmid. If shmat() is successful, the actual shared memory segment
3673 // will be deleted when it's detached by shmdt() or when the process
3674 // terminates. If shmat() is not successful this will remove the shared
3675 // segment immediately.
3676 shmctl(shmid, IPC_RMID, NULL);
3677
3678 return addr;
3679 }
3680
3681 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3682 int error) {
3683 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3684
3685 bool warn_on_failure = UseLargePages &&
3686 (!FLAG_IS_DEFAULT(UseLargePages) ||
3687 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3688 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3689
3690 if (warn_on_failure) {
3691 char msg[128];
3692 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3693 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3694 warning("%s", msg);
3695 }
3696 }
3697
3698 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3699 char* req_addr,
3700 bool exec) {
3701 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3702 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3703 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3704
3705 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3706 char* addr = (char*)::mmap(req_addr, bytes, prot,
3707 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3708 -1, 0);
3709
3710 if (addr == MAP_FAILED) {
3711 warn_on_large_pages_failure(req_addr, bytes, errno);
3712 return NULL;
3713 }
3714
3715 assert(is_aligned(addr, os::large_page_size()), "Must be");
3716
3717 return addr;
3718 }
3719
3720 // Reserve memory using mmap(MAP_HUGETLB).
3721 // - bytes shall be a multiple of alignment.
3722 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3723 // - alignment sets the alignment at which memory shall be allocated.
3724 // It must be a multiple of allocation granularity.
3725 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3726 // req_addr or NULL.
3727 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3728 size_t alignment,
3729 char* req_addr,
3730 bool exec) {
3731 size_t large_page_size = os::large_page_size();
3732 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3733
3734 assert(is_aligned(req_addr, alignment), "Must be");
3735 assert(is_aligned(bytes, alignment), "Must be");
3736
3737 // First reserve - but not commit - the address range in small pages.
3738 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3739
3740 if (start == NULL) {
3741 return NULL;
3742 }
3743
3744 assert(is_aligned(start, alignment), "Must be");
3745
3746 char* end = start + bytes;
3747
3748 // Find the regions of the allocated chunk that can be promoted to large pages.
3749 char* lp_start = align_up(start, large_page_size);
3750 char* lp_end = align_down(end, large_page_size);
3751
3752 size_t lp_bytes = lp_end - lp_start;
3753
3754 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3755
3756 if (lp_bytes == 0) {
3757 // The mapped region doesn't even span the start and the end of a large page.
3758 // Fall back to allocate a non-special area.
3759 ::munmap(start, end - start);
3760 return NULL;
3761 }
3762
3763 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3764
3765 void* result;
3766
3767 // Commit small-paged leading area.
3768 if (start != lp_start) {
3769 result = ::mmap(start, lp_start - start, prot,
3770 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3771 -1, 0);
3772 if (result == MAP_FAILED) {
3773 ::munmap(lp_start, end - lp_start);
3774 return NULL;
3775 }
3776 }
3777
3778 // Commit large-paged area.
3779 result = ::mmap(lp_start, lp_bytes, prot,
3780 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3781 -1, 0);
3782 if (result == MAP_FAILED) {
3783 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3784 // If the mmap above fails, the large pages region will be unmapped and we
3785 // have regions before and after with small pages. Release these regions.
3786 //
3787 // | mapped | unmapped | mapped |
3788 // ^ ^ ^ ^
3789 // start lp_start lp_end end
3790 //
3791 ::munmap(start, lp_start - start);
3792 ::munmap(lp_end, end - lp_end);
3793 return NULL;
3794 }
3795
3796 // Commit small-paged trailing area.
3797 if (lp_end != end) {
3798 result = ::mmap(lp_end, end - lp_end, prot,
3799 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3800 -1, 0);
3801 if (result == MAP_FAILED) {
3802 ::munmap(start, lp_end - start);
3803 return NULL;
3804 }
3805 }
3806
3807 return start;
3808 }
3809
3810 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3811 size_t alignment,
3812 char* req_addr,
3813 bool exec) {
3814 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3815 assert(is_aligned(req_addr, alignment), "Must be");
3816 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3817 assert(is_power_of_2(os::large_page_size()), "Must be");
3818 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3819
3820 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3821 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3822 } else {
3823 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3824 }
3825 }
3826
3827 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3828 char* req_addr, bool exec) {
3829 assert(UseLargePages, "only for large pages");
3830
3831 char* addr;
3832 if (UseSHM) {
3833 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3834 } else {
3835 assert(UseHugeTLBFS, "must be");
3836 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3837 }
3838
3839 if (addr != NULL) {
3840 if (UseNUMAInterleaving) {
3841 numa_make_global(addr, bytes);
3842 }
3843
3844 // The memory is committed
3845 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3846 }
3847
3848 return addr;
3849 }
3850
3851 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3852 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3853 return shmdt(base) == 0;
3854 }
3855
3856 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3857 return pd_release_memory(base, bytes);
3858 }
3859
3860 bool os::release_memory_special(char* base, size_t bytes) {
3861 bool res;
3862 if (MemTracker::tracking_level() > NMT_minimal) {
3863 Tracker tkr(Tracker::release);
3864 res = os::Linux::release_memory_special_impl(base, bytes);
3865 if (res) {
3866 tkr.record((address)base, bytes);
3867 }
3868
3869 } else {
3870 res = os::Linux::release_memory_special_impl(base, bytes);
3871 }
3872 return res;
3873 }
3874
3875 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3876 assert(UseLargePages, "only for large pages");
3877 bool res;
3878
3879 if (UseSHM) {
3880 res = os::Linux::release_memory_special_shm(base, bytes);
3881 } else {
3882 assert(UseHugeTLBFS, "must be");
3883 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3884 }
3885 return res;
3886 }
3887
3888 size_t os::large_page_size() {
3889 return _large_page_size;
3890 }
3891
3892 // With SysV SHM the entire memory region must be allocated as shared
3893 // memory.
3894 // HugeTLBFS allows application to commit large page memory on demand.
3895 // However, when committing memory with HugeTLBFS fails, the region
3896 // that was supposed to be committed will lose the old reservation
3897 // and allow other threads to steal that memory region. Because of this
3898 // behavior we can't commit HugeTLBFS memory.
3899 bool os::can_commit_large_page_memory() {
3900 return UseTransparentHugePages;
3901 }
3902
3903 bool os::can_execute_large_page_memory() {
3904 return UseTransparentHugePages || UseHugeTLBFS;
3905 }
3906
3907 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3908 assert(file_desc >= 0, "file_desc is not valid");
3909 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3910 if (result != NULL) {
3911 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3912 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3913 }
3914 }
3915 return result;
3916 }
3917
3918 // Reserve memory at an arbitrary address, only if that area is
3919 // available (and not reserved for something else).
3920
3921 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3922 const int max_tries = 10;
3923 char* base[max_tries];
3924 size_t size[max_tries];
3925 const size_t gap = 0x000000;
3926
3927 // Assert only that the size is a multiple of the page size, since
3928 // that's all that mmap requires, and since that's all we really know
3929 // about at this low abstraction level. If we need higher alignment,
3930 // we can either pass an alignment to this method or verify alignment
3931 // in one of the methods further up the call chain. See bug 5044738.
3932 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3933
3934 // Repeatedly allocate blocks until the block is allocated at the
3935 // right spot.
3936
3937 // Linux mmap allows caller to pass an address as hint; give it a try first,
3938 // if kernel honors the hint then we can return immediately.
3939 char * addr = anon_mmap(requested_addr, bytes, false);
3940 if (addr == requested_addr) {
3941 return requested_addr;
3942 }
3943
3944 if (addr != NULL) {
3945 // mmap() is successful but it fails to reserve at the requested address
3946 anon_munmap(addr, bytes);
3947 }
3948
3949 int i;
3950 for (i = 0; i < max_tries; ++i) {
3951 base[i] = reserve_memory(bytes);
3952
3953 if (base[i] != NULL) {
3954 // Is this the block we wanted?
3955 if (base[i] == requested_addr) {
3956 size[i] = bytes;
3957 break;
3958 }
3959
3960 // Does this overlap the block we wanted? Give back the overlapped
3961 // parts and try again.
3962
3963 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3964 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3965 unmap_memory(base[i], top_overlap);
3966 base[i] += top_overlap;
3967 size[i] = bytes - top_overlap;
3968 } else {
3969 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3970 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3971 unmap_memory(requested_addr, bottom_overlap);
3972 size[i] = bytes - bottom_overlap;
3973 } else {
3974 size[i] = bytes;
3975 }
3976 }
3977 }
3978 }
3979
3980 // Give back the unused reserved pieces.
3981
3982 for (int j = 0; j < i; ++j) {
3983 if (base[j] != NULL) {
3984 unmap_memory(base[j], size[j]);
3985 }
3986 }
3987
3988 if (i < max_tries) {
3989 return requested_addr;
3990 } else {
3991 return NULL;
3992 }
3993 }
3994
3995 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3996 return ::read(fd, buf, nBytes);
3997 }
3998
3999 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4000 return ::pread(fd, buf, nBytes, offset);
4001 }
4002
4003 // Short sleep, direct OS call.
4004 //
4005 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4006 // sched_yield(2) will actually give up the CPU:
4007 //
4008 // * Alone on this pariticular CPU, keeps running.
4009 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4010 // (pre 2.6.39).
4011 //
4012 // So calling this with 0 is an alternative.
4013 //
4014 void os::naked_short_sleep(jlong ms) {
4015 struct timespec req;
4016
4017 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4018 req.tv_sec = 0;
4019 if (ms > 0) {
4020 req.tv_nsec = (ms % 1000) * 1000000;
4021 } else {
4022 req.tv_nsec = 1;
4023 }
4024
4025 nanosleep(&req, NULL);
4026
4027 return;
4028 }
4029
4030 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4031 void os::infinite_sleep() {
4032 while (true) { // sleep forever ...
4033 ::sleep(100); // ... 100 seconds at a time
4034 }
4035 }
4036
4037 // Used to convert frequent JVM_Yield() to nops
4038 bool os::dont_yield() {
4039 return DontYieldALot;
4040 }
4041
4042 void os::naked_yield() {
4043 sched_yield();
4044 }
4045
4046 ////////////////////////////////////////////////////////////////////////////////
4047 // thread priority support
4048
4049 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4050 // only supports dynamic priority, static priority must be zero. For real-time
4051 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4052 // However, for large multi-threaded applications, SCHED_RR is not only slower
4053 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4054 // of 5 runs - Sep 2005).
4055 //
4056 // The following code actually changes the niceness of kernel-thread/LWP. It
4057 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4058 // not the entire user process, and user level threads are 1:1 mapped to kernel
4059 // threads. It has always been the case, but could change in the future. For
4060 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4061 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4062
4063 int os::java_to_os_priority[CriticalPriority + 1] = {
4064 19, // 0 Entry should never be used
4065
4066 4, // 1 MinPriority
4067 3, // 2
4068 2, // 3
4069
4070 1, // 4
4071 0, // 5 NormPriority
4072 -1, // 6
4073
4074 -2, // 7
4075 -3, // 8
4076 -4, // 9 NearMaxPriority
4077
4078 -5, // 10 MaxPriority
4079
4080 -5 // 11 CriticalPriority
4081 };
4082
4083 static int prio_init() {
4084 if (ThreadPriorityPolicy == 1) {
4085 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4086 // if effective uid is not root. Perhaps, a more elegant way of doing
4087 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4088 if (geteuid() != 0) {
4089 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4090 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4091 }
4092 ThreadPriorityPolicy = 0;
4093 }
4094 }
4095 if (UseCriticalJavaThreadPriority) {
4096 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4097 }
4098 return 0;
4099 }
4100
4101 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4102 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4103
4104 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4105 return (ret == 0) ? OS_OK : OS_ERR;
4106 }
4107
4108 OSReturn os::get_native_priority(const Thread* const thread,
4109 int *priority_ptr) {
4110 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4111 *priority_ptr = java_to_os_priority[NormPriority];
4112 return OS_OK;
4113 }
4114
4115 errno = 0;
4116 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4117 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4118 }
4119
4120 // Hint to the underlying OS that a task switch would not be good.
4121 // Void return because it's a hint and can fail.
4122 void os::hint_no_preempt() {}
4123
4124 ////////////////////////////////////////////////////////////////////////////////
4125 // suspend/resume support
4126
4127 // The low-level signal-based suspend/resume support is a remnant from the
4128 // old VM-suspension that used to be for java-suspension, safepoints etc,
4129 // within hotspot. Currently used by JFR's OSThreadSampler
4130 //
4131 // The remaining code is greatly simplified from the more general suspension
4132 // code that used to be used.
4133 //
4134 // The protocol is quite simple:
4135 // - suspend:
4136 // - sends a signal to the target thread
4137 // - polls the suspend state of the osthread using a yield loop
4138 // - target thread signal handler (SR_handler) sets suspend state
4139 // and blocks in sigsuspend until continued
4140 // - resume:
4141 // - sets target osthread state to continue
4142 // - sends signal to end the sigsuspend loop in the SR_handler
4143 //
4144 // Note that the SR_lock plays no role in this suspend/resume protocol,
4145 // but is checked for NULL in SR_handler as a thread termination indicator.
4146 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4147 //
4148 // Note that resume_clear_context() and suspend_save_context() are needed
4149 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4150 // which in part is used by:
4151 // - Forte Analyzer: AsyncGetCallTrace()
4152 // - StackBanging: get_frame_at_stack_banging_point()
4153
4154 static void resume_clear_context(OSThread *osthread) {
4155 osthread->set_ucontext(NULL);
4156 osthread->set_siginfo(NULL);
4157 }
4158
4159 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4160 ucontext_t* context) {
4161 osthread->set_ucontext(context);
4162 osthread->set_siginfo(siginfo);
4163 }
4164
4165 // Handler function invoked when a thread's execution is suspended or
4166 // resumed. We have to be careful that only async-safe functions are
4167 // called here (Note: most pthread functions are not async safe and
4168 // should be avoided.)
4169 //
4170 // Note: sigwait() is a more natural fit than sigsuspend() from an
4171 // interface point of view, but sigwait() prevents the signal hander
4172 // from being run. libpthread would get very confused by not having
4173 // its signal handlers run and prevents sigwait()'s use with the
4174 // mutex granting granting signal.
4175 //
4176 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4177 //
4178 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4179 // Save and restore errno to avoid confusing native code with EINTR
4180 // after sigsuspend.
4181 int old_errno = errno;
4182
4183 Thread* thread = Thread::current_or_null_safe();
4184 assert(thread != NULL, "Missing current thread in SR_handler");
4185
4186 // On some systems we have seen signal delivery get "stuck" until the signal
4187 // mask is changed as part of thread termination. Check that the current thread
4188 // has not already terminated (via SR_lock()) - else the following assertion
4189 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4190 // destructor has completed.
4191
4192 if (thread->SR_lock() == NULL) {
4193 return;
4194 }
4195
4196 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4197
4198 OSThread* osthread = thread->osthread();
4199
4200 os::SuspendResume::State current = osthread->sr.state();
4201 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4202 suspend_save_context(osthread, siginfo, context);
4203
4204 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4205 os::SuspendResume::State state = osthread->sr.suspended();
4206 if (state == os::SuspendResume::SR_SUSPENDED) {
4207 sigset_t suspend_set; // signals for sigsuspend()
4208 sigemptyset(&suspend_set);
4209 // get current set of blocked signals and unblock resume signal
4210 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4211 sigdelset(&suspend_set, SR_signum);
4212
4213 sr_semaphore.signal();
4214 // wait here until we are resumed
4215 while (1) {
4216 sigsuspend(&suspend_set);
4217
4218 os::SuspendResume::State result = osthread->sr.running();
4219 if (result == os::SuspendResume::SR_RUNNING) {
4220 sr_semaphore.signal();
4221 break;
4222 }
4223 }
4224
4225 } else if (state == os::SuspendResume::SR_RUNNING) {
4226 // request was cancelled, continue
4227 } else {
4228 ShouldNotReachHere();
4229 }
4230
4231 resume_clear_context(osthread);
4232 } else if (current == os::SuspendResume::SR_RUNNING) {
4233 // request was cancelled, continue
4234 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4235 // ignore
4236 } else {
4237 // ignore
4238 }
4239
4240 errno = old_errno;
4241 }
4242
4243 static int SR_initialize() {
4244 struct sigaction act;
4245 char *s;
4246
4247 // Get signal number to use for suspend/resume
4248 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4249 int sig = ::strtol(s, 0, 10);
4250 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4251 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4252 SR_signum = sig;
4253 } else {
4254 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4255 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4256 }
4257 }
4258
4259 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4260 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4261
4262 sigemptyset(&SR_sigset);
4263 sigaddset(&SR_sigset, SR_signum);
4264
4265 // Set up signal handler for suspend/resume
4266 act.sa_flags = SA_RESTART|SA_SIGINFO;
4267 act.sa_handler = (void (*)(int)) SR_handler;
4268
4269 // SR_signum is blocked by default.
4270 // 4528190 - We also need to block pthread restart signal (32 on all
4271 // supported Linux platforms). Note that LinuxThreads need to block
4272 // this signal for all threads to work properly. So we don't have
4273 // to use hard-coded signal number when setting up the mask.
4274 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4275
4276 if (sigaction(SR_signum, &act, 0) == -1) {
4277 return -1;
4278 }
4279
4280 // Save signal flag
4281 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4282 return 0;
4283 }
4284
4285 static int sr_notify(OSThread* osthread) {
4286 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4287 assert_status(status == 0, status, "pthread_kill");
4288 return status;
4289 }
4290
4291 // "Randomly" selected value for how long we want to spin
4292 // before bailing out on suspending a thread, also how often
4293 // we send a signal to a thread we want to resume
4294 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4295 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4296
4297 // returns true on success and false on error - really an error is fatal
4298 // but this seems the normal response to library errors
4299 static bool do_suspend(OSThread* osthread) {
4300 assert(osthread->sr.is_running(), "thread should be running");
4301 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4302
4303 // mark as suspended and send signal
4304 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4305 // failed to switch, state wasn't running?
4306 ShouldNotReachHere();
4307 return false;
4308 }
4309
4310 if (sr_notify(osthread) != 0) {
4311 ShouldNotReachHere();
4312 }
4313
4314 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4315 while (true) {
4316 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4317 break;
4318 } else {
4319 // timeout
4320 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4321 if (cancelled == os::SuspendResume::SR_RUNNING) {
4322 return false;
4323 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4324 // make sure that we consume the signal on the semaphore as well
4325 sr_semaphore.wait();
4326 break;
4327 } else {
4328 ShouldNotReachHere();
4329 return false;
4330 }
4331 }
4332 }
4333
4334 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4335 return true;
4336 }
4337
4338 static void do_resume(OSThread* osthread) {
4339 assert(osthread->sr.is_suspended(), "thread should be suspended");
4340 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4341
4342 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4343 // failed to switch to WAKEUP_REQUEST
4344 ShouldNotReachHere();
4345 return;
4346 }
4347
4348 while (true) {
4349 if (sr_notify(osthread) == 0) {
4350 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4351 if (osthread->sr.is_running()) {
4352 return;
4353 }
4354 }
4355 } else {
4356 ShouldNotReachHere();
4357 }
4358 }
4359
4360 guarantee(osthread->sr.is_running(), "Must be running!");
4361 }
4362
4363 ///////////////////////////////////////////////////////////////////////////////////
4364 // signal handling (except suspend/resume)
4365
4366 // This routine may be used by user applications as a "hook" to catch signals.
4367 // The user-defined signal handler must pass unrecognized signals to this
4368 // routine, and if it returns true (non-zero), then the signal handler must
4369 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4370 // routine will never retun false (zero), but instead will execute a VM panic
4371 // routine kill the process.
4372 //
4373 // If this routine returns false, it is OK to call it again. This allows
4374 // the user-defined signal handler to perform checks either before or after
4375 // the VM performs its own checks. Naturally, the user code would be making
4376 // a serious error if it tried to handle an exception (such as a null check
4377 // or breakpoint) that the VM was generating for its own correct operation.
4378 //
4379 // This routine may recognize any of the following kinds of signals:
4380 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4381 // It should be consulted by handlers for any of those signals.
4382 //
4383 // The caller of this routine must pass in the three arguments supplied
4384 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4385 // field of the structure passed to sigaction(). This routine assumes that
4386 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4387 //
4388 // Note that the VM will print warnings if it detects conflicting signal
4389 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4390 //
4391 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4392 siginfo_t* siginfo,
4393 void* ucontext,
4394 int abort_if_unrecognized);
4395
4396 void signalHandler(int sig, siginfo_t* info, void* uc) {
4397 assert(info != NULL && uc != NULL, "it must be old kernel");
4398 int orig_errno = errno; // Preserve errno value over signal handler.
4399 JVM_handle_linux_signal(sig, info, uc, true);
4400 errno = orig_errno;
4401 }
4402
4403
4404 // This boolean allows users to forward their own non-matching signals
4405 // to JVM_handle_linux_signal, harmlessly.
4406 bool os::Linux::signal_handlers_are_installed = false;
4407
4408 // For signal-chaining
4409 struct sigaction sigact[NSIG];
4410 uint64_t sigs = 0;
4411 #if (64 < NSIG-1)
4412 #error "Not all signals can be encoded in sigs. Adapt its type!"
4413 #endif
4414 bool os::Linux::libjsig_is_loaded = false;
4415 typedef struct sigaction *(*get_signal_t)(int);
4416 get_signal_t os::Linux::get_signal_action = NULL;
4417
4418 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4419 struct sigaction *actp = NULL;
4420
4421 if (libjsig_is_loaded) {
4422 // Retrieve the old signal handler from libjsig
4423 actp = (*get_signal_action)(sig);
4424 }
4425 if (actp == NULL) {
4426 // Retrieve the preinstalled signal handler from jvm
4427 actp = get_preinstalled_handler(sig);
4428 }
4429
4430 return actp;
4431 }
4432
4433 static bool call_chained_handler(struct sigaction *actp, int sig,
4434 siginfo_t *siginfo, void *context) {
4435 // Call the old signal handler
4436 if (actp->sa_handler == SIG_DFL) {
4437 // It's more reasonable to let jvm treat it as an unexpected exception
4438 // instead of taking the default action.
4439 return false;
4440 } else if (actp->sa_handler != SIG_IGN) {
4441 if ((actp->sa_flags & SA_NODEFER) == 0) {
4442 // automaticlly block the signal
4443 sigaddset(&(actp->sa_mask), sig);
4444 }
4445
4446 sa_handler_t hand = NULL;
4447 sa_sigaction_t sa = NULL;
4448 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4449 // retrieve the chained handler
4450 if (siginfo_flag_set) {
4451 sa = actp->sa_sigaction;
4452 } else {
4453 hand = actp->sa_handler;
4454 }
4455
4456 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4457 actp->sa_handler = SIG_DFL;
4458 }
4459
4460 // try to honor the signal mask
4461 sigset_t oset;
4462 sigemptyset(&oset);
4463 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4464
4465 // call into the chained handler
4466 if (siginfo_flag_set) {
4467 (*sa)(sig, siginfo, context);
4468 } else {
4469 (*hand)(sig);
4470 }
4471
4472 // restore the signal mask
4473 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4474 }
4475 // Tell jvm's signal handler the signal is taken care of.
4476 return true;
4477 }
4478
4479 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4480 bool chained = false;
4481 // signal-chaining
4482 if (UseSignalChaining) {
4483 struct sigaction *actp = get_chained_signal_action(sig);
4484 if (actp != NULL) {
4485 chained = call_chained_handler(actp, sig, siginfo, context);
4486 }
4487 }
4488 return chained;
4489 }
4490
4491 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4492 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4493 return &sigact[sig];
4494 }
4495 return NULL;
4496 }
4497
4498 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4499 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4500 sigact[sig] = oldAct;
4501 sigs |= (uint64_t)1 << (sig-1);
4502 }
4503
4504 // for diagnostic
4505 int sigflags[NSIG];
4506
4507 int os::Linux::get_our_sigflags(int sig) {
4508 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4509 return sigflags[sig];
4510 }
4511
4512 void os::Linux::set_our_sigflags(int sig, int flags) {
4513 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4514 if (sig > 0 && sig < NSIG) {
4515 sigflags[sig] = flags;
4516 }
4517 }
4518
4519 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4520 // Check for overwrite.
4521 struct sigaction oldAct;
4522 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4523
4524 void* oldhand = oldAct.sa_sigaction
4525 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4526 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4527 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4528 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4529 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4530 if (AllowUserSignalHandlers || !set_installed) {
4531 // Do not overwrite; user takes responsibility to forward to us.
4532 return;
4533 } else if (UseSignalChaining) {
4534 // save the old handler in jvm
4535 save_preinstalled_handler(sig, oldAct);
4536 // libjsig also interposes the sigaction() call below and saves the
4537 // old sigaction on it own.
4538 } else {
4539 fatal("Encountered unexpected pre-existing sigaction handler "
4540 "%#lx for signal %d.", (long)oldhand, sig);
4541 }
4542 }
4543
4544 struct sigaction sigAct;
4545 sigfillset(&(sigAct.sa_mask));
4546 sigAct.sa_handler = SIG_DFL;
4547 if (!set_installed) {
4548 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4549 } else {
4550 sigAct.sa_sigaction = signalHandler;
4551 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4552 }
4553 // Save flags, which are set by ours
4554 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4555 sigflags[sig] = sigAct.sa_flags;
4556
4557 int ret = sigaction(sig, &sigAct, &oldAct);
4558 assert(ret == 0, "check");
4559
4560 void* oldhand2 = oldAct.sa_sigaction
4561 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4562 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4563 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4564 }
4565
4566 // install signal handlers for signals that HotSpot needs to
4567 // handle in order to support Java-level exception handling.
4568
4569 void os::Linux::install_signal_handlers() {
4570 if (!signal_handlers_are_installed) {
4571 signal_handlers_are_installed = true;
4572
4573 // signal-chaining
4574 typedef void (*signal_setting_t)();
4575 signal_setting_t begin_signal_setting = NULL;
4576 signal_setting_t end_signal_setting = NULL;
4577 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4578 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4579 if (begin_signal_setting != NULL) {
4580 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4581 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4582 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4583 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4584 libjsig_is_loaded = true;
4585 assert(UseSignalChaining, "should enable signal-chaining");
4586 }
4587 if (libjsig_is_loaded) {
4588 // Tell libjsig jvm is setting signal handlers
4589 (*begin_signal_setting)();
4590 }
4591
4592 set_signal_handler(SIGSEGV, true);
4593 set_signal_handler(SIGPIPE, true);
4594 set_signal_handler(SIGBUS, true);
4595 set_signal_handler(SIGILL, true);
4596 set_signal_handler(SIGFPE, true);
4597 #if defined(PPC64)
4598 set_signal_handler(SIGTRAP, true);
4599 #endif
4600 set_signal_handler(SIGXFSZ, true);
4601
4602 if (libjsig_is_loaded) {
4603 // Tell libjsig jvm finishes setting signal handlers
4604 (*end_signal_setting)();
4605 }
4606
4607 // We don't activate signal checker if libjsig is in place, we trust ourselves
4608 // and if UserSignalHandler is installed all bets are off.
4609 // Log that signal checking is off only if -verbose:jni is specified.
4610 if (CheckJNICalls) {
4611 if (libjsig_is_loaded) {
4612 if (PrintJNIResolving) {
4613 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4614 }
4615 check_signals = false;
4616 }
4617 if (AllowUserSignalHandlers) {
4618 if (PrintJNIResolving) {
4619 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4620 }
4621 check_signals = false;
4622 }
4623 }
4624 }
4625 }
4626
4627 // This is the fastest way to get thread cpu time on Linux.
4628 // Returns cpu time (user+sys) for any thread, not only for current.
4629 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4630 // It might work on 2.6.10+ with a special kernel/glibc patch.
4631 // For reference, please, see IEEE Std 1003.1-2004:
4632 // http://www.unix.org/single_unix_specification
4633
4634 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4635 struct timespec tp;
4636 int rc = os::Linux::clock_gettime(clockid, &tp);
4637 assert(rc == 0, "clock_gettime is expected to return 0 code");
4638
4639 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4640 }
4641
4642 void os::Linux::initialize_os_info() {
4643 assert(_os_version == 0, "OS info already initialized");
4644
4645 struct utsname _uname;
4646
4647 uint32_t major;
4648 uint32_t minor;
4649 uint32_t fix;
4650
4651 int rc;
4652
4653 // Kernel version is unknown if
4654 // verification below fails.
4655 _os_version = 0x01000000;
4656
4657 rc = uname(&_uname);
4658 if (rc != -1) {
4659
4660 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4661 if (rc == 3) {
4662
4663 if (major < 256 && minor < 256 && fix < 256) {
4664 // Kernel version format is as expected,
4665 // set it overriding unknown state.
4666 _os_version = (major << 16) |
4667 (minor << 8 ) |
4668 (fix << 0 ) ;
4669 }
4670 }
4671 }
4672 }
4673
4674 uint32_t os::Linux::os_version() {
4675 assert(_os_version != 0, "not initialized");
4676 return _os_version & 0x00FFFFFF;
4677 }
4678
4679 bool os::Linux::os_version_is_known() {
4680 assert(_os_version != 0, "not initialized");
4681 return _os_version & 0x01000000 ? false : true;
4682 }
4683
4684 /////
4685 // glibc on Linux platform uses non-documented flag
4686 // to indicate, that some special sort of signal
4687 // trampoline is used.
4688 // We will never set this flag, and we should
4689 // ignore this flag in our diagnostic
4690 #ifdef SIGNIFICANT_SIGNAL_MASK
4691 #undef SIGNIFICANT_SIGNAL_MASK
4692 #endif
4693 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4694
4695 static const char* get_signal_handler_name(address handler,
4696 char* buf, int buflen) {
4697 int offset = 0;
4698 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4699 if (found) {
4700 // skip directory names
4701 const char *p1, *p2;
4702 p1 = buf;
4703 size_t len = strlen(os::file_separator());
4704 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4705 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4706 } else {
4707 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4708 }
4709 return buf;
4710 }
4711
4712 static void print_signal_handler(outputStream* st, int sig,
4713 char* buf, size_t buflen) {
4714 struct sigaction sa;
4715
4716 sigaction(sig, NULL, &sa);
4717
4718 // See comment for SIGNIFICANT_SIGNAL_MASK define
4719 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4720
4721 st->print("%s: ", os::exception_name(sig, buf, buflen));
4722
4723 address handler = (sa.sa_flags & SA_SIGINFO)
4724 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4725 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4726
4727 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4728 st->print("SIG_DFL");
4729 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4730 st->print("SIG_IGN");
4731 } else {
4732 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4733 }
4734
4735 st->print(", sa_mask[0]=");
4736 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4737
4738 address rh = VMError::get_resetted_sighandler(sig);
4739 // May be, handler was resetted by VMError?
4740 if (rh != NULL) {
4741 handler = rh;
4742 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4743 }
4744
4745 st->print(", sa_flags=");
4746 os::Posix::print_sa_flags(st, sa.sa_flags);
4747
4748 // Check: is it our handler?
4749 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4750 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4751 // It is our signal handler
4752 // check for flags, reset system-used one!
4753 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4754 st->print(
4755 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4756 os::Linux::get_our_sigflags(sig));
4757 }
4758 }
4759 st->cr();
4760 }
4761
4762
4763 #define DO_SIGNAL_CHECK(sig) \
4764 do { \
4765 if (!sigismember(&check_signal_done, sig)) { \
4766 os::Linux::check_signal_handler(sig); \
4767 } \
4768 } while (0)
4769
4770 // This method is a periodic task to check for misbehaving JNI applications
4771 // under CheckJNI, we can add any periodic checks here
4772
4773 void os::run_periodic_checks() {
4774 if (check_signals == false) return;
4775
4776 // SEGV and BUS if overridden could potentially prevent
4777 // generation of hs*.log in the event of a crash, debugging
4778 // such a case can be very challenging, so we absolutely
4779 // check the following for a good measure:
4780 DO_SIGNAL_CHECK(SIGSEGV);
4781 DO_SIGNAL_CHECK(SIGILL);
4782 DO_SIGNAL_CHECK(SIGFPE);
4783 DO_SIGNAL_CHECK(SIGBUS);
4784 DO_SIGNAL_CHECK(SIGPIPE);
4785 DO_SIGNAL_CHECK(SIGXFSZ);
4786 #if defined(PPC64)
4787 DO_SIGNAL_CHECK(SIGTRAP);
4788 #endif
4789
4790 // ReduceSignalUsage allows the user to override these handlers
4791 // see comments at the very top and jvm_md.h
4792 if (!ReduceSignalUsage) {
4793 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4794 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4795 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4796 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4797 }
4798
4799 DO_SIGNAL_CHECK(SR_signum);
4800 }
4801
4802 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4803
4804 static os_sigaction_t os_sigaction = NULL;
4805
4806 void os::Linux::check_signal_handler(int sig) {
4807 char buf[O_BUFLEN];
4808 address jvmHandler = NULL;
4809
4810
4811 struct sigaction act;
4812 if (os_sigaction == NULL) {
4813 // only trust the default sigaction, in case it has been interposed
4814 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4815 if (os_sigaction == NULL) return;
4816 }
4817
4818 os_sigaction(sig, (struct sigaction*)NULL, &act);
4819
4820
4821 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4822
4823 address thisHandler = (act.sa_flags & SA_SIGINFO)
4824 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4825 : CAST_FROM_FN_PTR(address, act.sa_handler);
4826
4827
4828 switch (sig) {
4829 case SIGSEGV:
4830 case SIGBUS:
4831 case SIGFPE:
4832 case SIGPIPE:
4833 case SIGILL:
4834 case SIGXFSZ:
4835 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4836 break;
4837
4838 case SHUTDOWN1_SIGNAL:
4839 case SHUTDOWN2_SIGNAL:
4840 case SHUTDOWN3_SIGNAL:
4841 case BREAK_SIGNAL:
4842 jvmHandler = (address)user_handler();
4843 break;
4844
4845 default:
4846 if (sig == SR_signum) {
4847 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4848 } else {
4849 return;
4850 }
4851 break;
4852 }
4853
4854 if (thisHandler != jvmHandler) {
4855 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4856 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4857 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4858 // No need to check this sig any longer
4859 sigaddset(&check_signal_done, sig);
4860 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4861 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4862 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4863 exception_name(sig, buf, O_BUFLEN));
4864 }
4865 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4866 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4867 tty->print("expected:");
4868 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4869 tty->cr();
4870 tty->print(" found:");
4871 os::Posix::print_sa_flags(tty, act.sa_flags);
4872 tty->cr();
4873 // No need to check this sig any longer
4874 sigaddset(&check_signal_done, sig);
4875 }
4876
4877 // Dump all the signal
4878 if (sigismember(&check_signal_done, sig)) {
4879 print_signal_handlers(tty, buf, O_BUFLEN);
4880 }
4881 }
4882
4883 extern void report_error(char* file_name, int line_no, char* title,
4884 char* format, ...);
4885
4886 // this is called _before_ most of the global arguments have been parsed
4887 void os::init(void) {
4888 char dummy; // used to get a guess on initial stack address
4889
4890 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4891
4892 init_random(1234567);
4893
4894 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4895 if (Linux::page_size() == -1) {
4896 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4897 os::strerror(errno));
4898 }
4899 init_page_sizes((size_t) Linux::page_size());
4900
4901 Linux::initialize_system_info();
4902
4903 Linux::initialize_os_info();
4904
4905 // _main_thread points to the thread that created/loaded the JVM.
4906 Linux::_main_thread = pthread_self();
4907
4908 Linux::clock_init();
4909 initial_time_count = javaTimeNanos();
4910
4911 // retrieve entry point for pthread_setname_np
4912 Linux::_pthread_setname_np =
4913 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4914
4915 os::Posix::init();
4916 }
4917
4918 // To install functions for atexit system call
4919 extern "C" {
4920 static void perfMemory_exit_helper() {
4921 perfMemory_exit();
4922 }
4923 }
4924
4925 void os::pd_init_container_support() {
4926 OSContainer::init();
4927 }
4928
4929 // this is called _after_ the global arguments have been parsed
4930 jint os::init_2(void) {
4931
4932 os::Posix::init_2();
4933
4934 Linux::fast_thread_clock_init();
4935
4936 // initialize suspend/resume support - must do this before signal_sets_init()
4937 if (SR_initialize() != 0) {
4938 perror("SR_initialize failed");
4939 return JNI_ERR;
4940 }
4941
4942 Linux::signal_sets_init();
4943 Linux::install_signal_handlers();
4944
4945 // Check and sets minimum stack sizes against command line options
4946 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4947 return JNI_ERR;
4948 }
4949
4950 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
4951 if (!suppress_primordial_thread_resolution) {
4952 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4953 }
4954
4955 #if defined(IA32)
4956 workaround_expand_exec_shield_cs_limit();
4957 #endif
4958
4959 Linux::libpthread_init();
4960 Linux::sched_getcpu_init();
4961 log_info(os)("HotSpot is running with %s, %s",
4962 Linux::glibc_version(), Linux::libpthread_version());
4963
4964 if (UseNUMA) {
4965 if (!Linux::libnuma_init()) {
4966 UseNUMA = false;
4967 } else {
4968 if ((Linux::numa_max_node() < 1)) {
4969 // There's only one node(they start from 0), disable NUMA.
4970 UseNUMA = false;
4971 }
4972 }
4973
4974 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4975 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4976 // we can make the adaptive lgrp chunk resizing work. If the user specified both
4977 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
4978 // and disable adaptive resizing.
4979 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4980 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
4981 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4982 UseAdaptiveSizePolicy = false;
4983 UseAdaptiveNUMAChunkSizing = false;
4984 }
4985 }
4986
4987 if (!UseNUMA && ForceNUMA) {
4988 UseNUMA = true;
4989 }
4990 }
4991
4992 if (MaxFDLimit) {
4993 // set the number of file descriptors to max. print out error
4994 // if getrlimit/setrlimit fails but continue regardless.
4995 struct rlimit nbr_files;
4996 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4997 if (status != 0) {
4998 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4999 } else {
5000 nbr_files.rlim_cur = nbr_files.rlim_max;
5001 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5002 if (status != 0) {
5003 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5004 }
5005 }
5006 }
5007
5008 // Initialize lock used to serialize thread creation (see os::create_thread)
5009 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5010
5011 // at-exit methods are called in the reverse order of their registration.
5012 // atexit functions are called on return from main or as a result of a
5013 // call to exit(3C). There can be only 32 of these functions registered
5014 // and atexit() does not set errno.
5015
5016 if (PerfAllowAtExitRegistration) {
5017 // only register atexit functions if PerfAllowAtExitRegistration is set.
5018 // atexit functions can be delayed until process exit time, which
5019 // can be problematic for embedded VM situations. Embedded VMs should
5020 // call DestroyJavaVM() to assure that VM resources are released.
5021
5022 // note: perfMemory_exit_helper atexit function may be removed in
5023 // the future if the appropriate cleanup code can be added to the
5024 // VM_Exit VMOperation's doit method.
5025 if (atexit(perfMemory_exit_helper) != 0) {
5026 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5027 }
5028 }
5029
5030 // initialize thread priority policy
5031 prio_init();
5032
5033 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5034 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5035 }
5036 return JNI_OK;
5037 }
5038
5039 // Mark the polling page as unreadable
5040 void os::make_polling_page_unreadable(void) {
5041 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5042 fatal("Could not disable polling page");
5043 }
5044 }
5045
5046 // Mark the polling page as readable
5047 void os::make_polling_page_readable(void) {
5048 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5049 fatal("Could not enable polling page");
5050 }
5051 }
5052
5053 // older glibc versions don't have this macro (which expands to
5054 // an optimized bit-counting function) so we have to roll our own
5055 #ifndef CPU_COUNT
5056
5057 static int _cpu_count(const cpu_set_t* cpus) {
5058 int count = 0;
5059 // only look up to the number of configured processors
5060 for (int i = 0; i < os::processor_count(); i++) {
5061 if (CPU_ISSET(i, cpus)) {
5062 count++;
5063 }
5064 }
5065 return count;
5066 }
5067
5068 #define CPU_COUNT(cpus) _cpu_count(cpus)
5069
5070 #endif // CPU_COUNT
5071
5072 // Get the current number of available processors for this process.
5073 // This value can change at any time during a process's lifetime.
5074 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5075 // If it appears there may be more than 1024 processors then we do a
5076 // dynamic check - see 6515172 for details.
5077 // If anything goes wrong we fallback to returning the number of online
5078 // processors - which can be greater than the number available to the process.
5079 int os::Linux::active_processor_count() {
5080 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5081 cpu_set_t* cpus_p = &cpus;
5082 int cpus_size = sizeof(cpu_set_t);
5083
5084 int configured_cpus = os::processor_count(); // upper bound on available cpus
5085 int cpu_count = 0;
5086
5087 // old build platforms may not support dynamic cpu sets
5088 #ifdef CPU_ALLOC
5089
5090 // To enable easy testing of the dynamic path on different platforms we
5091 // introduce a diagnostic flag: UseCpuAllocPath
5092 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5093 // kernel may use a mask bigger than cpu_set_t
5094 log_trace(os)("active_processor_count: using dynamic path %s"
5095 "- configured processors: %d",
5096 UseCpuAllocPath ? "(forced) " : "",
5097 configured_cpus);
5098 cpus_p = CPU_ALLOC(configured_cpus);
5099 if (cpus_p != NULL) {
5100 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5101 // zero it just to be safe
5102 CPU_ZERO_S(cpus_size, cpus_p);
5103 }
5104 else {
5105 // failed to allocate so fallback to online cpus
5106 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5107 log_trace(os)("active_processor_count: "
5108 "CPU_ALLOC failed (%s) - using "
5109 "online processor count: %d",
5110 os::strerror(errno), online_cpus);
5111 return online_cpus;
5112 }
5113 }
5114 else {
5115 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5116 configured_cpus);
5117 }
5118 #else // CPU_ALLOC
5119 // these stubs won't be executed
5120 #define CPU_COUNT_S(size, cpus) -1
5121 #define CPU_FREE(cpus)
5122
5123 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5124 configured_cpus);
5125 #endif // CPU_ALLOC
5126
5127 // pid 0 means the current thread - which we have to assume represents the process
5128 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5129 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5130 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5131 }
5132 else {
5133 cpu_count = CPU_COUNT(cpus_p);
5134 }
5135 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5136 }
5137 else {
5138 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5139 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5140 "which may exceed available processors", os::strerror(errno), cpu_count);
5141 }
5142
5143 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5144 CPU_FREE(cpus_p);
5145 }
5146
5147 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5148 return cpu_count;
5149 }
5150
5151 // Determine the active processor count from one of
5152 // three different sources:
5153 //
5154 // 1. User option -XX:ActiveProcessorCount
5155 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5156 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5157 //
5158 // Option 1, if specified, will always override.
5159 // If the cgroup subsystem is active and configured, we
5160 // will return the min of the cgroup and option 2 results.
5161 // This is required since tools, such as numactl, that
5162 // alter cpu affinity do not update cgroup subsystem
5163 // cpuset configuration files.
5164 int os::active_processor_count() {
5165 // User has overridden the number of active processors
5166 if (ActiveProcessorCount > 0) {
5167 log_trace(os)("active_processor_count: "
5168 "active processor count set by user : %d",
5169 ActiveProcessorCount);
5170 return ActiveProcessorCount;
5171 }
5172
5173 int active_cpus;
5174 if (OSContainer::is_containerized()) {
5175 active_cpus = OSContainer::active_processor_count();
5176 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5177 active_cpus);
5178 } else {
5179 active_cpus = os::Linux::active_processor_count();
5180 }
5181
5182 return active_cpus;
5183 }
5184
5185 void os::set_native_thread_name(const char *name) {
5186 if (Linux::_pthread_setname_np) {
5187 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5188 snprintf(buf, sizeof(buf), "%s", name);
5189 buf[sizeof(buf) - 1] = '\0';
5190 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5191 // ERANGE should not happen; all other errors should just be ignored.
5192 assert(rc != ERANGE, "pthread_setname_np failed");
5193 }
5194 }
5195
5196 bool os::distribute_processes(uint length, uint* distribution) {
5197 // Not yet implemented.
5198 return false;
5199 }
5200
5201 bool os::bind_to_processor(uint processor_id) {
5202 // Not yet implemented.
5203 return false;
5204 }
5205
5206 ///
5207
5208 void os::SuspendedThreadTask::internal_do_task() {
5209 if (do_suspend(_thread->osthread())) {
5210 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5211 do_task(context);
5212 do_resume(_thread->osthread());
5213 }
5214 }
5215
5216 ////////////////////////////////////////////////////////////////////////////////
5217 // debug support
5218
5219 bool os::find(address addr, outputStream* st) {
5220 Dl_info dlinfo;
5221 memset(&dlinfo, 0, sizeof(dlinfo));
5222 if (dladdr(addr, &dlinfo) != 0) {
5223 st->print(PTR_FORMAT ": ", p2i(addr));
5224 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5225 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5226 p2i(addr) - p2i(dlinfo.dli_saddr));
5227 } else if (dlinfo.dli_fbase != NULL) {
5228 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5229 } else {
5230 st->print("<absolute address>");
5231 }
5232 if (dlinfo.dli_fname != NULL) {
5233 st->print(" in %s", dlinfo.dli_fname);
5234 }
5235 if (dlinfo.dli_fbase != NULL) {
5236 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5237 }
5238 st->cr();
5239
5240 if (Verbose) {
5241 // decode some bytes around the PC
5242 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5243 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5244 address lowest = (address) dlinfo.dli_sname;
5245 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5246 if (begin < lowest) begin = lowest;
5247 Dl_info dlinfo2;
5248 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5249 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5250 end = (address) dlinfo2.dli_saddr;
5251 }
5252 Disassembler::decode(begin, end, st);
5253 }
5254 return true;
5255 }
5256 return false;
5257 }
5258
5259 ////////////////////////////////////////////////////////////////////////////////
5260 // misc
5261
5262 // This does not do anything on Linux. This is basically a hook for being
5263 // able to use structured exception handling (thread-local exception filters)
5264 // on, e.g., Win32.
5265 void
5266 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5267 JavaCallArguments* args, Thread* thread) {
5268 f(value, method, args, thread);
5269 }
5270
5271 void os::print_statistics() {
5272 }
5273
5274 bool os::message_box(const char* title, const char* message) {
5275 int i;
5276 fdStream err(defaultStream::error_fd());
5277 for (i = 0; i < 78; i++) err.print_raw("=");
5278 err.cr();
5279 err.print_raw_cr(title);
5280 for (i = 0; i < 78; i++) err.print_raw("-");
5281 err.cr();
5282 err.print_raw_cr(message);
5283 for (i = 0; i < 78; i++) err.print_raw("=");
5284 err.cr();
5285
5286 char buf[16];
5287 // Prevent process from exiting upon "read error" without consuming all CPU
5288 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5289
5290 return buf[0] == 'y' || buf[0] == 'Y';
5291 }
5292
5293 int os::stat(const char *path, struct stat *sbuf) {
5294 char pathbuf[MAX_PATH];
5295 if (strlen(path) > MAX_PATH - 1) {
5296 errno = ENAMETOOLONG;
5297 return -1;
5298 }
5299 os::native_path(strcpy(pathbuf, path));
5300 return ::stat(pathbuf, sbuf);
5301 }
5302
5303 // Is a (classpath) directory empty?
5304 bool os::dir_is_empty(const char* path) {
5305 DIR *dir = NULL;
5306 struct dirent *ptr;
5307
5308 dir = opendir(path);
5309 if (dir == NULL) return true;
5310
5311 // Scan the directory
5312 bool result = true;
5313 char buf[sizeof(struct dirent) + MAX_PATH];
5314 while (result && (ptr = ::readdir(dir)) != NULL) {
5315 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5316 result = false;
5317 }
5318 }
5319 closedir(dir);
5320 return result;
5321 }
5322
5323 // This code originates from JDK's sysOpen and open64_w
5324 // from src/solaris/hpi/src/system_md.c
5325
5326 int os::open(const char *path, int oflag, int mode) {
5327 if (strlen(path) > MAX_PATH - 1) {
5328 errno = ENAMETOOLONG;
5329 return -1;
5330 }
5331
5332 // All file descriptors that are opened in the Java process and not
5333 // specifically destined for a subprocess should have the close-on-exec
5334 // flag set. If we don't set it, then careless 3rd party native code
5335 // might fork and exec without closing all appropriate file descriptors
5336 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5337 // turn might:
5338 //
5339 // - cause end-of-file to fail to be detected on some file
5340 // descriptors, resulting in mysterious hangs, or
5341 //
5342 // - might cause an fopen in the subprocess to fail on a system
5343 // suffering from bug 1085341.
5344 //
5345 // (Yes, the default setting of the close-on-exec flag is a Unix
5346 // design flaw)
5347 //
5348 // See:
5349 // 1085341: 32-bit stdio routines should support file descriptors >255
5350 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5351 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5352 //
5353 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5354 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5355 // because it saves a system call and removes a small window where the flag
5356 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5357 // and we fall back to using FD_CLOEXEC (see below).
5358 #ifdef O_CLOEXEC
5359 oflag |= O_CLOEXEC;
5360 #endif
5361
5362 int fd = ::open64(path, oflag, mode);
5363 if (fd == -1) return -1;
5364
5365 //If the open succeeded, the file might still be a directory
5366 {
5367 struct stat64 buf64;
5368 int ret = ::fstat64(fd, &buf64);
5369 int st_mode = buf64.st_mode;
5370
5371 if (ret != -1) {
5372 if ((st_mode & S_IFMT) == S_IFDIR) {
5373 errno = EISDIR;
5374 ::close(fd);
5375 return -1;
5376 }
5377 } else {
5378 ::close(fd);
5379 return -1;
5380 }
5381 }
5382
5383 #ifdef FD_CLOEXEC
5384 // Validate that the use of the O_CLOEXEC flag on open above worked.
5385 // With recent kernels, we will perform this check exactly once.
5386 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5387 if (!O_CLOEXEC_is_known_to_work) {
5388 int flags = ::fcntl(fd, F_GETFD);
5389 if (flags != -1) {
5390 if ((flags & FD_CLOEXEC) != 0)
5391 O_CLOEXEC_is_known_to_work = 1;
5392 else
5393 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5394 }
5395 }
5396 #endif
5397
5398 return fd;
5399 }
5400
5401
5402 // create binary file, rewriting existing file if required
5403 int os::create_binary_file(const char* path, bool rewrite_existing) {
5404 int oflags = O_WRONLY | O_CREAT;
5405 if (!rewrite_existing) {
5406 oflags |= O_EXCL;
5407 }
5408 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5409 }
5410
5411 // return current position of file pointer
5412 jlong os::current_file_offset(int fd) {
5413 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5414 }
5415
5416 // move file pointer to the specified offset
5417 jlong os::seek_to_file_offset(int fd, jlong offset) {
5418 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5419 }
5420
5421 // This code originates from JDK's sysAvailable
5422 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5423
5424 int os::available(int fd, jlong *bytes) {
5425 jlong cur, end;
5426 int mode;
5427 struct stat64 buf64;
5428
5429 if (::fstat64(fd, &buf64) >= 0) {
5430 mode = buf64.st_mode;
5431 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5432 int n;
5433 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5434 *bytes = n;
5435 return 1;
5436 }
5437 }
5438 }
5439 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5440 return 0;
5441 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5442 return 0;
5443 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5444 return 0;
5445 }
5446 *bytes = end - cur;
5447 return 1;
5448 }
5449
5450 // Map a block of memory.
5451 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5452 char *addr, size_t bytes, bool read_only,
5453 bool allow_exec) {
5454 int prot;
5455 int flags = MAP_PRIVATE;
5456
5457 if (read_only) {
5458 prot = PROT_READ;
5459 } else {
5460 prot = PROT_READ | PROT_WRITE;
5461 }
5462
5463 if (allow_exec) {
5464 prot |= PROT_EXEC;
5465 }
5466
5467 if (addr != NULL) {
5468 flags |= MAP_FIXED;
5469 }
5470
5471 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5472 fd, file_offset);
5473 if (mapped_address == MAP_FAILED) {
5474 return NULL;
5475 }
5476 return mapped_address;
5477 }
5478
5479
5480 // Remap a block of memory.
5481 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5482 char *addr, size_t bytes, bool read_only,
5483 bool allow_exec) {
5484 // same as map_memory() on this OS
5485 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5486 allow_exec);
5487 }
5488
5489
5490 // Unmap a block of memory.
5491 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5492 return munmap(addr, bytes) == 0;
5493 }
5494
5495 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5496
5497 static clockid_t thread_cpu_clockid(Thread* thread) {
5498 pthread_t tid = thread->osthread()->pthread_id();
5499 clockid_t clockid;
5500
5501 // Get thread clockid
5502 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5503 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5504 return clockid;
5505 }
5506
5507 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5508 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5509 // of a thread.
5510 //
5511 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5512 // the fast estimate available on the platform.
5513
5514 jlong os::current_thread_cpu_time() {
5515 if (os::Linux::supports_fast_thread_cpu_time()) {
5516 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5517 } else {
5518 // return user + sys since the cost is the same
5519 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5520 }
5521 }
5522
5523 jlong os::thread_cpu_time(Thread* thread) {
5524 // consistent with what current_thread_cpu_time() returns
5525 if (os::Linux::supports_fast_thread_cpu_time()) {
5526 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5527 } else {
5528 return slow_thread_cpu_time(thread, true /* user + sys */);
5529 }
5530 }
5531
5532 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5533 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5534 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5535 } else {
5536 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5537 }
5538 }
5539
5540 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5541 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5542 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5543 } else {
5544 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5545 }
5546 }
5547
5548 // -1 on error.
5549 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5550 pid_t tid = thread->osthread()->thread_id();
5551 char *s;
5552 char stat[2048];
5553 int statlen;
5554 char proc_name[64];
5555 int count;
5556 long sys_time, user_time;
5557 char cdummy;
5558 int idummy;
5559 long ldummy;
5560 FILE *fp;
5561
5562 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5563 fp = fopen(proc_name, "r");
5564 if (fp == NULL) return -1;
5565 statlen = fread(stat, 1, 2047, fp);
5566 stat[statlen] = '\0';
5567 fclose(fp);
5568
5569 // Skip pid and the command string. Note that we could be dealing with
5570 // weird command names, e.g. user could decide to rename java launcher
5571 // to "java 1.4.2 :)", then the stat file would look like
5572 // 1234 (java 1.4.2 :)) R ... ...
5573 // We don't really need to know the command string, just find the last
5574 // occurrence of ")" and then start parsing from there. See bug 4726580.
5575 s = strrchr(stat, ')');
5576 if (s == NULL) return -1;
5577
5578 // Skip blank chars
5579 do { s++; } while (s && isspace(*s));
5580
5581 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5582 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5583 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5584 &user_time, &sys_time);
5585 if (count != 13) return -1;
5586 if (user_sys_cpu_time) {
5587 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5588 } else {
5589 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5590 }
5591 }
5592
5593 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5594 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5595 info_ptr->may_skip_backward = false; // elapsed time not wall time
5596 info_ptr->may_skip_forward = false; // elapsed time not wall time
5597 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5598 }
5599
5600 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5601 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5602 info_ptr->may_skip_backward = false; // elapsed time not wall time
5603 info_ptr->may_skip_forward = false; // elapsed time not wall time
5604 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5605 }
5606
5607 bool os::is_thread_cpu_time_supported() {
5608 return true;
5609 }
5610
5611 // System loadavg support. Returns -1 if load average cannot be obtained.
5612 // Linux doesn't yet have a (official) notion of processor sets,
5613 // so just return the system wide load average.
5614 int os::loadavg(double loadavg[], int nelem) {
5615 return ::getloadavg(loadavg, nelem);
5616 }
5617
5618 void os::pause() {
5619 char filename[MAX_PATH];
5620 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5621 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5622 } else {
5623 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5624 }
5625
5626 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5627 if (fd != -1) {
5628 struct stat buf;
5629 ::close(fd);
5630 while (::stat(filename, &buf) == 0) {
5631 (void)::poll(NULL, 0, 100);
5632 }
5633 } else {
5634 jio_fprintf(stderr,
5635 "Could not open pause file '%s', continuing immediately.\n", filename);
5636 }
5637 }
5638
5639 extern char** environ;
5640
5641 // Run the specified command in a separate process. Return its exit value,
5642 // or -1 on failure (e.g. can't fork a new process).
5643 // Unlike system(), this function can be called from signal handler. It
5644 // doesn't block SIGINT et al.
5645 int os::fork_and_exec(char* cmd) {
5646 const char * argv[4] = {"sh", "-c", cmd, NULL};
5647
5648 pid_t pid = fork();
5649
5650 if (pid < 0) {
5651 // fork failed
5652 return -1;
5653
5654 } else if (pid == 0) {
5655 // child process
5656
5657 execve("/bin/sh", (char* const*)argv, environ);
5658
5659 // execve failed
5660 _exit(-1);
5661
5662 } else {
5663 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5664 // care about the actual exit code, for now.
5665
5666 int status;
5667
5668 // Wait for the child process to exit. This returns immediately if
5669 // the child has already exited. */
5670 while (waitpid(pid, &status, 0) < 0) {
5671 switch (errno) {
5672 case ECHILD: return 0;
5673 case EINTR: break;
5674 default: return -1;
5675 }
5676 }
5677
5678 if (WIFEXITED(status)) {
5679 // The child exited normally; get its exit code.
5680 return WEXITSTATUS(status);
5681 } else if (WIFSIGNALED(status)) {
5682 // The child exited because of a signal
5683 // The best value to return is 0x80 + signal number,
5684 // because that is what all Unix shells do, and because
5685 // it allows callers to distinguish between process exit and
5686 // process death by signal.
5687 return 0x80 + WTERMSIG(status);
5688 } else {
5689 // Unknown exit code; pass it through
5690 return status;
5691 }
5692 }
5693 }
5694
5695 // Get the default path to the core file
5696 // Returns the length of the string
5697 int os::get_core_path(char* buffer, size_t bufferSize) {
5698 /*
5699 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5700 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5701 */
5702 const int core_pattern_len = 129;
5703 char core_pattern[core_pattern_len] = {0};
5704
5705 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5706 if (core_pattern_file == -1) {
5707 return -1;
5708 }
5709
5710 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5711 ::close(core_pattern_file);
5712 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5713 return -1;
5714 }
5715 if (core_pattern[ret-1] == '\n') {
5716 core_pattern[ret-1] = '\0';
5717 } else {
5718 core_pattern[ret] = '\0';
5719 }
5720
5721 char *pid_pos = strstr(core_pattern, "%p");
5722 int written;
5723
5724 if (core_pattern[0] == '/') {
5725 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5726 } else {
5727 char cwd[PATH_MAX];
5728
5729 const char* p = get_current_directory(cwd, PATH_MAX);
5730 if (p == NULL) {
5731 return -1;
5732 }
5733
5734 if (core_pattern[0] == '|') {
5735 written = jio_snprintf(buffer, bufferSize,
5736 "\"%s\" (or dumping to %s/core.%d)",
5737 &core_pattern[1], p, current_process_id());
5738 } else {
5739 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5740 }
5741 }
5742
5743 if (written < 0) {
5744 return -1;
5745 }
5746
5747 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5748 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5749
5750 if (core_uses_pid_file != -1) {
5751 char core_uses_pid = 0;
5752 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5753 ::close(core_uses_pid_file);
5754
5755 if (core_uses_pid == '1') {
5756 jio_snprintf(buffer + written, bufferSize - written,
5757 ".%d", current_process_id());
5758 }
5759 }
5760 }
5761
5762 return strlen(buffer);
5763 }
5764
5765 bool os::start_debugging(char *buf, int buflen) {
5766 int len = (int)strlen(buf);
5767 char *p = &buf[len];
5768
5769 jio_snprintf(p, buflen-len,
5770 "\n\n"
5771 "Do you want to debug the problem?\n\n"
5772 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5773 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5774 "Otherwise, press RETURN to abort...",
5775 os::current_process_id(), os::current_process_id(),
5776 os::current_thread_id(), os::current_thread_id());
5777
5778 bool yes = os::message_box("Unexpected Error", buf);
5779
5780 if (yes) {
5781 // yes, user asked VM to launch debugger
5782 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5783 os::current_process_id(), os::current_process_id());
5784
5785 os::fork_and_exec(buf);
5786 yes = false;
5787 }
5788 return yes;
5789 }
5790
5791
5792 // Java/Compiler thread:
5793 //
5794 // Low memory addresses
5795 // P0 +------------------------+
5796 // | |\ Java thread created by VM does not have glibc
5797 // | glibc guard page | - guard page, attached Java thread usually has
5798 // | |/ 1 glibc guard page.
5799 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5800 // | |\
5801 // | HotSpot Guard Pages | - red, yellow and reserved pages
5802 // | |/
5803 // +------------------------+ JavaThread::stack_reserved_zone_base()
5804 // | |\
5805 // | Normal Stack | -
5806 // | |/
5807 // P2 +------------------------+ Thread::stack_base()
5808 //
5809 // Non-Java thread:
5810 //
5811 // Low memory addresses
5812 // P0 +------------------------+
5813 // | |\
5814 // | glibc guard page | - usually 1 page
5815 // | |/
5816 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5817 // | |\
5818 // | Normal Stack | -
5819 // | |/
5820 // P2 +------------------------+ Thread::stack_base()
5821 //
5822 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5823 // returned from pthread_attr_getstack().
5824 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5825 // of the stack size given in pthread_attr. We work around this for
5826 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5827 //
5828 #ifndef ZERO
5829 static void current_stack_region(address * bottom, size_t * size) {
5830 if (os::is_primordial_thread()) {
5831 // primordial thread needs special handling because pthread_getattr_np()
5832 // may return bogus value.
5833 *bottom = os::Linux::initial_thread_stack_bottom();
5834 *size = os::Linux::initial_thread_stack_size();
5835 } else {
5836 pthread_attr_t attr;
5837
5838 int rslt = pthread_getattr_np(pthread_self(), &attr);
5839
5840 // JVM needs to know exact stack location, abort if it fails
5841 if (rslt != 0) {
5842 if (rslt == ENOMEM) {
5843 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5844 } else {
5845 fatal("pthread_getattr_np failed with error = %d", rslt);
5846 }
5847 }
5848
5849 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5850 fatal("Cannot locate current stack attributes!");
5851 }
5852
5853 // Work around NPTL stack guard error.
5854 size_t guard_size = 0;
5855 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5856 if (rslt != 0) {
5857 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5858 }
5859 *bottom += guard_size;
5860 *size -= guard_size;
5861
5862 pthread_attr_destroy(&attr);
5863
5864 }
5865 assert(os::current_stack_pointer() >= *bottom &&
5866 os::current_stack_pointer() < *bottom + *size, "just checking");
5867 }
5868
5869 address os::current_stack_base() {
5870 address bottom;
5871 size_t size;
5872 current_stack_region(&bottom, &size);
5873 return (bottom + size);
5874 }
5875
5876 size_t os::current_stack_size() {
5877 // This stack size includes the usable stack and HotSpot guard pages
5878 // (for the threads that have Hotspot guard pages).
5879 address bottom;
5880 size_t size;
5881 current_stack_region(&bottom, &size);
5882 return size;
5883 }
5884 #endif
5885
5886 static inline struct timespec get_mtime(const char* filename) {
5887 struct stat st;
5888 int ret = os::stat(filename, &st);
5889 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5890 return st.st_mtim;
5891 }
5892
5893 int os::compare_file_modified_times(const char* file1, const char* file2) {
5894 struct timespec filetime1 = get_mtime(file1);
5895 struct timespec filetime2 = get_mtime(file2);
5896 int diff = filetime1.tv_sec - filetime2.tv_sec;
5897 if (diff == 0) {
5898 return filetime1.tv_nsec - filetime2.tv_nsec;
5899 }
5900 return diff;
5901 }
5902
5903 /////////////// Unit tests ///////////////
5904
5905 #ifndef PRODUCT
5906
5907 #define test_log(...) \
5908 do { \
5909 if (VerboseInternalVMTests) { \
5910 tty->print_cr(__VA_ARGS__); \
5911 tty->flush(); \
5912 } \
5913 } while (false)
5914
5915 class TestReserveMemorySpecial : AllStatic {
5916 public:
5917 static void small_page_write(void* addr, size_t size) {
5918 size_t page_size = os::vm_page_size();
5919
5920 char* end = (char*)addr + size;
5921 for (char* p = (char*)addr; p < end; p += page_size) {
5922 *p = 1;
5923 }
5924 }
5925
5926 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5927 if (!UseHugeTLBFS) {
5928 return;
5929 }
5930
5931 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5932
5933 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5934
5935 if (addr != NULL) {
5936 small_page_write(addr, size);
5937
5938 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5939 }
5940 }
5941
5942 static void test_reserve_memory_special_huge_tlbfs_only() {
5943 if (!UseHugeTLBFS) {
5944 return;
5945 }
5946
5947 size_t lp = os::large_page_size();
5948
5949 for (size_t size = lp; size <= lp * 10; size += lp) {
5950 test_reserve_memory_special_huge_tlbfs_only(size);
5951 }
5952 }
5953
5954 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5955 size_t lp = os::large_page_size();
5956 size_t ag = os::vm_allocation_granularity();
5957
5958 // sizes to test
5959 const size_t sizes[] = {
5960 lp, lp + ag, lp + lp / 2, lp * 2,
5961 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5962 lp * 10, lp * 10 + lp / 2
5963 };
5964 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5965
5966 // For each size/alignment combination, we test three scenarios:
5967 // 1) with req_addr == NULL
5968 // 2) with a non-null req_addr at which we expect to successfully allocate
5969 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5970 // expect the allocation to either fail or to ignore req_addr
5971
5972 // Pre-allocate two areas; they shall be as large as the largest allocation
5973 // and aligned to the largest alignment we will be testing.
5974 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5975 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5976 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5977 -1, 0);
5978 assert(mapping1 != MAP_FAILED, "should work");
5979
5980 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5981 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5982 -1, 0);
5983 assert(mapping2 != MAP_FAILED, "should work");
5984
5985 // Unmap the first mapping, but leave the second mapping intact: the first
5986 // mapping will serve as a value for a "good" req_addr (case 2). The second
5987 // mapping, still intact, as "bad" req_addr (case 3).
5988 ::munmap(mapping1, mapping_size);
5989
5990 // Case 1
5991 test_log("%s, req_addr NULL:", __FUNCTION__);
5992 test_log("size align result");
5993
5994 for (int i = 0; i < num_sizes; i++) {
5995 const size_t size = sizes[i];
5996 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5997 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5998 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
5999 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6000 if (p != NULL) {
6001 assert(is_aligned(p, alignment), "must be");
6002 small_page_write(p, size);
6003 os::Linux::release_memory_special_huge_tlbfs(p, size);
6004 }
6005 }
6006 }
6007
6008 // Case 2
6009 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6010 test_log("size align req_addr result");
6011
6012 for (int i = 0; i < num_sizes; i++) {
6013 const size_t size = sizes[i];
6014 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6015 char* const req_addr = align_up(mapping1, alignment);
6016 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6017 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6018 size, alignment, p2i(req_addr), p2i(p),
6019 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6020 if (p != NULL) {
6021 assert(p == req_addr, "must be");
6022 small_page_write(p, size);
6023 os::Linux::release_memory_special_huge_tlbfs(p, size);
6024 }
6025 }
6026 }
6027
6028 // Case 3
6029 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6030 test_log("size align req_addr result");
6031
6032 for (int i = 0; i < num_sizes; i++) {
6033 const size_t size = sizes[i];
6034 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6035 char* const req_addr = align_up(mapping2, alignment);
6036 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6037 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6038 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6039 // as the area around req_addr contains already existing mappings, the API should always
6040 // return NULL (as per contract, it cannot return another address)
6041 assert(p == NULL, "must be");
6042 }
6043 }
6044
6045 ::munmap(mapping2, mapping_size);
6046
6047 }
6048
6049 static void test_reserve_memory_special_huge_tlbfs() {
6050 if (!UseHugeTLBFS) {
6051 return;
6052 }
6053
6054 test_reserve_memory_special_huge_tlbfs_only();
6055 test_reserve_memory_special_huge_tlbfs_mixed();
6056 }
6057
6058 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6059 if (!UseSHM) {
6060 return;
6061 }
6062
6063 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6064
6065 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6066
6067 if (addr != NULL) {
6068 assert(is_aligned(addr, alignment), "Check");
6069 assert(is_aligned(addr, os::large_page_size()), "Check");
6070
6071 small_page_write(addr, size);
6072
6073 os::Linux::release_memory_special_shm(addr, size);
6074 }
6075 }
6076
6077 static void test_reserve_memory_special_shm() {
6078 size_t lp = os::large_page_size();
6079 size_t ag = os::vm_allocation_granularity();
6080
6081 for (size_t size = ag; size < lp * 3; size += ag) {
6082 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6083 test_reserve_memory_special_shm(size, alignment);
6084 }
6085 }
6086 }
6087
6088 static void test() {
6089 test_reserve_memory_special_huge_tlbfs();
6090 test_reserve_memory_special_shm();
6091 }
6092 };
6093
6094 void TestReserveMemorySpecial_test() {
6095 TestReserveMemorySpecial::test();
6096 }
6097
6098 #endif