1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/threadSMR.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/align.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/vmError.hpp"
76
77 // put OS-includes here
78 # include <sys/types.h>
79 # include <sys/mman.h>
80 # include <sys/stat.h>
81 # include <sys/select.h>
82 # include <pthread.h>
83 # include <signal.h>
84 # include <errno.h>
85 # include <dlfcn.h>
86 # include <stdio.h>
87 # include <unistd.h>
88 # include <sys/resource.h>
89 # include <pthread.h>
90 # include <sys/stat.h>
91 # include <sys/time.h>
92 # include <sys/times.h>
93 # include <sys/utsname.h>
94 # include <sys/socket.h>
95 # include <sys/wait.h>
96 # include <pwd.h>
97 # include <poll.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 #define DAX_SHARED_BIT (1 << 8)
133 ////////////////////////////////////////////////////////////////////////////////
134 // global variables
135 julong os::Linux::_physical_memory = 0;
136
137 address os::Linux::_initial_thread_stack_bottom = NULL;
138 uintptr_t os::Linux::_initial_thread_stack_size = 0;
139
140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
143 Mutex* os::Linux::_createThread_lock = NULL;
144 pthread_t os::Linux::_main_thread;
145 int os::Linux::_page_size = -1;
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 uint32_t os::Linux::_os_version = 0;
148 const char * os::Linux::_glibc_version = NULL;
149 const char * os::Linux::_libpthread_version = NULL;
150
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 // Signal number used to suspend/resume a thread
160
161 // do not use any signal number less than SIGSEGV, see 4355769
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 // utility functions
166
167 static int SR_initialize();
168
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
172
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 julong avail_mem;
177
178 if (OSContainer::is_containerized()) {
179 jlong mem_limit, mem_usage;
180 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
181 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
182 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
183 }
184 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
185 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
186 }
187 if (mem_limit > 0 && mem_usage > 0 ) {
188 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
189 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
190 return avail_mem;
191 }
192 }
193
194 sysinfo(&si);
195 avail_mem = (julong)si.freeram * si.mem_unit;
196 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
197 return avail_mem;
198 }
199
200 julong os::physical_memory() {
201 jlong phys_mem = 0;
202 if (OSContainer::is_containerized()) {
203 jlong mem_limit;
204 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
205 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
206 return mem_limit;
207 }
208 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
209 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
210 }
211
212 phys_mem = Linux::physical_memory();
213 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
214 return phys_mem;
215 }
216
217 // Return true if user is running as root.
218
219 bool os::have_special_privileges() {
220 static bool init = false;
221 static bool privileges = false;
222 if (!init) {
223 privileges = (getuid() != geteuid()) || (getgid() != getegid());
224 init = true;
225 }
226 return privileges;
227 }
228
229
230 #ifndef SYS_gettid
231 // i386: 224, ia64: 1105, amd64: 186, sparc 143
232 #ifdef __ia64__
233 #define SYS_gettid 1105
234 #else
235 #ifdef __i386__
236 #define SYS_gettid 224
237 #else
238 #ifdef __amd64__
239 #define SYS_gettid 186
240 #else
241 #ifdef __sparc__
242 #define SYS_gettid 143
243 #else
244 #error define gettid for the arch
245 #endif
246 #endif
247 #endif
248 #endif
249 #endif
250
251
252 // pid_t gettid()
253 //
254 // Returns the kernel thread id of the currently running thread. Kernel
255 // thread id is used to access /proc.
256 pid_t os::Linux::gettid() {
257 int rslt = syscall(SYS_gettid);
258 assert(rslt != -1, "must be."); // old linuxthreads implementation?
259 return (pid_t)rslt;
260 }
261
262 // Most versions of linux have a bug where the number of processors are
263 // determined by looking at the /proc file system. In a chroot environment,
264 // the system call returns 1. This causes the VM to act as if it is
265 // a single processor and elide locking (see is_MP() call).
266 static bool unsafe_chroot_detected = false;
267 static const char *unstable_chroot_error = "/proc file system not found.\n"
268 "Java may be unstable running multithreaded in a chroot "
269 "environment on Linux when /proc filesystem is not mounted.";
270
271 void os::Linux::initialize_system_info() {
272 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
273 if (processor_count() == 1) {
274 pid_t pid = os::Linux::gettid();
275 char fname[32];
276 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
277 FILE *fp = fopen(fname, "r");
278 if (fp == NULL) {
279 unsafe_chroot_detected = true;
280 } else {
281 fclose(fp);
282 }
283 }
284 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
285 assert(processor_count() > 0, "linux error");
286 }
287
288 void os::init_system_properties_values() {
289 // The next steps are taken in the product version:
290 //
291 // Obtain the JAVA_HOME value from the location of libjvm.so.
292 // This library should be located at:
293 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
294 //
295 // If "/jre/lib/" appears at the right place in the path, then we
296 // assume libjvm.so is installed in a JDK and we use this path.
297 //
298 // Otherwise exit with message: "Could not create the Java virtual machine."
299 //
300 // The following extra steps are taken in the debugging version:
301 //
302 // If "/jre/lib/" does NOT appear at the right place in the path
303 // instead of exit check for $JAVA_HOME environment variable.
304 //
305 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
306 // then we append a fake suffix "hotspot/libjvm.so" to this path so
307 // it looks like libjvm.so is installed there
308 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
309 //
310 // Otherwise exit.
311 //
312 // Important note: if the location of libjvm.so changes this
313 // code needs to be changed accordingly.
314
315 // See ld(1):
316 // The linker uses the following search paths to locate required
317 // shared libraries:
318 // 1: ...
319 // ...
320 // 7: The default directories, normally /lib and /usr/lib.
321 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
322 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
323 #else
324 #define DEFAULT_LIBPATH "/lib:/usr/lib"
325 #endif
326
327 // Base path of extensions installed on the system.
328 #define SYS_EXT_DIR "/usr/java/packages"
329 #define EXTENSIONS_DIR "/lib/ext"
330
331 // Buffer that fits several sprintfs.
332 // Note that the space for the colon and the trailing null are provided
333 // by the nulls included by the sizeof operator.
334 const size_t bufsize =
335 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
336 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
337 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
338
339 // sysclasspath, java_home, dll_dir
340 {
341 char *pslash;
342 os::jvm_path(buf, bufsize);
343
344 // Found the full path to libjvm.so.
345 // Now cut the path to <java_home>/jre if we can.
346 pslash = strrchr(buf, '/');
347 if (pslash != NULL) {
348 *pslash = '\0'; // Get rid of /libjvm.so.
349 }
350 pslash = strrchr(buf, '/');
351 if (pslash != NULL) {
352 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
353 }
354 Arguments::set_dll_dir(buf);
355
356 if (pslash != NULL) {
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /lib.
360 }
361 }
362 Arguments::set_java_home(buf);
363 set_boot_path('/', ':');
364 }
365
366 // Where to look for native libraries.
367 //
368 // Note: Due to a legacy implementation, most of the library path
369 // is set in the launcher. This was to accomodate linking restrictions
370 // on legacy Linux implementations (which are no longer supported).
371 // Eventually, all the library path setting will be done here.
372 //
373 // However, to prevent the proliferation of improperly built native
374 // libraries, the new path component /usr/java/packages is added here.
375 // Eventually, all the library path setting will be done here.
376 {
377 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
378 // should always exist (until the legacy problem cited above is
379 // addressed).
380 const char *v = ::getenv("LD_LIBRARY_PATH");
381 const char *v_colon = ":";
382 if (v == NULL) { v = ""; v_colon = ""; }
383 // That's +1 for the colon and +1 for the trailing '\0'.
384 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
385 strlen(v) + 1 +
386 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
387 mtInternal);
388 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
389 Arguments::set_library_path(ld_library_path);
390 FREE_C_HEAP_ARRAY(char, ld_library_path);
391 }
392
393 // Extensions directories.
394 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
395 Arguments::set_ext_dirs(buf);
396
397 FREE_C_HEAP_ARRAY(char, buf);
398
399 #undef DEFAULT_LIBPATH
400 #undef SYS_EXT_DIR
401 #undef EXTENSIONS_DIR
402 }
403
404 ////////////////////////////////////////////////////////////////////////////////
405 // breakpoint support
406
407 void os::breakpoint() {
408 BREAKPOINT;
409 }
410
411 extern "C" void breakpoint() {
412 // use debugger to set breakpoint here
413 }
414
415 ////////////////////////////////////////////////////////////////////////////////
416 // signal support
417
418 debug_only(static bool signal_sets_initialized = false);
419 static sigset_t unblocked_sigs, vm_sigs;
420
421 bool os::Linux::is_sig_ignored(int sig) {
422 struct sigaction oact;
423 sigaction(sig, (struct sigaction*)NULL, &oact);
424 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
425 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
426 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
427 return true;
428 } else {
429 return false;
430 }
431 }
432
433 void os::Linux::signal_sets_init() {
434 // Should also have an assertion stating we are still single-threaded.
435 assert(!signal_sets_initialized, "Already initialized");
436 // Fill in signals that are necessarily unblocked for all threads in
437 // the VM. Currently, we unblock the following signals:
438 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
439 // by -Xrs (=ReduceSignalUsage));
440 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
441 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
442 // the dispositions or masks wrt these signals.
443 // Programs embedding the VM that want to use the above signals for their
444 // own purposes must, at this time, use the "-Xrs" option to prevent
445 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
446 // (See bug 4345157, and other related bugs).
447 // In reality, though, unblocking these signals is really a nop, since
448 // these signals are not blocked by default.
449 sigemptyset(&unblocked_sigs);
450 sigaddset(&unblocked_sigs, SIGILL);
451 sigaddset(&unblocked_sigs, SIGSEGV);
452 sigaddset(&unblocked_sigs, SIGBUS);
453 sigaddset(&unblocked_sigs, SIGFPE);
454 #if defined(PPC64)
455 sigaddset(&unblocked_sigs, SIGTRAP);
456 #endif
457 sigaddset(&unblocked_sigs, SR_signum);
458
459 if (!ReduceSignalUsage) {
460 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
461 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
462 }
463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
468 }
469 }
470 // Fill in signals that are blocked by all but the VM thread.
471 sigemptyset(&vm_sigs);
472 if (!ReduceSignalUsage) {
473 sigaddset(&vm_sigs, BREAK_SIGNAL);
474 }
475 debug_only(signal_sets_initialized = true);
476
477 }
478
479 // These are signals that are unblocked while a thread is running Java.
480 // (For some reason, they get blocked by default.)
481 sigset_t* os::Linux::unblocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &unblocked_sigs;
484 }
485
486 // These are the signals that are blocked while a (non-VM) thread is
487 // running Java. Only the VM thread handles these signals.
488 sigset_t* os::Linux::vm_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &vm_sigs;
491 }
492
493 void os::Linux::hotspot_sigmask(Thread* thread) {
494
495 //Save caller's signal mask before setting VM signal mask
496 sigset_t caller_sigmask;
497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
498
499 OSThread* osthread = thread->osthread();
500 osthread->set_caller_sigmask(caller_sigmask);
501
502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
503
504 if (!ReduceSignalUsage) {
505 if (thread->is_VM_thread()) {
506 // Only the VM thread handles BREAK_SIGNAL ...
507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
508 } else {
509 // ... all other threads block BREAK_SIGNAL
510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
511 }
512 }
513 }
514
515 //////////////////////////////////////////////////////////////////////////////
516 // detecting pthread library
517
518 void os::Linux::libpthread_init() {
519 // Save glibc and pthread version strings.
520 #if !defined(_CS_GNU_LIBC_VERSION) || \
521 !defined(_CS_GNU_LIBPTHREAD_VERSION)
522 #error "glibc too old (< 2.3.2)"
523 #endif
524
525 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
526 assert(n > 0, "cannot retrieve glibc version");
527 char *str = (char *)malloc(n, mtInternal);
528 confstr(_CS_GNU_LIBC_VERSION, str, n);
529 os::Linux::set_glibc_version(str);
530
531 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
532 assert(n > 0, "cannot retrieve pthread version");
533 str = (char *)malloc(n, mtInternal);
534 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
535 os::Linux::set_libpthread_version(str);
536 }
537
538 /////////////////////////////////////////////////////////////////////////////
539 // thread stack expansion
540
541 // os::Linux::manually_expand_stack() takes care of expanding the thread
542 // stack. Note that this is normally not needed: pthread stacks allocate
543 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
544 // committed. Therefore it is not necessary to expand the stack manually.
545 //
546 // Manually expanding the stack was historically needed on LinuxThreads
547 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
548 // it is kept to deal with very rare corner cases:
549 //
550 // For one, user may run the VM on an own implementation of threads
551 // whose stacks are - like the old LinuxThreads - implemented using
552 // mmap(MAP_GROWSDOWN).
553 //
554 // Also, this coding may be needed if the VM is running on the primordial
555 // thread. Normally we avoid running on the primordial thread; however,
556 // user may still invoke the VM on the primordial thread.
557 //
558 // The following historical comment describes the details about running
559 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
560
561
562 // Force Linux kernel to expand current thread stack. If "bottom" is close
563 // to the stack guard, caller should block all signals.
564 //
565 // MAP_GROWSDOWN:
566 // A special mmap() flag that is used to implement thread stacks. It tells
567 // kernel that the memory region should extend downwards when needed. This
568 // allows early versions of LinuxThreads to only mmap the first few pages
569 // when creating a new thread. Linux kernel will automatically expand thread
570 // stack as needed (on page faults).
571 //
572 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
573 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
574 // region, it's hard to tell if the fault is due to a legitimate stack
575 // access or because of reading/writing non-exist memory (e.g. buffer
576 // overrun). As a rule, if the fault happens below current stack pointer,
577 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
578 // application (see Linux kernel fault.c).
579 //
580 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
581 // stack overflow detection.
582 //
583 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
584 // not use MAP_GROWSDOWN.
585 //
586 // To get around the problem and allow stack banging on Linux, we need to
587 // manually expand thread stack after receiving the SIGSEGV.
588 //
589 // There are two ways to expand thread stack to address "bottom", we used
590 // both of them in JVM before 1.5:
591 // 1. adjust stack pointer first so that it is below "bottom", and then
592 // touch "bottom"
593 // 2. mmap() the page in question
594 //
595 // Now alternate signal stack is gone, it's harder to use 2. For instance,
596 // if current sp is already near the lower end of page 101, and we need to
597 // call mmap() to map page 100, it is possible that part of the mmap() frame
598 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
599 // That will destroy the mmap() frame and cause VM to crash.
600 //
601 // The following code works by adjusting sp first, then accessing the "bottom"
602 // page to force a page fault. Linux kernel will then automatically expand the
603 // stack mapping.
604 //
605 // _expand_stack_to() assumes its frame size is less than page size, which
606 // should always be true if the function is not inlined.
607
608 static void NOINLINE _expand_stack_to(address bottom) {
609 address sp;
610 size_t size;
611 volatile char *p;
612
613 // Adjust bottom to point to the largest address within the same page, it
614 // gives us a one-page buffer if alloca() allocates slightly more memory.
615 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
616 bottom += os::Linux::page_size() - 1;
617
618 // sp might be slightly above current stack pointer; if that's the case, we
619 // will alloca() a little more space than necessary, which is OK. Don't use
620 // os::current_stack_pointer(), as its result can be slightly below current
621 // stack pointer, causing us to not alloca enough to reach "bottom".
622 sp = (address)&sp;
623
624 if (sp > bottom) {
625 size = sp - bottom;
626 p = (volatile char *)alloca(size);
627 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
628 p[0] = '\0';
629 }
630 }
631
632 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
633 assert(t!=NULL, "just checking");
634 assert(t->osthread()->expanding_stack(), "expand should be set");
635 assert(t->stack_base() != NULL, "stack_base was not initialized");
636
637 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
638 sigset_t mask_all, old_sigset;
639 sigfillset(&mask_all);
640 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
641 _expand_stack_to(addr);
642 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
643 return true;
644 }
645 return false;
646 }
647
648 //////////////////////////////////////////////////////////////////////////////
649 // create new thread
650
651 // Thread start routine for all newly created threads
652 static void *thread_native_entry(Thread *thread) {
653 // Try to randomize the cache line index of hot stack frames.
654 // This helps when threads of the same stack traces evict each other's
655 // cache lines. The threads can be either from the same JVM instance, or
656 // from different JVM instances. The benefit is especially true for
657 // processors with hyperthreading technology.
658 static int counter = 0;
659 int pid = os::current_process_id();
660 alloca(((pid ^ counter++) & 7) * 128);
661
662 thread->initialize_thread_current();
663
664 OSThread* osthread = thread->osthread();
665 Monitor* sync = osthread->startThread_lock();
666
667 osthread->set_thread_id(os::current_thread_id());
668
669 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
670 os::current_thread_id(), (uintx) pthread_self());
671
672 if (UseNUMA) {
673 int lgrp_id = os::numa_get_group_id();
674 if (lgrp_id != -1) {
675 thread->set_lgrp_id(lgrp_id);
676 }
677 }
678 // initialize signal mask for this thread
679 os::Linux::hotspot_sigmask(thread);
680
681 // initialize floating point control register
682 os::Linux::init_thread_fpu_state();
683
684 // handshaking with parent thread
685 {
686 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
687
688 // notify parent thread
689 osthread->set_state(INITIALIZED);
690 sync->notify_all();
691
692 // wait until os::start_thread()
693 while (osthread->get_state() == INITIALIZED) {
694 sync->wait(Mutex::_no_safepoint_check_flag);
695 }
696 }
697
698 // call one more level start routine
699 thread->run();
700
701 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
702 os::current_thread_id(), (uintx) pthread_self());
703
704 // If a thread has not deleted itself ("delete this") as part of its
705 // termination sequence, we have to ensure thread-local-storage is
706 // cleared before we actually terminate. No threads should ever be
707 // deleted asynchronously with respect to their termination.
708 if (Thread::current_or_null_safe() != NULL) {
709 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
710 thread->clear_thread_current();
711 }
712
713 return 0;
714 }
715
716 bool os::create_thread(Thread* thread, ThreadType thr_type,
717 size_t req_stack_size) {
718 assert(thread->osthread() == NULL, "caller responsible");
719
720 // Allocate the OSThread object
721 OSThread* osthread = new OSThread(NULL, NULL);
722 if (osthread == NULL) {
723 return false;
724 }
725
726 // set the correct thread state
727 osthread->set_thread_type(thr_type);
728
729 // Initial state is ALLOCATED but not INITIALIZED
730 osthread->set_state(ALLOCATED);
731
732 thread->set_osthread(osthread);
733
734 // init thread attributes
735 pthread_attr_t attr;
736 pthread_attr_init(&attr);
737 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
738
739 // Calculate stack size if it's not specified by caller.
740 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
741 // In the Linux NPTL pthread implementation the guard size mechanism
742 // is not implemented properly. The posix standard requires adding
743 // the size of the guard pages to the stack size, instead Linux
744 // takes the space out of 'stacksize'. Thus we adapt the requested
745 // stack_size by the size of the guard pages to mimick proper
746 // behaviour. However, be careful not to end up with a size
747 // of zero due to overflow. Don't add the guard page in that case.
748 size_t guard_size = os::Linux::default_guard_size(thr_type);
749 if (stack_size <= SIZE_MAX - guard_size) {
750 stack_size += guard_size;
751 }
752 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
753
754 int status = pthread_attr_setstacksize(&attr, stack_size);
755 assert_status(status == 0, status, "pthread_attr_setstacksize");
756
757 // Configure glibc guard page.
758 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
759
760 ThreadState state;
761
762 {
763 pthread_t tid;
764 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
765
766 char buf[64];
767 if (ret == 0) {
768 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
769 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
770 } else {
771 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
772 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
773 }
774
775 pthread_attr_destroy(&attr);
776
777 if (ret != 0) {
778 // Need to clean up stuff we've allocated so far
779 thread->set_osthread(NULL);
780 delete osthread;
781 return false;
782 }
783
784 // Store pthread info into the OSThread
785 osthread->set_pthread_id(tid);
786
787 // Wait until child thread is either initialized or aborted
788 {
789 Monitor* sync_with_child = osthread->startThread_lock();
790 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
791 while ((state = osthread->get_state()) == ALLOCATED) {
792 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
793 }
794 }
795 }
796
797 // Aborted due to thread limit being reached
798 if (state == ZOMBIE) {
799 thread->set_osthread(NULL);
800 delete osthread;
801 return false;
802 }
803
804 // The thread is returned suspended (in state INITIALIZED),
805 // and is started higher up in the call chain
806 assert(state == INITIALIZED, "race condition");
807 return true;
808 }
809
810 /////////////////////////////////////////////////////////////////////////////
811 // attach existing thread
812
813 // bootstrap the main thread
814 bool os::create_main_thread(JavaThread* thread) {
815 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
816 return create_attached_thread(thread);
817 }
818
819 bool os::create_attached_thread(JavaThread* thread) {
820 #ifdef ASSERT
821 thread->verify_not_published();
822 #endif
823
824 // Allocate the OSThread object
825 OSThread* osthread = new OSThread(NULL, NULL);
826
827 if (osthread == NULL) {
828 return false;
829 }
830
831 // Store pthread info into the OSThread
832 osthread->set_thread_id(os::Linux::gettid());
833 osthread->set_pthread_id(::pthread_self());
834
835 // initialize floating point control register
836 os::Linux::init_thread_fpu_state();
837
838 // Initial thread state is RUNNABLE
839 osthread->set_state(RUNNABLE);
840
841 thread->set_osthread(osthread);
842
843 if (UseNUMA) {
844 int lgrp_id = os::numa_get_group_id();
845 if (lgrp_id != -1) {
846 thread->set_lgrp_id(lgrp_id);
847 }
848 }
849
850 if (os::is_primordial_thread()) {
851 // If current thread is primordial thread, its stack is mapped on demand,
852 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
853 // the entire stack region to avoid SEGV in stack banging.
854 // It is also useful to get around the heap-stack-gap problem on SuSE
855 // kernel (see 4821821 for details). We first expand stack to the top
856 // of yellow zone, then enable stack yellow zone (order is significant,
857 // enabling yellow zone first will crash JVM on SuSE Linux), so there
858 // is no gap between the last two virtual memory regions.
859
860 JavaThread *jt = (JavaThread *)thread;
861 address addr = jt->stack_reserved_zone_base();
862 assert(addr != NULL, "initialization problem?");
863 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
864
865 osthread->set_expanding_stack();
866 os::Linux::manually_expand_stack(jt, addr);
867 osthread->clear_expanding_stack();
868 }
869
870 // initialize signal mask for this thread
871 // and save the caller's signal mask
872 os::Linux::hotspot_sigmask(thread);
873
874 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
875 os::current_thread_id(), (uintx) pthread_self());
876
877 return true;
878 }
879
880 void os::pd_start_thread(Thread* thread) {
881 OSThread * osthread = thread->osthread();
882 assert(osthread->get_state() != INITIALIZED, "just checking");
883 Monitor* sync_with_child = osthread->startThread_lock();
884 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
885 sync_with_child->notify();
886 }
887
888 // Free Linux resources related to the OSThread
889 void os::free_thread(OSThread* osthread) {
890 assert(osthread != NULL, "osthread not set");
891
892 // We are told to free resources of the argument thread,
893 // but we can only really operate on the current thread.
894 assert(Thread::current()->osthread() == osthread,
895 "os::free_thread but not current thread");
896
897 #ifdef ASSERT
898 sigset_t current;
899 sigemptyset(¤t);
900 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
901 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
902 #endif
903
904 // Restore caller's signal mask
905 sigset_t sigmask = osthread->caller_sigmask();
906 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
907
908 delete osthread;
909 }
910
911 //////////////////////////////////////////////////////////////////////////////
912 // primordial thread
913
914 // Check if current thread is the primordial thread, similar to Solaris thr_main.
915 bool os::is_primordial_thread(void) {
916 char dummy;
917 // If called before init complete, thread stack bottom will be null.
918 // Can be called if fatal error occurs before initialization.
919 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
920 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
921 os::Linux::initial_thread_stack_size() != 0,
922 "os::init did not locate primordial thread's stack region");
923 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
924 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
925 os::Linux::initial_thread_stack_size()) {
926 return true;
927 } else {
928 return false;
929 }
930 }
931
932 // Find the virtual memory area that contains addr
933 static bool find_vma(address addr, address* vma_low, address* vma_high) {
934 FILE *fp = fopen("/proc/self/maps", "r");
935 if (fp) {
936 address low, high;
937 while (!feof(fp)) {
938 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
939 if (low <= addr && addr < high) {
940 if (vma_low) *vma_low = low;
941 if (vma_high) *vma_high = high;
942 fclose(fp);
943 return true;
944 }
945 }
946 for (;;) {
947 int ch = fgetc(fp);
948 if (ch == EOF || ch == (int)'\n') break;
949 }
950 }
951 fclose(fp);
952 }
953 return false;
954 }
955
956 // Locate primordial thread stack. This special handling of primordial thread stack
957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
958 // bogus value for the primordial process thread. While the launcher has created
959 // the VM in a new thread since JDK 6, we still have to allow for the use of the
960 // JNI invocation API from a primordial thread.
961 void os::Linux::capture_initial_stack(size_t max_size) {
962
963 // max_size is either 0 (which means accept OS default for thread stacks) or
964 // a user-specified value known to be at least the minimum needed. If we
965 // are actually on the primordial thread we can make it appear that we have a
966 // smaller max_size stack by inserting the guard pages at that location. But we
967 // cannot do anything to emulate a larger stack than what has been provided by
968 // the OS or threading library. In fact if we try to use a stack greater than
969 // what is set by rlimit then we will crash the hosting process.
970
971 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
972 // If this is "unlimited" then it will be a huge value.
973 struct rlimit rlim;
974 getrlimit(RLIMIT_STACK, &rlim);
975 size_t stack_size = rlim.rlim_cur;
976
977 // 6308388: a bug in ld.so will relocate its own .data section to the
978 // lower end of primordial stack; reduce ulimit -s value a little bit
979 // so we won't install guard page on ld.so's data section.
980 // But ensure we don't underflow the stack size - allow 1 page spare
981 if (stack_size >= (size_t)(3 * page_size())) {
982 stack_size -= 2 * page_size();
983 }
984
985 // Try to figure out where the stack base (top) is. This is harder.
986 //
987 // When an application is started, glibc saves the initial stack pointer in
988 // a global variable "__libc_stack_end", which is then used by system
989 // libraries. __libc_stack_end should be pretty close to stack top. The
990 // variable is available since the very early days. However, because it is
991 // a private interface, it could disappear in the future.
992 //
993 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
994 // to __libc_stack_end, it is very close to stack top, but isn't the real
995 // stack top. Note that /proc may not exist if VM is running as a chroot
996 // program, so reading /proc/<pid>/stat could fail. Also the contents of
997 // /proc/<pid>/stat could change in the future (though unlikely).
998 //
999 // We try __libc_stack_end first. If that doesn't work, look for
1000 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1001 // as a hint, which should work well in most cases.
1002
1003 uintptr_t stack_start;
1004
1005 // try __libc_stack_end first
1006 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1007 if (p && *p) {
1008 stack_start = *p;
1009 } else {
1010 // see if we can get the start_stack field from /proc/self/stat
1011 FILE *fp;
1012 int pid;
1013 char state;
1014 int ppid;
1015 int pgrp;
1016 int session;
1017 int nr;
1018 int tpgrp;
1019 unsigned long flags;
1020 unsigned long minflt;
1021 unsigned long cminflt;
1022 unsigned long majflt;
1023 unsigned long cmajflt;
1024 unsigned long utime;
1025 unsigned long stime;
1026 long cutime;
1027 long cstime;
1028 long prio;
1029 long nice;
1030 long junk;
1031 long it_real;
1032 uintptr_t start;
1033 uintptr_t vsize;
1034 intptr_t rss;
1035 uintptr_t rsslim;
1036 uintptr_t scodes;
1037 uintptr_t ecode;
1038 int i;
1039
1040 // Figure what the primordial thread stack base is. Code is inspired
1041 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1042 // followed by command name surrounded by parentheses, state, etc.
1043 char stat[2048];
1044 int statlen;
1045
1046 fp = fopen("/proc/self/stat", "r");
1047 if (fp) {
1048 statlen = fread(stat, 1, 2047, fp);
1049 stat[statlen] = '\0';
1050 fclose(fp);
1051
1052 // Skip pid and the command string. Note that we could be dealing with
1053 // weird command names, e.g. user could decide to rename java launcher
1054 // to "java 1.4.2 :)", then the stat file would look like
1055 // 1234 (java 1.4.2 :)) R ... ...
1056 // We don't really need to know the command string, just find the last
1057 // occurrence of ")" and then start parsing from there. See bug 4726580.
1058 char * s = strrchr(stat, ')');
1059
1060 i = 0;
1061 if (s) {
1062 // Skip blank chars
1063 do { s++; } while (s && isspace(*s));
1064
1065 #define _UFM UINTX_FORMAT
1066 #define _DFM INTX_FORMAT
1067
1068 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1069 // 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
1070 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,
1071 &state, // 3 %c
1072 &ppid, // 4 %d
1073 &pgrp, // 5 %d
1074 &session, // 6 %d
1075 &nr, // 7 %d
1076 &tpgrp, // 8 %d
1077 &flags, // 9 %lu
1078 &minflt, // 10 %lu
1079 &cminflt, // 11 %lu
1080 &majflt, // 12 %lu
1081 &cmajflt, // 13 %lu
1082 &utime, // 14 %lu
1083 &stime, // 15 %lu
1084 &cutime, // 16 %ld
1085 &cstime, // 17 %ld
1086 &prio, // 18 %ld
1087 &nice, // 19 %ld
1088 &junk, // 20 %ld
1089 &it_real, // 21 %ld
1090 &start, // 22 UINTX_FORMAT
1091 &vsize, // 23 UINTX_FORMAT
1092 &rss, // 24 INTX_FORMAT
1093 &rsslim, // 25 UINTX_FORMAT
1094 &scodes, // 26 UINTX_FORMAT
1095 &ecode, // 27 UINTX_FORMAT
1096 &stack_start); // 28 UINTX_FORMAT
1097 }
1098
1099 #undef _UFM
1100 #undef _DFM
1101
1102 if (i != 28 - 2) {
1103 assert(false, "Bad conversion from /proc/self/stat");
1104 // product mode - assume we are the primordial thread, good luck in the
1105 // embedded case.
1106 warning("Can't detect primordial thread stack location - bad conversion");
1107 stack_start = (uintptr_t) &rlim;
1108 }
1109 } else {
1110 // For some reason we can't open /proc/self/stat (for example, running on
1111 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1112 // most cases, so don't abort:
1113 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1114 stack_start = (uintptr_t) &rlim;
1115 }
1116 }
1117
1118 // Now we have a pointer (stack_start) very close to the stack top, the
1119 // next thing to do is to figure out the exact location of stack top. We
1120 // can find out the virtual memory area that contains stack_start by
1121 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1122 // and its upper limit is the real stack top. (again, this would fail if
1123 // running inside chroot, because /proc may not exist.)
1124
1125 uintptr_t stack_top;
1126 address low, high;
1127 if (find_vma((address)stack_start, &low, &high)) {
1128 // success, "high" is the true stack top. (ignore "low", because initial
1129 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1130 stack_top = (uintptr_t)high;
1131 } else {
1132 // failed, likely because /proc/self/maps does not exist
1133 warning("Can't detect primordial thread stack location - find_vma failed");
1134 // best effort: stack_start is normally within a few pages below the real
1135 // stack top, use it as stack top, and reduce stack size so we won't put
1136 // guard page outside stack.
1137 stack_top = stack_start;
1138 stack_size -= 16 * page_size();
1139 }
1140
1141 // stack_top could be partially down the page so align it
1142 stack_top = align_up(stack_top, page_size());
1143
1144 // Allowed stack value is minimum of max_size and what we derived from rlimit
1145 if (max_size > 0) {
1146 _initial_thread_stack_size = MIN2(max_size, stack_size);
1147 } else {
1148 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1149 // clamp it at 8MB as we do on Solaris
1150 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1151 }
1152 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1153 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1154
1155 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1156
1157 if (log_is_enabled(Info, os, thread)) {
1158 // See if we seem to be on primordial process thread
1159 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1160 uintptr_t(&rlim) < stack_top;
1161
1162 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1163 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1164 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1165 stack_top, intptr_t(_initial_thread_stack_bottom));
1166 }
1167 }
1168
1169 ////////////////////////////////////////////////////////////////////////////////
1170 // time support
1171
1172 // Time since start-up in seconds to a fine granularity.
1173 // Used by VMSelfDestructTimer and the MemProfiler.
1174 double os::elapsedTime() {
1175
1176 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1177 }
1178
1179 jlong os::elapsed_counter() {
1180 return javaTimeNanos() - initial_time_count;
1181 }
1182
1183 jlong os::elapsed_frequency() {
1184 return NANOSECS_PER_SEC; // nanosecond resolution
1185 }
1186
1187 bool os::supports_vtime() { return true; }
1188 bool os::enable_vtime() { return false; }
1189 bool os::vtime_enabled() { return false; }
1190
1191 double os::elapsedVTime() {
1192 struct rusage usage;
1193 int retval = getrusage(RUSAGE_THREAD, &usage);
1194 if (retval == 0) {
1195 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);
1196 } else {
1197 // better than nothing, but not much
1198 return elapsedTime();
1199 }
1200 }
1201
1202 jlong os::javaTimeMillis() {
1203 timeval time;
1204 int status = gettimeofday(&time, NULL);
1205 assert(status != -1, "linux error");
1206 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1207 }
1208
1209 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1210 timeval time;
1211 int status = gettimeofday(&time, NULL);
1212 assert(status != -1, "linux error");
1213 seconds = jlong(time.tv_sec);
1214 nanos = jlong(time.tv_usec) * 1000;
1215 }
1216
1217
1218 #ifndef CLOCK_MONOTONIC
1219 #define CLOCK_MONOTONIC (1)
1220 #endif
1221
1222 void os::Linux::clock_init() {
1223 // we do dlopen's in this particular order due to bug in linux
1224 // dynamical loader (see 6348968) leading to crash on exit
1225 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1226 if (handle == NULL) {
1227 handle = dlopen("librt.so", RTLD_LAZY);
1228 }
1229
1230 if (handle) {
1231 int (*clock_getres_func)(clockid_t, struct timespec*) =
1232 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1233 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1234 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1235 if (clock_getres_func && clock_gettime_func) {
1236 // See if monotonic clock is supported by the kernel. Note that some
1237 // early implementations simply return kernel jiffies (updated every
1238 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1239 // for nano time (though the monotonic property is still nice to have).
1240 // It's fixed in newer kernels, however clock_getres() still returns
1241 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1242 // resolution for now. Hopefully as people move to new kernels, this
1243 // won't be a problem.
1244 struct timespec res;
1245 struct timespec tp;
1246 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1247 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1248 // yes, monotonic clock is supported
1249 _clock_gettime = clock_gettime_func;
1250 return;
1251 } else {
1252 // close librt if there is no monotonic clock
1253 dlclose(handle);
1254 }
1255 }
1256 }
1257 warning("No monotonic clock was available - timed services may " \
1258 "be adversely affected if the time-of-day clock changes");
1259 }
1260
1261 #ifndef SYS_clock_getres
1262 #if defined(X86) || defined(PPC64) || defined(S390)
1263 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1264 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1265 #else
1266 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1267 #define sys_clock_getres(x,y) -1
1268 #endif
1269 #else
1270 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1271 #endif
1272
1273 void os::Linux::fast_thread_clock_init() {
1274 if (!UseLinuxPosixThreadCPUClocks) {
1275 return;
1276 }
1277 clockid_t clockid;
1278 struct timespec tp;
1279 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1280 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1281
1282 // Switch to using fast clocks for thread cpu time if
1283 // the sys_clock_getres() returns 0 error code.
1284 // Note, that some kernels may support the current thread
1285 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1286 // returned by the pthread_getcpuclockid().
1287 // If the fast Posix clocks are supported then the sys_clock_getres()
1288 // must return at least tp.tv_sec == 0 which means a resolution
1289 // better than 1 sec. This is extra check for reliability.
1290
1291 if (pthread_getcpuclockid_func &&
1292 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1293 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1294 _supports_fast_thread_cpu_time = true;
1295 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1296 }
1297 }
1298
1299 jlong os::javaTimeNanos() {
1300 if (os::supports_monotonic_clock()) {
1301 struct timespec tp;
1302 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1303 assert(status == 0, "gettime error");
1304 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1305 return result;
1306 } else {
1307 timeval time;
1308 int status = gettimeofday(&time, NULL);
1309 assert(status != -1, "linux error");
1310 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1311 return 1000 * usecs;
1312 }
1313 }
1314
1315 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1316 if (os::supports_monotonic_clock()) {
1317 info_ptr->max_value = ALL_64_BITS;
1318
1319 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1320 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1321 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1322 } else {
1323 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1324 info_ptr->max_value = ALL_64_BITS;
1325
1326 // gettimeofday is a real time clock so it skips
1327 info_ptr->may_skip_backward = true;
1328 info_ptr->may_skip_forward = true;
1329 }
1330
1331 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1332 }
1333
1334 // Return the real, user, and system times in seconds from an
1335 // arbitrary fixed point in the past.
1336 bool os::getTimesSecs(double* process_real_time,
1337 double* process_user_time,
1338 double* process_system_time) {
1339 struct tms ticks;
1340 clock_t real_ticks = times(&ticks);
1341
1342 if (real_ticks == (clock_t) (-1)) {
1343 return false;
1344 } else {
1345 double ticks_per_second = (double) clock_tics_per_sec;
1346 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1347 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1348 *process_real_time = ((double) real_ticks) / ticks_per_second;
1349
1350 return true;
1351 }
1352 }
1353
1354
1355 char * os::local_time_string(char *buf, size_t buflen) {
1356 struct tm t;
1357 time_t long_time;
1358 time(&long_time);
1359 localtime_r(&long_time, &t);
1360 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1361 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1362 t.tm_hour, t.tm_min, t.tm_sec);
1363 return buf;
1364 }
1365
1366 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1367 return localtime_r(clock, res);
1368 }
1369
1370 ////////////////////////////////////////////////////////////////////////////////
1371 // runtime exit support
1372
1373 // Note: os::shutdown() might be called very early during initialization, or
1374 // called from signal handler. Before adding something to os::shutdown(), make
1375 // sure it is async-safe and can handle partially initialized VM.
1376 void os::shutdown() {
1377
1378 // allow PerfMemory to attempt cleanup of any persistent resources
1379 perfMemory_exit();
1380
1381 // needs to remove object in file system
1382 AttachListener::abort();
1383
1384 // flush buffered output, finish log files
1385 ostream_abort();
1386
1387 // Check for abort hook
1388 abort_hook_t abort_hook = Arguments::abort_hook();
1389 if (abort_hook != NULL) {
1390 abort_hook();
1391 }
1392
1393 }
1394
1395 // Note: os::abort() might be called very early during initialization, or
1396 // called from signal handler. Before adding something to os::abort(), make
1397 // sure it is async-safe and can handle partially initialized VM.
1398 void os::abort(bool dump_core, void* siginfo, const void* context) {
1399 os::shutdown();
1400 if (dump_core) {
1401 #ifndef PRODUCT
1402 fdStream out(defaultStream::output_fd());
1403 out.print_raw("Current thread is ");
1404 char buf[16];
1405 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1406 out.print_raw_cr(buf);
1407 out.print_raw_cr("Dumping core ...");
1408 #endif
1409 ::abort(); // dump core
1410 }
1411
1412 ::exit(1);
1413 }
1414
1415 // Die immediately, no exit hook, no abort hook, no cleanup.
1416 void os::die() {
1417 ::abort();
1418 }
1419
1420
1421 // This method is a copy of JDK's sysGetLastErrorString
1422 // from src/solaris/hpi/src/system_md.c
1423
1424 size_t os::lasterror(char *buf, size_t len) {
1425 if (errno == 0) return 0;
1426
1427 const char *s = os::strerror(errno);
1428 size_t n = ::strlen(s);
1429 if (n >= len) {
1430 n = len - 1;
1431 }
1432 ::strncpy(buf, s, n);
1433 buf[n] = '\0';
1434 return n;
1435 }
1436
1437 // thread_id is kernel thread id (similar to Solaris LWP id)
1438 intx os::current_thread_id() { return os::Linux::gettid(); }
1439 int os::current_process_id() {
1440 return ::getpid();
1441 }
1442
1443 // DLL functions
1444
1445 const char* os::dll_file_extension() { return ".so"; }
1446
1447 // This must be hard coded because it's the system's temporary
1448 // directory not the java application's temp directory, ala java.io.tmpdir.
1449 const char* os::get_temp_directory() { return "/tmp"; }
1450
1451 static bool file_exists(const char* filename) {
1452 struct stat statbuf;
1453 if (filename == NULL || strlen(filename) == 0) {
1454 return false;
1455 }
1456 return os::stat(filename, &statbuf) == 0;
1457 }
1458
1459 // check if addr is inside libjvm.so
1460 bool os::address_is_in_vm(address addr) {
1461 static address libjvm_base_addr;
1462 Dl_info dlinfo;
1463
1464 if (libjvm_base_addr == NULL) {
1465 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1466 libjvm_base_addr = (address)dlinfo.dli_fbase;
1467 }
1468 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1469 }
1470
1471 if (dladdr((void *)addr, &dlinfo) != 0) {
1472 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1473 }
1474
1475 return false;
1476 }
1477
1478 bool os::dll_address_to_function_name(address addr, char *buf,
1479 int buflen, int *offset,
1480 bool demangle) {
1481 // buf is not optional, but offset is optional
1482 assert(buf != NULL, "sanity check");
1483
1484 Dl_info dlinfo;
1485
1486 if (dladdr((void*)addr, &dlinfo) != 0) {
1487 // see if we have a matching symbol
1488 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1489 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1490 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1491 }
1492 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1493 return true;
1494 }
1495 // no matching symbol so try for just file info
1496 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1497 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1498 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1499 return true;
1500 }
1501 }
1502 }
1503
1504 buf[0] = '\0';
1505 if (offset != NULL) *offset = -1;
1506 return false;
1507 }
1508
1509 struct _address_to_library_name {
1510 address addr; // input : memory address
1511 size_t buflen; // size of fname
1512 char* fname; // output: library name
1513 address base; // library base addr
1514 };
1515
1516 static int address_to_library_name_callback(struct dl_phdr_info *info,
1517 size_t size, void *data) {
1518 int i;
1519 bool found = false;
1520 address libbase = NULL;
1521 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1522
1523 // iterate through all loadable segments
1524 for (i = 0; i < info->dlpi_phnum; i++) {
1525 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1526 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1527 // base address of a library is the lowest address of its loaded
1528 // segments.
1529 if (libbase == NULL || libbase > segbase) {
1530 libbase = segbase;
1531 }
1532 // see if 'addr' is within current segment
1533 if (segbase <= d->addr &&
1534 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1535 found = true;
1536 }
1537 }
1538 }
1539
1540 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1541 // so dll_address_to_library_name() can fall through to use dladdr() which
1542 // can figure out executable name from argv[0].
1543 if (found && info->dlpi_name && info->dlpi_name[0]) {
1544 d->base = libbase;
1545 if (d->fname) {
1546 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1547 }
1548 return 1;
1549 }
1550 return 0;
1551 }
1552
1553 bool os::dll_address_to_library_name(address addr, char* buf,
1554 int buflen, int* offset) {
1555 // buf is not optional, but offset is optional
1556 assert(buf != NULL, "sanity check");
1557
1558 Dl_info dlinfo;
1559 struct _address_to_library_name data;
1560
1561 // There is a bug in old glibc dladdr() implementation that it could resolve
1562 // to wrong library name if the .so file has a base address != NULL. Here
1563 // we iterate through the program headers of all loaded libraries to find
1564 // out which library 'addr' really belongs to. This workaround can be
1565 // removed once the minimum requirement for glibc is moved to 2.3.x.
1566 data.addr = addr;
1567 data.fname = buf;
1568 data.buflen = buflen;
1569 data.base = NULL;
1570 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1571
1572 if (rslt) {
1573 // buf already contains library name
1574 if (offset) *offset = addr - data.base;
1575 return true;
1576 }
1577 if (dladdr((void*)addr, &dlinfo) != 0) {
1578 if (dlinfo.dli_fname != NULL) {
1579 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1580 }
1581 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1582 *offset = addr - (address)dlinfo.dli_fbase;
1583 }
1584 return true;
1585 }
1586
1587 buf[0] = '\0';
1588 if (offset) *offset = -1;
1589 return false;
1590 }
1591
1592 // Loads .dll/.so and
1593 // in case of error it checks if .dll/.so was built for the
1594 // same architecture as Hotspot is running on
1595
1596
1597 // Remember the stack's state. The Linux dynamic linker will change
1598 // the stack to 'executable' at most once, so we must safepoint only once.
1599 bool os::Linux::_stack_is_executable = false;
1600
1601 // VM operation that loads a library. This is necessary if stack protection
1602 // of the Java stacks can be lost during loading the library. If we
1603 // do not stop the Java threads, they can stack overflow before the stacks
1604 // are protected again.
1605 class VM_LinuxDllLoad: public VM_Operation {
1606 private:
1607 const char *_filename;
1608 char *_ebuf;
1609 int _ebuflen;
1610 void *_lib;
1611 public:
1612 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1613 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1614 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1615 void doit() {
1616 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1617 os::Linux::_stack_is_executable = true;
1618 }
1619 void* loaded_library() { return _lib; }
1620 };
1621
1622 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1623 void * result = NULL;
1624 bool load_attempted = false;
1625
1626 // Check whether the library to load might change execution rights
1627 // of the stack. If they are changed, the protection of the stack
1628 // guard pages will be lost. We need a safepoint to fix this.
1629 //
1630 // See Linux man page execstack(8) for more info.
1631 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1632 if (!ElfFile::specifies_noexecstack(filename)) {
1633 if (!is_init_completed()) {
1634 os::Linux::_stack_is_executable = true;
1635 // This is OK - No Java threads have been created yet, and hence no
1636 // stack guard pages to fix.
1637 //
1638 // This should happen only when you are building JDK7 using a very
1639 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1640 //
1641 // Dynamic loader will make all stacks executable after
1642 // this function returns, and will not do that again.
1643 #ifdef ASSERT
1644 ThreadsListHandle tlh;
1645 assert(tlh.length() == 0, "no Java threads should exist yet.");
1646 #endif
1647 } else {
1648 warning("You have loaded library %s which might have disabled stack guard. "
1649 "The VM will try to fix the stack guard now.\n"
1650 "It's highly recommended that you fix the library with "
1651 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1652 filename);
1653
1654 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1655 JavaThread *jt = JavaThread::current();
1656 if (jt->thread_state() != _thread_in_native) {
1657 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1658 // that requires ExecStack. Cannot enter safe point. Let's give up.
1659 warning("Unable to fix stack guard. Giving up.");
1660 } else {
1661 if (!LoadExecStackDllInVMThread) {
1662 // This is for the case where the DLL has an static
1663 // constructor function that executes JNI code. We cannot
1664 // load such DLLs in the VMThread.
1665 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1666 }
1667
1668 ThreadInVMfromNative tiv(jt);
1669 debug_only(VMNativeEntryWrapper vew;)
1670
1671 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1672 VMThread::execute(&op);
1673 if (LoadExecStackDllInVMThread) {
1674 result = op.loaded_library();
1675 }
1676 load_attempted = true;
1677 }
1678 }
1679 }
1680 }
1681
1682 if (!load_attempted) {
1683 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1684 }
1685
1686 if (result != NULL) {
1687 // Successful loading
1688 return result;
1689 }
1690
1691 Elf32_Ehdr elf_head;
1692 int diag_msg_max_length=ebuflen-strlen(ebuf);
1693 char* diag_msg_buf=ebuf+strlen(ebuf);
1694
1695 if (diag_msg_max_length==0) {
1696 // No more space in ebuf for additional diagnostics message
1697 return NULL;
1698 }
1699
1700
1701 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1702
1703 if (file_descriptor < 0) {
1704 // Can't open library, report dlerror() message
1705 return NULL;
1706 }
1707
1708 bool failed_to_read_elf_head=
1709 (sizeof(elf_head)!=
1710 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1711
1712 ::close(file_descriptor);
1713 if (failed_to_read_elf_head) {
1714 // file i/o error - report dlerror() msg
1715 return NULL;
1716 }
1717
1718 typedef struct {
1719 Elf32_Half code; // Actual value as defined in elf.h
1720 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1721 unsigned char elf_class; // 32 or 64 bit
1722 unsigned char endianess; // MSB or LSB
1723 char* name; // String representation
1724 } arch_t;
1725
1726 #ifndef EM_486
1727 #define EM_486 6 /* Intel 80486 */
1728 #endif
1729 #ifndef EM_AARCH64
1730 #define EM_AARCH64 183 /* ARM AARCH64 */
1731 #endif
1732
1733 static const arch_t arch_array[]={
1734 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1736 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1737 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1738 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1740 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1741 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1742 #if defined(VM_LITTLE_ENDIAN)
1743 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1744 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1745 #else
1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1747 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1748 #endif
1749 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1750 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1751 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1752 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1753 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1754 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1755 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1756 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1757 {EM_RISCV, EM_RISCV, ELFCLASS64, ELFDATA2LSB, (char*)"RISCV"},
1758 };
1759
1760 #if (defined IA32)
1761 static Elf32_Half running_arch_code=EM_386;
1762 #elif (defined AMD64)
1763 static Elf32_Half running_arch_code=EM_X86_64;
1764 #elif (defined IA64)
1765 static Elf32_Half running_arch_code=EM_IA_64;
1766 #elif (defined __sparc) && (defined _LP64)
1767 static Elf32_Half running_arch_code=EM_SPARCV9;
1768 #elif (defined __sparc) && (!defined _LP64)
1769 static Elf32_Half running_arch_code=EM_SPARC;
1770 #elif (defined __powerpc64__)
1771 static Elf32_Half running_arch_code=EM_PPC64;
1772 #elif (defined __powerpc__)
1773 static Elf32_Half running_arch_code=EM_PPC;
1774 #elif (defined RISCV)
1775 static Elf32_Half running_arch_code=EM_RISCV;
1776 #elif (defined AARCH64)
1777 static Elf32_Half running_arch_code=EM_AARCH64;
1778 #elif (defined ARM)
1779 static Elf32_Half running_arch_code=EM_ARM;
1780 #elif (defined S390)
1781 static Elf32_Half running_arch_code=EM_S390;
1782 #elif (defined ALPHA)
1783 static Elf32_Half running_arch_code=EM_ALPHA;
1784 #elif (defined MIPSEL)
1785 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1786 #elif (defined PARISC)
1787 static Elf32_Half running_arch_code=EM_PARISC;
1788 #elif (defined MIPS)
1789 static Elf32_Half running_arch_code=EM_MIPS;
1790 #elif (defined M68K)
1791 static Elf32_Half running_arch_code=EM_68K;
1792 #elif (defined SH)
1793 static Elf32_Half running_arch_code=EM_SH;
1794 #else
1795 #error Method os::dll_load requires that one of following is defined:\
1796 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1797 #endif
1798
1799 // Identify compatability class for VM's architecture and library's architecture
1800 // Obtain string descriptions for architectures
1801
1802 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1803 int running_arch_index=-1;
1804
1805 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1806 if (running_arch_code == arch_array[i].code) {
1807 running_arch_index = i;
1808 }
1809 if (lib_arch.code == arch_array[i].code) {
1810 lib_arch.compat_class = arch_array[i].compat_class;
1811 lib_arch.name = arch_array[i].name;
1812 }
1813 }
1814
1815 assert(running_arch_index != -1,
1816 "Didn't find running architecture code (running_arch_code) in arch_array");
1817 if (running_arch_index == -1) {
1818 // Even though running architecture detection failed
1819 // we may still continue with reporting dlerror() message
1820 return NULL;
1821 }
1822
1823 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1824 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1825 return NULL;
1826 }
1827
1828 #ifndef S390
1829 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1830 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1831 return NULL;
1832 }
1833 #endif // !S390
1834
1835 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1836 if (lib_arch.name!=NULL) {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1839 lib_arch.name, arch_array[running_arch_index].name);
1840 } else {
1841 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1843 lib_arch.code,
1844 arch_array[running_arch_index].name);
1845 }
1846 }
1847
1848 return NULL;
1849 }
1850
1851 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1852 int ebuflen) {
1853 void * result = ::dlopen(filename, RTLD_LAZY);
1854 if (result == NULL) {
1855 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1856 ebuf[ebuflen-1] = '\0';
1857 }
1858 return result;
1859 }
1860
1861 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1862 int ebuflen) {
1863 void * result = NULL;
1864 if (LoadExecStackDllInVMThread) {
1865 result = dlopen_helper(filename, ebuf, ebuflen);
1866 }
1867
1868 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1869 // library that requires an executable stack, or which does not have this
1870 // stack attribute set, dlopen changes the stack attribute to executable. The
1871 // read protection of the guard pages gets lost.
1872 //
1873 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1874 // may have been queued at the same time.
1875
1876 if (!_stack_is_executable) {
1877 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1878 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1879 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1880 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1881 warning("Attempt to reguard stack yellow zone failed.");
1882 }
1883 }
1884 }
1885 }
1886
1887 return result;
1888 }
1889
1890 void* os::dll_lookup(void* handle, const char* name) {
1891 void* res = dlsym(handle, name);
1892 return res;
1893 }
1894
1895 void* os::get_default_process_handle() {
1896 return (void*)::dlopen(NULL, RTLD_LAZY);
1897 }
1898
1899 static bool _print_ascii_file(const char* filename, outputStream* st) {
1900 int fd = ::open(filename, O_RDONLY);
1901 if (fd == -1) {
1902 return false;
1903 }
1904
1905 char buf[33];
1906 int bytes;
1907 buf[32] = '\0';
1908 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1909 st->print_raw(buf, bytes);
1910 }
1911
1912 ::close(fd);
1913
1914 return true;
1915 }
1916
1917 void os::print_dll_info(outputStream *st) {
1918 st->print_cr("Dynamic libraries:");
1919
1920 char fname[32];
1921 pid_t pid = os::Linux::gettid();
1922
1923 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1924
1925 if (!_print_ascii_file(fname, st)) {
1926 st->print("Can not get library information for pid = %d\n", pid);
1927 }
1928 }
1929
1930 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1931 FILE *procmapsFile = NULL;
1932
1933 // Open the procfs maps file for the current process
1934 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1935 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1936 char line[PATH_MAX + 100];
1937
1938 // Read line by line from 'file'
1939 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1940 u8 base, top, offset, inode;
1941 char permissions[5];
1942 char device[6];
1943 char name[PATH_MAX + 1];
1944
1945 // Parse fields from line
1946 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1947 &base, &top, permissions, &offset, device, &inode, name);
1948
1949 // Filter by device id '00:00' so that we only get file system mapped files.
1950 if (strcmp(device, "00:00") != 0) {
1951
1952 // Call callback with the fields of interest
1953 if(callback(name, (address)base, (address)top, param)) {
1954 // Oops abort, callback aborted
1955 fclose(procmapsFile);
1956 return 1;
1957 }
1958 }
1959 }
1960 fclose(procmapsFile);
1961 }
1962 return 0;
1963 }
1964
1965 void os::print_os_info_brief(outputStream* st) {
1966 os::Linux::print_distro_info(st);
1967
1968 os::Posix::print_uname_info(st);
1969
1970 os::Linux::print_libversion_info(st);
1971
1972 }
1973
1974 void os::print_os_info(outputStream* st) {
1975 st->print("OS:");
1976
1977 os::Linux::print_distro_info(st);
1978
1979 os::Posix::print_uname_info(st);
1980
1981 // Print warning if unsafe chroot environment detected
1982 if (unsafe_chroot_detected) {
1983 st->print("WARNING!! ");
1984 st->print_cr("%s", unstable_chroot_error);
1985 }
1986
1987 os::Linux::print_libversion_info(st);
1988
1989 os::Posix::print_rlimit_info(st);
1990
1991 os::Posix::print_load_average(st);
1992
1993 os::Linux::print_full_memory_info(st);
1994
1995 os::Linux::print_container_info(st);
1996 }
1997
1998 // Try to identify popular distros.
1999 // Most Linux distributions have a /etc/XXX-release file, which contains
2000 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2001 // file that also contains the OS version string. Some have more than one
2002 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2003 // /etc/redhat-release.), so the order is important.
2004 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2005 // their own specific XXX-release file as well as a redhat-release file.
2006 // Because of this the XXX-release file needs to be searched for before the
2007 // redhat-release file.
2008 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2009 // search for redhat-release / SuSE-release needs to be before lsb-release.
2010 // Since the lsb-release file is the new standard it needs to be searched
2011 // before the older style release files.
2012 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2013 // next to last resort. The os-release file is a new standard that contains
2014 // distribution information and the system-release file seems to be an old
2015 // standard that has been replaced by the lsb-release and os-release files.
2016 // Searching for the debian_version file is the last resort. It contains
2017 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2018 // "Debian " is printed before the contents of the debian_version file.
2019
2020 const char* distro_files[] = {
2021 "/etc/oracle-release",
2022 "/etc/mandriva-release",
2023 "/etc/mandrake-release",
2024 "/etc/sun-release",
2025 "/etc/redhat-release",
2026 "/etc/SuSE-release",
2027 "/etc/lsb-release",
2028 "/etc/turbolinux-release",
2029 "/etc/gentoo-release",
2030 "/etc/ltib-release",
2031 "/etc/angstrom-version",
2032 "/etc/system-release",
2033 "/etc/os-release",
2034 NULL };
2035
2036 void os::Linux::print_distro_info(outputStream* st) {
2037 for (int i = 0;; i++) {
2038 const char* file = distro_files[i];
2039 if (file == NULL) {
2040 break; // done
2041 }
2042 // If file prints, we found it.
2043 if (_print_ascii_file(file, st)) {
2044 return;
2045 }
2046 }
2047
2048 if (file_exists("/etc/debian_version")) {
2049 st->print("Debian ");
2050 _print_ascii_file("/etc/debian_version", st);
2051 } else {
2052 st->print("Linux");
2053 }
2054 st->cr();
2055 }
2056
2057 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2058 char buf[256];
2059 while (fgets(buf, sizeof(buf), fp)) {
2060 // Edit out extra stuff in expected format
2061 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2062 char* ptr = strstr(buf, "\""); // the name is in quotes
2063 if (ptr != NULL) {
2064 ptr++; // go beyond first quote
2065 char* nl = strchr(ptr, '\"');
2066 if (nl != NULL) *nl = '\0';
2067 strncpy(distro, ptr, length);
2068 } else {
2069 ptr = strstr(buf, "=");
2070 ptr++; // go beyond equals then
2071 char* nl = strchr(ptr, '\n');
2072 if (nl != NULL) *nl = '\0';
2073 strncpy(distro, ptr, length);
2074 }
2075 return;
2076 } else if (get_first_line) {
2077 char* nl = strchr(buf, '\n');
2078 if (nl != NULL) *nl = '\0';
2079 strncpy(distro, buf, length);
2080 return;
2081 }
2082 }
2083 // print last line and close
2084 char* nl = strchr(buf, '\n');
2085 if (nl != NULL) *nl = '\0';
2086 strncpy(distro, buf, length);
2087 }
2088
2089 static void parse_os_info(char* distro, size_t length, const char* file) {
2090 FILE* fp = fopen(file, "r");
2091 if (fp != NULL) {
2092 // if suse format, print out first line
2093 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2094 parse_os_info_helper(fp, distro, length, get_first_line);
2095 fclose(fp);
2096 }
2097 }
2098
2099 void os::get_summary_os_info(char* buf, size_t buflen) {
2100 for (int i = 0;; i++) {
2101 const char* file = distro_files[i];
2102 if (file == NULL) {
2103 break; // ran out of distro_files
2104 }
2105 if (file_exists(file)) {
2106 parse_os_info(buf, buflen, file);
2107 return;
2108 }
2109 }
2110 // special case for debian
2111 if (file_exists("/etc/debian_version")) {
2112 strncpy(buf, "Debian ", buflen);
2113 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2114 } else {
2115 strncpy(buf, "Linux", buflen);
2116 }
2117 }
2118
2119 void os::Linux::print_libversion_info(outputStream* st) {
2120 // libc, pthread
2121 st->print("libc:");
2122 st->print("%s ", os::Linux::glibc_version());
2123 st->print("%s ", os::Linux::libpthread_version());
2124 st->cr();
2125 }
2126
2127 void os::Linux::print_full_memory_info(outputStream* st) {
2128 st->print("\n/proc/meminfo:\n");
2129 _print_ascii_file("/proc/meminfo", st);
2130 st->cr();
2131 }
2132
2133 void os::Linux::print_container_info(outputStream* st) {
2134 if (!OSContainer::is_containerized()) {
2135 return;
2136 }
2137
2138 st->print("container (cgroup) information:\n");
2139
2140 const char *p_ct = OSContainer::container_type();
2141 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2142
2143 char *p = OSContainer::cpu_cpuset_cpus();
2144 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2145 free(p);
2146
2147 p = OSContainer::cpu_cpuset_memory_nodes();
2148 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2149 free(p);
2150
2151 int i = OSContainer::active_processor_count();
2152 if (i > 0) {
2153 st->print("active_processor_count: %d\n", i);
2154 } else {
2155 st->print("active_processor_count: failed\n");
2156 }
2157
2158 i = OSContainer::cpu_quota();
2159 st->print("cpu_quota: %d\n", i);
2160
2161 i = OSContainer::cpu_period();
2162 st->print("cpu_period: %d\n", i);
2163
2164 i = OSContainer::cpu_shares();
2165 st->print("cpu_shares: %d\n", i);
2166
2167 jlong j = OSContainer::memory_limit_in_bytes();
2168 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2169
2170 j = OSContainer::memory_and_swap_limit_in_bytes();
2171 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2172
2173 j = OSContainer::memory_soft_limit_in_bytes();
2174 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2175
2176 j = OSContainer::OSContainer::memory_usage_in_bytes();
2177 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2178
2179 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2180 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2181 st->cr();
2182 }
2183
2184 void os::print_memory_info(outputStream* st) {
2185
2186 st->print("Memory:");
2187 st->print(" %dk page", os::vm_page_size()>>10);
2188
2189 // values in struct sysinfo are "unsigned long"
2190 struct sysinfo si;
2191 sysinfo(&si);
2192
2193 st->print(", physical " UINT64_FORMAT "k",
2194 os::physical_memory() >> 10);
2195 st->print("(" UINT64_FORMAT "k free)",
2196 os::available_memory() >> 10);
2197 st->print(", swap " UINT64_FORMAT "k",
2198 ((jlong)si.totalswap * si.mem_unit) >> 10);
2199 st->print("(" UINT64_FORMAT "k free)",
2200 ((jlong)si.freeswap * si.mem_unit) >> 10);
2201 st->cr();
2202 }
2203
2204 // Print the first "model name" line and the first "flags" line
2205 // that we find and nothing more. We assume "model name" comes
2206 // before "flags" so if we find a second "model name", then the
2207 // "flags" field is considered missing.
2208 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2209 #if defined(IA32) || defined(AMD64)
2210 // Other platforms have less repetitive cpuinfo files
2211 FILE *fp = fopen("/proc/cpuinfo", "r");
2212 if (fp) {
2213 while (!feof(fp)) {
2214 if (fgets(buf, buflen, fp)) {
2215 // Assume model name comes before flags
2216 bool model_name_printed = false;
2217 if (strstr(buf, "model name") != NULL) {
2218 if (!model_name_printed) {
2219 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2220 st->print_raw(buf);
2221 model_name_printed = true;
2222 } else {
2223 // model name printed but not flags? Odd, just return
2224 fclose(fp);
2225 return true;
2226 }
2227 }
2228 // print the flags line too
2229 if (strstr(buf, "flags") != NULL) {
2230 st->print_raw(buf);
2231 fclose(fp);
2232 return true;
2233 }
2234 }
2235 }
2236 fclose(fp);
2237 }
2238 #endif // x86 platforms
2239 return false;
2240 }
2241
2242 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2243 // Only print the model name if the platform provides this as a summary
2244 if (!print_model_name_and_flags(st, buf, buflen)) {
2245 st->print("\n/proc/cpuinfo:\n");
2246 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2247 st->print_cr(" <Not Available>");
2248 }
2249 }
2250 }
2251
2252 #if defined(AMD64) || defined(IA32) || defined(X32)
2253 const char* search_string = "model name";
2254 #elif defined(M68K)
2255 const char* search_string = "CPU";
2256 #elif defined(PPC64)
2257 const char* search_string = "cpu";
2258 #elif defined(S390)
2259 const char* search_string = "processor";
2260 #elif defined(SPARC)
2261 const char* search_string = "cpu";
2262 #else
2263 const char* search_string = "Processor";
2264 #endif
2265
2266 // Parses the cpuinfo file for string representing the model name.
2267 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2268 FILE* fp = fopen("/proc/cpuinfo", "r");
2269 if (fp != NULL) {
2270 while (!feof(fp)) {
2271 char buf[256];
2272 if (fgets(buf, sizeof(buf), fp)) {
2273 char* start = strstr(buf, search_string);
2274 if (start != NULL) {
2275 char *ptr = start + strlen(search_string);
2276 char *end = buf + strlen(buf);
2277 while (ptr != end) {
2278 // skip whitespace and colon for the rest of the name.
2279 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2280 break;
2281 }
2282 ptr++;
2283 }
2284 if (ptr != end) {
2285 // reasonable string, get rid of newline and keep the rest
2286 char* nl = strchr(buf, '\n');
2287 if (nl != NULL) *nl = '\0';
2288 strncpy(cpuinfo, ptr, length);
2289 fclose(fp);
2290 return;
2291 }
2292 }
2293 }
2294 }
2295 fclose(fp);
2296 }
2297 // cpuinfo not found or parsing failed, just print generic string. The entire
2298 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2299 #if defined(AARCH64)
2300 strncpy(cpuinfo, "AArch64", length);
2301 #elif defined(AMD64)
2302 strncpy(cpuinfo, "x86_64", length);
2303 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2304 strncpy(cpuinfo, "ARM", length);
2305 #elif defined(IA32)
2306 strncpy(cpuinfo, "x86_32", length);
2307 #elif defined(IA64)
2308 strncpy(cpuinfo, "IA64", length);
2309 #elif defined(PPC)
2310 strncpy(cpuinfo, "PPC64", length);
2311 #elif defined(RISCV)
2312 strncpy(cpuinfo, "RISCV", length);
2313 #elif defined(S390)
2314 strncpy(cpuinfo, "S390", length);
2315 #elif defined(SPARC)
2316 strncpy(cpuinfo, "sparcv9", length);
2317 #elif defined(ZERO_LIBARCH)
2318 strncpy(cpuinfo, ZERO_LIBARCH, length);
2319 #else
2320 strncpy(cpuinfo, "unknown", length);
2321 #endif
2322 }
2323
2324 static void print_signal_handler(outputStream* st, int sig,
2325 char* buf, size_t buflen);
2326
2327 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2328 st->print_cr("Signal Handlers:");
2329 print_signal_handler(st, SIGSEGV, buf, buflen);
2330 print_signal_handler(st, SIGBUS , buf, buflen);
2331 print_signal_handler(st, SIGFPE , buf, buflen);
2332 print_signal_handler(st, SIGPIPE, buf, buflen);
2333 print_signal_handler(st, SIGXFSZ, buf, buflen);
2334 print_signal_handler(st, SIGILL , buf, buflen);
2335 print_signal_handler(st, SR_signum, buf, buflen);
2336 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2337 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2338 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2339 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2340 #if defined(PPC64)
2341 print_signal_handler(st, SIGTRAP, buf, buflen);
2342 #endif
2343 }
2344
2345 static char saved_jvm_path[MAXPATHLEN] = {0};
2346
2347 // Find the full path to the current module, libjvm.so
2348 void os::jvm_path(char *buf, jint buflen) {
2349 // Error checking.
2350 if (buflen < MAXPATHLEN) {
2351 assert(false, "must use a large-enough buffer");
2352 buf[0] = '\0';
2353 return;
2354 }
2355 // Lazy resolve the path to current module.
2356 if (saved_jvm_path[0] != 0) {
2357 strcpy(buf, saved_jvm_path);
2358 return;
2359 }
2360
2361 char dli_fname[MAXPATHLEN];
2362 bool ret = dll_address_to_library_name(
2363 CAST_FROM_FN_PTR(address, os::jvm_path),
2364 dli_fname, sizeof(dli_fname), NULL);
2365 assert(ret, "cannot locate libjvm");
2366 char *rp = NULL;
2367 if (ret && dli_fname[0] != '\0') {
2368 rp = os::Posix::realpath(dli_fname, buf, buflen);
2369 }
2370 if (rp == NULL) {
2371 return;
2372 }
2373
2374 if (Arguments::sun_java_launcher_is_altjvm()) {
2375 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2376 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2377 // If "/jre/lib/" appears at the right place in the string, then
2378 // assume we are installed in a JDK and we're done. Otherwise, check
2379 // for a JAVA_HOME environment variable and fix up the path so it
2380 // looks like libjvm.so is installed there (append a fake suffix
2381 // hotspot/libjvm.so).
2382 const char *p = buf + strlen(buf) - 1;
2383 for (int count = 0; p > buf && count < 5; ++count) {
2384 for (--p; p > buf && *p != '/'; --p)
2385 /* empty */ ;
2386 }
2387
2388 if (strncmp(p, "/jre/lib/", 9) != 0) {
2389 // Look for JAVA_HOME in the environment.
2390 char* java_home_var = ::getenv("JAVA_HOME");
2391 if (java_home_var != NULL && java_home_var[0] != 0) {
2392 char* jrelib_p;
2393 int len;
2394
2395 // Check the current module name "libjvm.so".
2396 p = strrchr(buf, '/');
2397 if (p == NULL) {
2398 return;
2399 }
2400 assert(strstr(p, "/libjvm") == p, "invalid library name");
2401
2402 rp = os::Posix::realpath(java_home_var, buf, buflen);
2403 if (rp == NULL) {
2404 return;
2405 }
2406
2407 // determine if this is a legacy image or modules image
2408 // modules image doesn't have "jre" subdirectory
2409 len = strlen(buf);
2410 assert(len < buflen, "Ran out of buffer room");
2411 jrelib_p = buf + len;
2412 snprintf(jrelib_p, buflen-len, "/jre/lib");
2413 if (0 != access(buf, F_OK)) {
2414 snprintf(jrelib_p, buflen-len, "/lib");
2415 }
2416
2417 if (0 == access(buf, F_OK)) {
2418 // Use current module name "libjvm.so"
2419 len = strlen(buf);
2420 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2421 } else {
2422 // Go back to path of .so
2423 rp = os::Posix::realpath(dli_fname, buf, buflen);
2424 if (rp == NULL) {
2425 return;
2426 }
2427 }
2428 }
2429 }
2430 }
2431
2432 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2433 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2434 }
2435
2436 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2437 // no prefix required, not even "_"
2438 }
2439
2440 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2441 // no suffix required
2442 }
2443
2444 ////////////////////////////////////////////////////////////////////////////////
2445 // sun.misc.Signal support
2446
2447 static volatile jint sigint_count = 0;
2448
2449 static void UserHandler(int sig, void *siginfo, void *context) {
2450 // 4511530 - sem_post is serialized and handled by the manager thread. When
2451 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2452 // don't want to flood the manager thread with sem_post requests.
2453 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2454 return;
2455 }
2456
2457 // Ctrl-C is pressed during error reporting, likely because the error
2458 // handler fails to abort. Let VM die immediately.
2459 if (sig == SIGINT && VMError::is_error_reported()) {
2460 os::die();
2461 }
2462
2463 os::signal_notify(sig);
2464 }
2465
2466 void* os::user_handler() {
2467 return CAST_FROM_FN_PTR(void*, UserHandler);
2468 }
2469
2470 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2471 struct timespec ts;
2472 // Semaphore's are always associated with CLOCK_REALTIME
2473 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2474 // see os_posix.cpp for discussion on overflow checking
2475 if (sec >= MAX_SECS) {
2476 ts.tv_sec += MAX_SECS;
2477 ts.tv_nsec = 0;
2478 } else {
2479 ts.tv_sec += sec;
2480 ts.tv_nsec += nsec;
2481 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2482 ts.tv_nsec -= NANOSECS_PER_SEC;
2483 ++ts.tv_sec; // note: this must be <= max_secs
2484 }
2485 }
2486
2487 return ts;
2488 }
2489
2490 extern "C" {
2491 typedef void (*sa_handler_t)(int);
2492 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2493 }
2494
2495 void* os::signal(int signal_number, void* handler) {
2496 struct sigaction sigAct, oldSigAct;
2497
2498 sigfillset(&(sigAct.sa_mask));
2499 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2500 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2501
2502 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2503 // -1 means registration failed
2504 return (void *)-1;
2505 }
2506
2507 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2508 }
2509
2510 void os::signal_raise(int signal_number) {
2511 ::raise(signal_number);
2512 }
2513
2514 // The following code is moved from os.cpp for making this
2515 // code platform specific, which it is by its very nature.
2516
2517 // Will be modified when max signal is changed to be dynamic
2518 int os::sigexitnum_pd() {
2519 return NSIG;
2520 }
2521
2522 // a counter for each possible signal value
2523 static volatile jint pending_signals[NSIG+1] = { 0 };
2524
2525 // Linux(POSIX) specific hand shaking semaphore.
2526 static Semaphore* sig_sem = NULL;
2527 static PosixSemaphore sr_semaphore;
2528
2529 void os::signal_init_pd() {
2530 // Initialize signal structures
2531 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2532
2533 // Initialize signal semaphore
2534 sig_sem = new Semaphore();
2535 }
2536
2537 void os::signal_notify(int sig) {
2538 if (sig_sem != NULL) {
2539 Atomic::inc(&pending_signals[sig]);
2540 sig_sem->signal();
2541 } else {
2542 // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2543 // initialization isn't called.
2544 assert(ReduceSignalUsage, "signal semaphore should be created");
2545 }
2546 }
2547
2548 static int check_pending_signals() {
2549 Atomic::store(0, &sigint_count);
2550 for (;;) {
2551 for (int i = 0; i < NSIG + 1; i++) {
2552 jint n = pending_signals[i];
2553 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2554 return i;
2555 }
2556 }
2557 JavaThread *thread = JavaThread::current();
2558 ThreadBlockInVM tbivm(thread);
2559
2560 bool threadIsSuspended;
2561 do {
2562 thread->set_suspend_equivalent();
2563 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2564 sig_sem->wait();
2565
2566 // were we externally suspended while we were waiting?
2567 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2568 if (threadIsSuspended) {
2569 // The semaphore has been incremented, but while we were waiting
2570 // another thread suspended us. We don't want to continue running
2571 // while suspended because that would surprise the thread that
2572 // suspended us.
2573 sig_sem->signal();
2574
2575 thread->java_suspend_self();
2576 }
2577 } while (threadIsSuspended);
2578 }
2579 }
2580
2581 int os::signal_wait() {
2582 return check_pending_signals();
2583 }
2584
2585 ////////////////////////////////////////////////////////////////////////////////
2586 // Virtual Memory
2587
2588 int os::vm_page_size() {
2589 // Seems redundant as all get out
2590 assert(os::Linux::page_size() != -1, "must call os::init");
2591 return os::Linux::page_size();
2592 }
2593
2594 // Solaris allocates memory by pages.
2595 int os::vm_allocation_granularity() {
2596 assert(os::Linux::page_size() != -1, "must call os::init");
2597 return os::Linux::page_size();
2598 }
2599
2600 // Rationale behind this function:
2601 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2602 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2603 // samples for JITted code. Here we create private executable mapping over the code cache
2604 // and then we can use standard (well, almost, as mapping can change) way to provide
2605 // info for the reporting script by storing timestamp and location of symbol
2606 void linux_wrap_code(char* base, size_t size) {
2607 static volatile jint cnt = 0;
2608
2609 if (!UseOprofile) {
2610 return;
2611 }
2612
2613 char buf[PATH_MAX+1];
2614 int num = Atomic::add(1, &cnt);
2615
2616 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2617 os::get_temp_directory(), os::current_process_id(), num);
2618 unlink(buf);
2619
2620 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2621
2622 if (fd != -1) {
2623 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2624 if (rv != (off_t)-1) {
2625 if (::write(fd, "", 1) == 1) {
2626 mmap(base, size,
2627 PROT_READ|PROT_WRITE|PROT_EXEC,
2628 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2629 }
2630 }
2631 ::close(fd);
2632 unlink(buf);
2633 }
2634 }
2635
2636 static bool recoverable_mmap_error(int err) {
2637 // See if the error is one we can let the caller handle. This
2638 // list of errno values comes from JBS-6843484. I can't find a
2639 // Linux man page that documents this specific set of errno
2640 // values so while this list currently matches Solaris, it may
2641 // change as we gain experience with this failure mode.
2642 switch (err) {
2643 case EBADF:
2644 case EINVAL:
2645 case ENOTSUP:
2646 // let the caller deal with these errors
2647 return true;
2648
2649 default:
2650 // Any remaining errors on this OS can cause our reserved mapping
2651 // to be lost. That can cause confusion where different data
2652 // structures think they have the same memory mapped. The worst
2653 // scenario is if both the VM and a library think they have the
2654 // same memory mapped.
2655 return false;
2656 }
2657 }
2658
2659 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2660 int err) {
2661 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2662 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2663 os::strerror(err), err);
2664 }
2665
2666 static void warn_fail_commit_memory(char* addr, size_t size,
2667 size_t alignment_hint, bool exec,
2668 int err) {
2669 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2670 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2671 alignment_hint, exec, os::strerror(err), err);
2672 }
2673
2674 // NOTE: Linux kernel does not really reserve the pages for us.
2675 // All it does is to check if there are enough free pages
2676 // left at the time of mmap(). This could be a potential
2677 // problem.
2678 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2679 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2680 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2681 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2682 if (res != (uintptr_t) MAP_FAILED) {
2683 if (UseNUMAInterleaving) {
2684 numa_make_global(addr, size);
2685 }
2686 return 0;
2687 }
2688
2689 int err = errno; // save errno from mmap() call above
2690
2691 if (!recoverable_mmap_error(err)) {
2692 warn_fail_commit_memory(addr, size, exec, err);
2693 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2694 }
2695
2696 return err;
2697 }
2698
2699 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2700 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2701 }
2702
2703 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2704 const char* mesg) {
2705 assert(mesg != NULL, "mesg must be specified");
2706 int err = os::Linux::commit_memory_impl(addr, size, exec);
2707 if (err != 0) {
2708 // the caller wants all commit errors to exit with the specified mesg:
2709 warn_fail_commit_memory(addr, size, exec, err);
2710 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2711 }
2712 }
2713
2714 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2715 #ifndef MAP_HUGETLB
2716 #define MAP_HUGETLB 0x40000
2717 #endif
2718
2719 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2720 #ifndef MADV_HUGEPAGE
2721 #define MADV_HUGEPAGE 14
2722 #endif
2723
2724 int os::Linux::commit_memory_impl(char* addr, size_t size,
2725 size_t alignment_hint, bool exec) {
2726 int err = os::Linux::commit_memory_impl(addr, size, exec);
2727 if (err == 0) {
2728 realign_memory(addr, size, alignment_hint);
2729 }
2730 return err;
2731 }
2732
2733 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2734 bool exec) {
2735 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2736 }
2737
2738 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2739 size_t alignment_hint, bool exec,
2740 const char* mesg) {
2741 assert(mesg != NULL, "mesg must be specified");
2742 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2743 if (err != 0) {
2744 // the caller wants all commit errors to exit with the specified mesg:
2745 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2746 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2747 }
2748 }
2749
2750 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2751 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2752 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2753 // be supported or the memory may already be backed by huge pages.
2754 ::madvise(addr, bytes, MADV_HUGEPAGE);
2755 }
2756 }
2757
2758 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2759 // This method works by doing an mmap over an existing mmaping and effectively discarding
2760 // the existing pages. However it won't work for SHM-based large pages that cannot be
2761 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2762 // small pages on top of the SHM segment. This method always works for small pages, so we
2763 // allow that in any case.
2764 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2765 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2766 }
2767 }
2768
2769 void os::numa_make_global(char *addr, size_t bytes) {
2770 Linux::numa_interleave_memory(addr, bytes);
2771 }
2772
2773 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2774 // bind policy to MPOL_PREFERRED for the current thread.
2775 #define USE_MPOL_PREFERRED 0
2776
2777 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2778 // To make NUMA and large pages more robust when both enabled, we need to ease
2779 // the requirements on where the memory should be allocated. MPOL_BIND is the
2780 // default policy and it will force memory to be allocated on the specified
2781 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2782 // the specified node, but will not force it. Using this policy will prevent
2783 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2784 // free large pages.
2785 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2786 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2787 }
2788
2789 bool os::numa_topology_changed() { return false; }
2790
2791 size_t os::numa_get_groups_num() {
2792 // Return just the number of nodes in which it's possible to allocate memory
2793 // (in numa terminology, configured nodes).
2794 return Linux::numa_num_configured_nodes();
2795 }
2796
2797 int os::numa_get_group_id() {
2798 int cpu_id = Linux::sched_getcpu();
2799 if (cpu_id != -1) {
2800 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2801 if (lgrp_id != -1) {
2802 return lgrp_id;
2803 }
2804 }
2805 return 0;
2806 }
2807
2808 int os::Linux::get_existing_num_nodes() {
2809 size_t node;
2810 size_t highest_node_number = Linux::numa_max_node();
2811 int num_nodes = 0;
2812
2813 // Get the total number of nodes in the system including nodes without memory.
2814 for (node = 0; node <= highest_node_number; node++) {
2815 if (isnode_in_existing_nodes(node)) {
2816 num_nodes++;
2817 }
2818 }
2819 return num_nodes;
2820 }
2821
2822 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2823 size_t highest_node_number = Linux::numa_max_node();
2824 size_t i = 0;
2825
2826 // Map all node ids in which is possible to allocate memory. Also nodes are
2827 // not always consecutively available, i.e. available from 0 to the highest
2828 // node number.
2829 for (size_t node = 0; node <= highest_node_number; node++) {
2830 if (Linux::isnode_in_configured_nodes(node)) {
2831 ids[i++] = node;
2832 }
2833 }
2834 return i;
2835 }
2836
2837 bool os::get_page_info(char *start, page_info* info) {
2838 return false;
2839 }
2840
2841 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2842 page_info* page_found) {
2843 return end;
2844 }
2845
2846
2847 int os::Linux::sched_getcpu_syscall(void) {
2848 unsigned int cpu = 0;
2849 int retval = -1;
2850
2851 #if defined(IA32)
2852 #ifndef SYS_getcpu
2853 #define SYS_getcpu 318
2854 #endif
2855 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2856 #elif defined(AMD64)
2857 // Unfortunately we have to bring all these macros here from vsyscall.h
2858 // to be able to compile on old linuxes.
2859 #define __NR_vgetcpu 2
2860 #define VSYSCALL_START (-10UL << 20)
2861 #define VSYSCALL_SIZE 1024
2862 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2863 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2864 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2865 retval = vgetcpu(&cpu, NULL, NULL);
2866 #endif
2867
2868 return (retval == -1) ? retval : cpu;
2869 }
2870
2871 void os::Linux::sched_getcpu_init() {
2872 // sched_getcpu() should be in libc.
2873 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2874 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2875
2876 // If it's not, try a direct syscall.
2877 if (sched_getcpu() == -1) {
2878 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2879 (void*)&sched_getcpu_syscall));
2880 }
2881 }
2882
2883 // Something to do with the numa-aware allocator needs these symbols
2884 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2885 extern "C" JNIEXPORT void numa_error(char *where) { }
2886
2887 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2888 // load symbol from base version instead.
2889 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2890 void *f = dlvsym(handle, name, "libnuma_1.1");
2891 if (f == NULL) {
2892 f = dlsym(handle, name);
2893 }
2894 return f;
2895 }
2896
2897 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2898 // Return NULL if the symbol is not defined in this particular version.
2899 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2900 return dlvsym(handle, name, "libnuma_1.2");
2901 }
2902
2903 bool os::Linux::libnuma_init() {
2904 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2905 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2906 if (handle != NULL) {
2907 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2908 libnuma_dlsym(handle, "numa_node_to_cpus")));
2909 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2910 libnuma_dlsym(handle, "numa_max_node")));
2911 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2912 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2913 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2914 libnuma_dlsym(handle, "numa_available")));
2915 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2916 libnuma_dlsym(handle, "numa_tonode_memory")));
2917 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2918 libnuma_dlsym(handle, "numa_interleave_memory")));
2919 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2920 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2921 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2922 libnuma_dlsym(handle, "numa_set_bind_policy")));
2923 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2924 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2925 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2926 libnuma_dlsym(handle, "numa_distance")));
2927
2928 if (numa_available() != -1) {
2929 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2930 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2931 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2932 // Create an index -> node mapping, since nodes are not always consecutive
2933 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2934 rebuild_nindex_to_node_map();
2935 // Create a cpu -> node mapping
2936 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2937 rebuild_cpu_to_node_map();
2938 return true;
2939 }
2940 }
2941 }
2942 return false;
2943 }
2944
2945 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2946 // Creating guard page is very expensive. Java thread has HotSpot
2947 // guard pages, only enable glibc guard page for non-Java threads.
2948 // (Remember: compiler thread is a Java thread, too!)
2949 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2950 }
2951
2952 void os::Linux::rebuild_nindex_to_node_map() {
2953 int highest_node_number = Linux::numa_max_node();
2954
2955 nindex_to_node()->clear();
2956 for (int node = 0; node <= highest_node_number; node++) {
2957 if (Linux::isnode_in_existing_nodes(node)) {
2958 nindex_to_node()->append(node);
2959 }
2960 }
2961 }
2962
2963 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2964 // The table is later used in get_node_by_cpu().
2965 void os::Linux::rebuild_cpu_to_node_map() {
2966 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2967 // in libnuma (possible values are starting from 16,
2968 // and continuing up with every other power of 2, but less
2969 // than the maximum number of CPUs supported by kernel), and
2970 // is a subject to change (in libnuma version 2 the requirements
2971 // are more reasonable) we'll just hardcode the number they use
2972 // in the library.
2973 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2974
2975 size_t cpu_num = processor_count();
2976 size_t cpu_map_size = NCPUS / BitsPerCLong;
2977 size_t cpu_map_valid_size =
2978 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2979
2980 cpu_to_node()->clear();
2981 cpu_to_node()->at_grow(cpu_num - 1);
2982
2983 size_t node_num = get_existing_num_nodes();
2984
2985 int distance = 0;
2986 int closest_distance = INT_MAX;
2987 int closest_node = 0;
2988 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2989 for (size_t i = 0; i < node_num; i++) {
2990 // Check if node is configured (not a memory-less node). If it is not, find
2991 // the closest configured node.
2992 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2993 closest_distance = INT_MAX;
2994 // Check distance from all remaining nodes in the system. Ignore distance
2995 // from itself and from another non-configured node.
2996 for (size_t m = 0; m < node_num; m++) {
2997 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2998 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2999 // If a closest node is found, update. There is always at least one
3000 // configured node in the system so there is always at least one node
3001 // close.
3002 if (distance != 0 && distance < closest_distance) {
3003 closest_distance = distance;
3004 closest_node = nindex_to_node()->at(m);
3005 }
3006 }
3007 }
3008 } else {
3009 // Current node is already a configured node.
3010 closest_node = nindex_to_node()->at(i);
3011 }
3012
3013 // Get cpus from the original node and map them to the closest node. If node
3014 // is a configured node (not a memory-less node), then original node and
3015 // closest node are the same.
3016 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3017 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3018 if (cpu_map[j] != 0) {
3019 for (size_t k = 0; k < BitsPerCLong; k++) {
3020 if (cpu_map[j] & (1UL << k)) {
3021 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3022 }
3023 }
3024 }
3025 }
3026 }
3027 }
3028 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3029 }
3030
3031 int os::Linux::get_node_by_cpu(int cpu_id) {
3032 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3033 return cpu_to_node()->at(cpu_id);
3034 }
3035 return -1;
3036 }
3037
3038 GrowableArray<int>* os::Linux::_cpu_to_node;
3039 GrowableArray<int>* os::Linux::_nindex_to_node;
3040 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3041 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3042 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3043 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3044 os::Linux::numa_available_func_t os::Linux::_numa_available;
3045 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3046 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3047 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3048 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3049 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3050 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3051 unsigned long* os::Linux::_numa_all_nodes;
3052 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3053 struct bitmask* os::Linux::_numa_nodes_ptr;
3054
3055 bool os::pd_uncommit_memory(char* addr, size_t size) {
3056 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3057 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3058 return res != (uintptr_t) MAP_FAILED;
3059 }
3060
3061 static address get_stack_commited_bottom(address bottom, size_t size) {
3062 address nbot = bottom;
3063 address ntop = bottom + size;
3064
3065 size_t page_sz = os::vm_page_size();
3066 unsigned pages = size / page_sz;
3067
3068 unsigned char vec[1];
3069 unsigned imin = 1, imax = pages + 1, imid;
3070 int mincore_return_value = 0;
3071
3072 assert(imin <= imax, "Unexpected page size");
3073
3074 while (imin < imax) {
3075 imid = (imax + imin) / 2;
3076 nbot = ntop - (imid * page_sz);
3077
3078 // Use a trick with mincore to check whether the page is mapped or not.
3079 // mincore sets vec to 1 if page resides in memory and to 0 if page
3080 // is swapped output but if page we are asking for is unmapped
3081 // it returns -1,ENOMEM
3082 mincore_return_value = mincore(nbot, page_sz, vec);
3083
3084 if (mincore_return_value == -1) {
3085 // Page is not mapped go up
3086 // to find first mapped page
3087 if (errno != EAGAIN) {
3088 assert(errno == ENOMEM, "Unexpected mincore errno");
3089 imax = imid;
3090 }
3091 } else {
3092 // Page is mapped go down
3093 // to find first not mapped page
3094 imin = imid + 1;
3095 }
3096 }
3097
3098 nbot = nbot + page_sz;
3099
3100 // Adjust stack bottom one page up if last checked page is not mapped
3101 if (mincore_return_value == -1) {
3102 nbot = nbot + page_sz;
3103 }
3104
3105 return nbot;
3106 }
3107
3108
3109 // Linux uses a growable mapping for the stack, and if the mapping for
3110 // the stack guard pages is not removed when we detach a thread the
3111 // stack cannot grow beyond the pages where the stack guard was
3112 // mapped. If at some point later in the process the stack expands to
3113 // that point, the Linux kernel cannot expand the stack any further
3114 // because the guard pages are in the way, and a segfault occurs.
3115 //
3116 // However, it's essential not to split the stack region by unmapping
3117 // a region (leaving a hole) that's already part of the stack mapping,
3118 // so if the stack mapping has already grown beyond the guard pages at
3119 // the time we create them, we have to truncate the stack mapping.
3120 // So, we need to know the extent of the stack mapping when
3121 // create_stack_guard_pages() is called.
3122
3123 // We only need this for stacks that are growable: at the time of
3124 // writing thread stacks don't use growable mappings (i.e. those
3125 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3126 // only applies to the main thread.
3127
3128 // If the (growable) stack mapping already extends beyond the point
3129 // where we're going to put our guard pages, truncate the mapping at
3130 // that point by munmap()ping it. This ensures that when we later
3131 // munmap() the guard pages we don't leave a hole in the stack
3132 // mapping. This only affects the main/primordial thread
3133
3134 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3135 if (os::is_primordial_thread()) {
3136 // As we manually grow stack up to bottom inside create_attached_thread(),
3137 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3138 // we don't need to do anything special.
3139 // Check it first, before calling heavy function.
3140 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3141 unsigned char vec[1];
3142
3143 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3144 // Fallback to slow path on all errors, including EAGAIN
3145 stack_extent = (uintptr_t) get_stack_commited_bottom(
3146 os::Linux::initial_thread_stack_bottom(),
3147 (size_t)addr - stack_extent);
3148 }
3149
3150 if (stack_extent < (uintptr_t)addr) {
3151 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3152 }
3153 }
3154
3155 return os::commit_memory(addr, size, !ExecMem);
3156 }
3157
3158 // If this is a growable mapping, remove the guard pages entirely by
3159 // munmap()ping them. If not, just call uncommit_memory(). This only
3160 // affects the main/primordial thread, but guard against future OS changes.
3161 // It's safe to always unmap guard pages for primordial thread because we
3162 // always place it right after end of the mapped region.
3163
3164 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3165 uintptr_t stack_extent, stack_base;
3166
3167 if (os::is_primordial_thread()) {
3168 return ::munmap(addr, size) == 0;
3169 }
3170
3171 return os::uncommit_memory(addr, size);
3172 }
3173
3174 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3175 // at 'requested_addr'. If there are existing memory mappings at the same
3176 // location, however, they will be overwritten. If 'fixed' is false,
3177 // 'requested_addr' is only treated as a hint, the return value may or
3178 // may not start from the requested address. Unlike Linux mmap(), this
3179 // function returns NULL to indicate failure.
3180 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3181 char * addr;
3182 int flags;
3183
3184 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3185 if (fixed) {
3186 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3187 flags |= MAP_FIXED;
3188 }
3189
3190 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3191 // touch an uncommitted page. Otherwise, the read/write might
3192 // succeed if we have enough swap space to back the physical page.
3193 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3194 flags, -1, 0);
3195
3196 return addr == MAP_FAILED ? NULL : addr;
3197 }
3198
3199 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3200 // (req_addr != NULL) or with a given alignment.
3201 // - bytes shall be a multiple of alignment.
3202 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3203 // - alignment sets the alignment at which memory shall be allocated.
3204 // It must be a multiple of allocation granularity.
3205 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3206 // req_addr or NULL.
3207 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3208
3209 size_t extra_size = bytes;
3210 if (req_addr == NULL && alignment > 0) {
3211 extra_size += alignment;
3212 }
3213
3214 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3215 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3216 -1, 0);
3217 if (start == MAP_FAILED) {
3218 start = NULL;
3219 } else {
3220 if (req_addr != NULL) {
3221 if (start != req_addr) {
3222 ::munmap(start, extra_size);
3223 start = NULL;
3224 }
3225 } else {
3226 char* const start_aligned = align_up(start, alignment);
3227 char* const end_aligned = start_aligned + bytes;
3228 char* const end = start + extra_size;
3229 if (start_aligned > start) {
3230 ::munmap(start, start_aligned - start);
3231 }
3232 if (end_aligned < end) {
3233 ::munmap(end_aligned, end - end_aligned);
3234 }
3235 start = start_aligned;
3236 }
3237 }
3238 return start;
3239 }
3240
3241 static int anon_munmap(char * addr, size_t size) {
3242 return ::munmap(addr, size) == 0;
3243 }
3244
3245 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3246 size_t alignment_hint) {
3247 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3248 }
3249
3250 bool os::pd_release_memory(char* addr, size_t size) {
3251 return anon_munmap(addr, size);
3252 }
3253
3254 static bool linux_mprotect(char* addr, size_t size, int prot) {
3255 // Linux wants the mprotect address argument to be page aligned.
3256 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3257
3258 // According to SUSv3, mprotect() should only be used with mappings
3259 // established by mmap(), and mmap() always maps whole pages. Unaligned
3260 // 'addr' likely indicates problem in the VM (e.g. trying to change
3261 // protection of malloc'ed or statically allocated memory). Check the
3262 // caller if you hit this assert.
3263 assert(addr == bottom, "sanity check");
3264
3265 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3266 return ::mprotect(bottom, size, prot) == 0;
3267 }
3268
3269 // Set protections specified
3270 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3271 bool is_committed) {
3272 unsigned int p = 0;
3273 switch (prot) {
3274 case MEM_PROT_NONE: p = PROT_NONE; break;
3275 case MEM_PROT_READ: p = PROT_READ; break;
3276 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3277 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3278 default:
3279 ShouldNotReachHere();
3280 }
3281 // is_committed is unused.
3282 return linux_mprotect(addr, bytes, p);
3283 }
3284
3285 bool os::guard_memory(char* addr, size_t size) {
3286 return linux_mprotect(addr, size, PROT_NONE);
3287 }
3288
3289 bool os::unguard_memory(char* addr, size_t size) {
3290 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3291 }
3292
3293 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3294 size_t page_size) {
3295 bool result = false;
3296 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3297 MAP_ANONYMOUS|MAP_PRIVATE,
3298 -1, 0);
3299 if (p != MAP_FAILED) {
3300 void *aligned_p = align_up(p, page_size);
3301
3302 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3303
3304 munmap(p, page_size * 2);
3305 }
3306
3307 if (warn && !result) {
3308 warning("TransparentHugePages is not supported by the operating system.");
3309 }
3310
3311 return result;
3312 }
3313
3314 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3315 bool result = false;
3316 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3317 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3318 -1, 0);
3319
3320 if (p != MAP_FAILED) {
3321 // We don't know if this really is a huge page or not.
3322 FILE *fp = fopen("/proc/self/maps", "r");
3323 if (fp) {
3324 while (!feof(fp)) {
3325 char chars[257];
3326 long x = 0;
3327 if (fgets(chars, sizeof(chars), fp)) {
3328 if (sscanf(chars, "%lx-%*x", &x) == 1
3329 && x == (long)p) {
3330 if (strstr (chars, "hugepage")) {
3331 result = true;
3332 break;
3333 }
3334 }
3335 }
3336 }
3337 fclose(fp);
3338 }
3339 munmap(p, page_size);
3340 }
3341
3342 if (warn && !result) {
3343 warning("HugeTLBFS is not supported by the operating system.");
3344 }
3345
3346 return result;
3347 }
3348
3349 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3350 //
3351 // From the coredump_filter documentation:
3352 //
3353 // - (bit 0) anonymous private memory
3354 // - (bit 1) anonymous shared memory
3355 // - (bit 2) file-backed private memory
3356 // - (bit 3) file-backed shared memory
3357 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3358 // effective only if the bit 2 is cleared)
3359 // - (bit 5) hugetlb private memory
3360 // - (bit 6) hugetlb shared memory
3361 // - (bit 7) dax private memory
3362 // - (bit 8) dax shared memory
3363 //
3364 static void set_coredump_filter(bool largepages, bool dax_shared) {
3365 FILE *f;
3366 long cdm;
3367 bool filter_changed = false;
3368
3369 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3370 return;
3371 }
3372
3373 if (fscanf(f, "%lx", &cdm) != 1) {
3374 fclose(f);
3375 return;
3376 }
3377
3378 rewind(f);
3379
3380 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3381 cdm |= LARGEPAGES_BIT;
3382 filter_changed = true;
3383 }
3384 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3385 cdm |= DAX_SHARED_BIT;
3386 filter_changed = true;
3387 }
3388 if (filter_changed) {
3389 fprintf(f, "%#lx", cdm);
3390 }
3391
3392 fclose(f);
3393 }
3394
3395 // Large page support
3396
3397 static size_t _large_page_size = 0;
3398
3399 size_t os::Linux::find_large_page_size() {
3400 size_t large_page_size = 0;
3401
3402 // large_page_size on Linux is used to round up heap size. x86 uses either
3403 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3404 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3405 // page as large as 256M.
3406 //
3407 // Here we try to figure out page size by parsing /proc/meminfo and looking
3408 // for a line with the following format:
3409 // Hugepagesize: 2048 kB
3410 //
3411 // If we can't determine the value (e.g. /proc is not mounted, or the text
3412 // format has been changed), we'll use the largest page size supported by
3413 // the processor.
3414
3415 #ifndef ZERO
3416 large_page_size =
3417 AARCH64_ONLY(2 * M)
3418 AMD64_ONLY(2 * M)
3419 ARM32_ONLY(2 * M)
3420 IA32_ONLY(4 * M)
3421 IA64_ONLY(256 * M)
3422 PPC_ONLY(4 * M)
3423 S390_ONLY(1 * M)
3424 SPARC_ONLY(4 * M);
3425 #endif // ZERO
3426
3427 FILE *fp = fopen("/proc/meminfo", "r");
3428 if (fp) {
3429 while (!feof(fp)) {
3430 int x = 0;
3431 char buf[16];
3432 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3433 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3434 large_page_size = x * K;
3435 break;
3436 }
3437 } else {
3438 // skip to next line
3439 for (;;) {
3440 int ch = fgetc(fp);
3441 if (ch == EOF || ch == (int)'\n') break;
3442 }
3443 }
3444 }
3445 fclose(fp);
3446 }
3447
3448 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3449 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3450 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3451 proper_unit_for_byte_size(large_page_size));
3452 }
3453
3454 return large_page_size;
3455 }
3456
3457 size_t os::Linux::setup_large_page_size() {
3458 _large_page_size = Linux::find_large_page_size();
3459 const size_t default_page_size = (size_t)Linux::page_size();
3460 if (_large_page_size > default_page_size) {
3461 _page_sizes[0] = _large_page_size;
3462 _page_sizes[1] = default_page_size;
3463 _page_sizes[2] = 0;
3464 }
3465
3466 return _large_page_size;
3467 }
3468
3469 bool os::Linux::setup_large_page_type(size_t page_size) {
3470 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3471 FLAG_IS_DEFAULT(UseSHM) &&
3472 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3473
3474 // The type of large pages has not been specified by the user.
3475
3476 // Try UseHugeTLBFS and then UseSHM.
3477 UseHugeTLBFS = UseSHM = true;
3478
3479 // Don't try UseTransparentHugePages since there are known
3480 // performance issues with it turned on. This might change in the future.
3481 UseTransparentHugePages = false;
3482 }
3483
3484 if (UseTransparentHugePages) {
3485 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3486 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3487 UseHugeTLBFS = false;
3488 UseSHM = false;
3489 return true;
3490 }
3491 UseTransparentHugePages = false;
3492 }
3493
3494 if (UseHugeTLBFS) {
3495 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3496 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3497 UseSHM = false;
3498 return true;
3499 }
3500 UseHugeTLBFS = false;
3501 }
3502
3503 return UseSHM;
3504 }
3505
3506 void os::large_page_init() {
3507 if (!UseLargePages &&
3508 !UseTransparentHugePages &&
3509 !UseHugeTLBFS &&
3510 !UseSHM) {
3511 // Not using large pages.
3512 return;
3513 }
3514
3515 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3516 // The user explicitly turned off large pages.
3517 // Ignore the rest of the large pages flags.
3518 UseTransparentHugePages = false;
3519 UseHugeTLBFS = false;
3520 UseSHM = false;
3521 return;
3522 }
3523
3524 size_t large_page_size = Linux::setup_large_page_size();
3525 UseLargePages = Linux::setup_large_page_type(large_page_size);
3526
3527 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3528 }
3529
3530 #ifndef SHM_HUGETLB
3531 #define SHM_HUGETLB 04000
3532 #endif
3533
3534 #define shm_warning_format(format, ...) \
3535 do { \
3536 if (UseLargePages && \
3537 (!FLAG_IS_DEFAULT(UseLargePages) || \
3538 !FLAG_IS_DEFAULT(UseSHM) || \
3539 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3540 warning(format, __VA_ARGS__); \
3541 } \
3542 } while (0)
3543
3544 #define shm_warning(str) shm_warning_format("%s", str)
3545
3546 #define shm_warning_with_errno(str) \
3547 do { \
3548 int err = errno; \
3549 shm_warning_format(str " (error = %d)", err); \
3550 } while (0)
3551
3552 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3553 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3554
3555 if (!is_aligned(alignment, SHMLBA)) {
3556 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3557 return NULL;
3558 }
3559
3560 // To ensure that we get 'alignment' aligned memory from shmat,
3561 // we pre-reserve aligned virtual memory and then attach to that.
3562
3563 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3564 if (pre_reserved_addr == NULL) {
3565 // Couldn't pre-reserve aligned memory.
3566 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3567 return NULL;
3568 }
3569
3570 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3571 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3572
3573 if ((intptr_t)addr == -1) {
3574 int err = errno;
3575 shm_warning_with_errno("Failed to attach shared memory.");
3576
3577 assert(err != EACCES, "Unexpected error");
3578 assert(err != EIDRM, "Unexpected error");
3579 assert(err != EINVAL, "Unexpected error");
3580
3581 // Since we don't know if the kernel unmapped the pre-reserved memory area
3582 // we can't unmap it, since that would potentially unmap memory that was
3583 // mapped from other threads.
3584 return NULL;
3585 }
3586
3587 return addr;
3588 }
3589
3590 static char* shmat_at_address(int shmid, char* req_addr) {
3591 if (!is_aligned(req_addr, SHMLBA)) {
3592 assert(false, "Requested address needs to be SHMLBA aligned");
3593 return NULL;
3594 }
3595
3596 char* addr = (char*)shmat(shmid, req_addr, 0);
3597
3598 if ((intptr_t)addr == -1) {
3599 shm_warning_with_errno("Failed to attach shared memory.");
3600 return NULL;
3601 }
3602
3603 return addr;
3604 }
3605
3606 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3607 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3608 if (req_addr != NULL) {
3609 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3610 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3611 return shmat_at_address(shmid, req_addr);
3612 }
3613
3614 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3615 // return large page size aligned memory addresses when req_addr == NULL.
3616 // However, if the alignment is larger than the large page size, we have
3617 // to manually ensure that the memory returned is 'alignment' aligned.
3618 if (alignment > os::large_page_size()) {
3619 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3620 return shmat_with_alignment(shmid, bytes, alignment);
3621 } else {
3622 return shmat_at_address(shmid, NULL);
3623 }
3624 }
3625
3626 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3627 char* req_addr, bool exec) {
3628 // "exec" is passed in but not used. Creating the shared image for
3629 // the code cache doesn't have an SHM_X executable permission to check.
3630 assert(UseLargePages && UseSHM, "only for SHM large pages");
3631 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3632 assert(is_aligned(req_addr, alignment), "Unaligned address");
3633
3634 if (!is_aligned(bytes, os::large_page_size())) {
3635 return NULL; // Fallback to small pages.
3636 }
3637
3638 // Create a large shared memory region to attach to based on size.
3639 // Currently, size is the total size of the heap.
3640 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3641 if (shmid == -1) {
3642 // Possible reasons for shmget failure:
3643 // 1. shmmax is too small for Java heap.
3644 // > check shmmax value: cat /proc/sys/kernel/shmmax
3645 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3646 // 2. not enough large page memory.
3647 // > check available large pages: cat /proc/meminfo
3648 // > increase amount of large pages:
3649 // echo new_value > /proc/sys/vm/nr_hugepages
3650 // Note 1: different Linux may use different name for this property,
3651 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3652 // Note 2: it's possible there's enough physical memory available but
3653 // they are so fragmented after a long run that they can't
3654 // coalesce into large pages. Try to reserve large pages when
3655 // the system is still "fresh".
3656 shm_warning_with_errno("Failed to reserve shared memory.");
3657 return NULL;
3658 }
3659
3660 // Attach to the region.
3661 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3662
3663 // Remove shmid. If shmat() is successful, the actual shared memory segment
3664 // will be deleted when it's detached by shmdt() or when the process
3665 // terminates. If shmat() is not successful this will remove the shared
3666 // segment immediately.
3667 shmctl(shmid, IPC_RMID, NULL);
3668
3669 return addr;
3670 }
3671
3672 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3673 int error) {
3674 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3675
3676 bool warn_on_failure = UseLargePages &&
3677 (!FLAG_IS_DEFAULT(UseLargePages) ||
3678 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3679 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3680
3681 if (warn_on_failure) {
3682 char msg[128];
3683 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3684 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3685 warning("%s", msg);
3686 }
3687 }
3688
3689 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3690 char* req_addr,
3691 bool exec) {
3692 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3693 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3694 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3695
3696 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3697 char* addr = (char*)::mmap(req_addr, bytes, prot,
3698 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3699 -1, 0);
3700
3701 if (addr == MAP_FAILED) {
3702 warn_on_large_pages_failure(req_addr, bytes, errno);
3703 return NULL;
3704 }
3705
3706 assert(is_aligned(addr, os::large_page_size()), "Must be");
3707
3708 return addr;
3709 }
3710
3711 // Reserve memory using mmap(MAP_HUGETLB).
3712 // - bytes shall be a multiple of alignment.
3713 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3714 // - alignment sets the alignment at which memory shall be allocated.
3715 // It must be a multiple of allocation granularity.
3716 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3717 // req_addr or NULL.
3718 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3719 size_t alignment,
3720 char* req_addr,
3721 bool exec) {
3722 size_t large_page_size = os::large_page_size();
3723 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3724
3725 assert(is_aligned(req_addr, alignment), "Must be");
3726 assert(is_aligned(bytes, alignment), "Must be");
3727
3728 // First reserve - but not commit - the address range in small pages.
3729 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3730
3731 if (start == NULL) {
3732 return NULL;
3733 }
3734
3735 assert(is_aligned(start, alignment), "Must be");
3736
3737 char* end = start + bytes;
3738
3739 // Find the regions of the allocated chunk that can be promoted to large pages.
3740 char* lp_start = align_up(start, large_page_size);
3741 char* lp_end = align_down(end, large_page_size);
3742
3743 size_t lp_bytes = lp_end - lp_start;
3744
3745 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3746
3747 if (lp_bytes == 0) {
3748 // The mapped region doesn't even span the start and the end of a large page.
3749 // Fall back to allocate a non-special area.
3750 ::munmap(start, end - start);
3751 return NULL;
3752 }
3753
3754 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3755
3756 void* result;
3757
3758 // Commit small-paged leading area.
3759 if (start != lp_start) {
3760 result = ::mmap(start, lp_start - start, prot,
3761 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3762 -1, 0);
3763 if (result == MAP_FAILED) {
3764 ::munmap(lp_start, end - lp_start);
3765 return NULL;
3766 }
3767 }
3768
3769 // Commit large-paged area.
3770 result = ::mmap(lp_start, lp_bytes, prot,
3771 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3772 -1, 0);
3773 if (result == MAP_FAILED) {
3774 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3775 // If the mmap above fails, the large pages region will be unmapped and we
3776 // have regions before and after with small pages. Release these regions.
3777 //
3778 // | mapped | unmapped | mapped |
3779 // ^ ^ ^ ^
3780 // start lp_start lp_end end
3781 //
3782 ::munmap(start, lp_start - start);
3783 ::munmap(lp_end, end - lp_end);
3784 return NULL;
3785 }
3786
3787 // Commit small-paged trailing area.
3788 if (lp_end != end) {
3789 result = ::mmap(lp_end, end - lp_end, prot,
3790 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3791 -1, 0);
3792 if (result == MAP_FAILED) {
3793 ::munmap(start, lp_end - start);
3794 return NULL;
3795 }
3796 }
3797
3798 return start;
3799 }
3800
3801 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3802 size_t alignment,
3803 char* req_addr,
3804 bool exec) {
3805 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3806 assert(is_aligned(req_addr, alignment), "Must be");
3807 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3808 assert(is_power_of_2(os::large_page_size()), "Must be");
3809 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3810
3811 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3812 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3813 } else {
3814 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3815 }
3816 }
3817
3818 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3819 char* req_addr, bool exec) {
3820 assert(UseLargePages, "only for large pages");
3821
3822 char* addr;
3823 if (UseSHM) {
3824 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3825 } else {
3826 assert(UseHugeTLBFS, "must be");
3827 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3828 }
3829
3830 if (addr != NULL) {
3831 if (UseNUMAInterleaving) {
3832 numa_make_global(addr, bytes);
3833 }
3834
3835 // The memory is committed
3836 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3837 }
3838
3839 return addr;
3840 }
3841
3842 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3843 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3844 return shmdt(base) == 0;
3845 }
3846
3847 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3848 return pd_release_memory(base, bytes);
3849 }
3850
3851 bool os::release_memory_special(char* base, size_t bytes) {
3852 bool res;
3853 if (MemTracker::tracking_level() > NMT_minimal) {
3854 Tracker tkr(Tracker::release);
3855 res = os::Linux::release_memory_special_impl(base, bytes);
3856 if (res) {
3857 tkr.record((address)base, bytes);
3858 }
3859
3860 } else {
3861 res = os::Linux::release_memory_special_impl(base, bytes);
3862 }
3863 return res;
3864 }
3865
3866 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3867 assert(UseLargePages, "only for large pages");
3868 bool res;
3869
3870 if (UseSHM) {
3871 res = os::Linux::release_memory_special_shm(base, bytes);
3872 } else {
3873 assert(UseHugeTLBFS, "must be");
3874 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3875 }
3876 return res;
3877 }
3878
3879 size_t os::large_page_size() {
3880 return _large_page_size;
3881 }
3882
3883 // With SysV SHM the entire memory region must be allocated as shared
3884 // memory.
3885 // HugeTLBFS allows application to commit large page memory on demand.
3886 // However, when committing memory with HugeTLBFS fails, the region
3887 // that was supposed to be committed will lose the old reservation
3888 // and allow other threads to steal that memory region. Because of this
3889 // behavior we can't commit HugeTLBFS memory.
3890 bool os::can_commit_large_page_memory() {
3891 return UseTransparentHugePages;
3892 }
3893
3894 bool os::can_execute_large_page_memory() {
3895 return UseTransparentHugePages || UseHugeTLBFS;
3896 }
3897
3898 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3899 assert(file_desc >= 0, "file_desc is not valid");
3900 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3901 if (result != NULL) {
3902 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3903 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3904 }
3905 }
3906 return result;
3907 }
3908
3909 // Reserve memory at an arbitrary address, only if that area is
3910 // available (and not reserved for something else).
3911
3912 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3913 const int max_tries = 10;
3914 char* base[max_tries];
3915 size_t size[max_tries];
3916 const size_t gap = 0x000000;
3917
3918 // Assert only that the size is a multiple of the page size, since
3919 // that's all that mmap requires, and since that's all we really know
3920 // about at this low abstraction level. If we need higher alignment,
3921 // we can either pass an alignment to this method or verify alignment
3922 // in one of the methods further up the call chain. See bug 5044738.
3923 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3924
3925 // Repeatedly allocate blocks until the block is allocated at the
3926 // right spot.
3927
3928 // Linux mmap allows caller to pass an address as hint; give it a try first,
3929 // if kernel honors the hint then we can return immediately.
3930 char * addr = anon_mmap(requested_addr, bytes, false);
3931 if (addr == requested_addr) {
3932 return requested_addr;
3933 }
3934
3935 if (addr != NULL) {
3936 // mmap() is successful but it fails to reserve at the requested address
3937 anon_munmap(addr, bytes);
3938 }
3939
3940 int i;
3941 for (i = 0; i < max_tries; ++i) {
3942 base[i] = reserve_memory(bytes);
3943
3944 if (base[i] != NULL) {
3945 // Is this the block we wanted?
3946 if (base[i] == requested_addr) {
3947 size[i] = bytes;
3948 break;
3949 }
3950
3951 // Does this overlap the block we wanted? Give back the overlapped
3952 // parts and try again.
3953
3954 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3955 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3956 unmap_memory(base[i], top_overlap);
3957 base[i] += top_overlap;
3958 size[i] = bytes - top_overlap;
3959 } else {
3960 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3961 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3962 unmap_memory(requested_addr, bottom_overlap);
3963 size[i] = bytes - bottom_overlap;
3964 } else {
3965 size[i] = bytes;
3966 }
3967 }
3968 }
3969 }
3970
3971 // Give back the unused reserved pieces.
3972
3973 for (int j = 0; j < i; ++j) {
3974 if (base[j] != NULL) {
3975 unmap_memory(base[j], size[j]);
3976 }
3977 }
3978
3979 if (i < max_tries) {
3980 return requested_addr;
3981 } else {
3982 return NULL;
3983 }
3984 }
3985
3986 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3987 return ::read(fd, buf, nBytes);
3988 }
3989
3990 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3991 return ::pread(fd, buf, nBytes, offset);
3992 }
3993
3994 // Short sleep, direct OS call.
3995 //
3996 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3997 // sched_yield(2) will actually give up the CPU:
3998 //
3999 // * Alone on this pariticular CPU, keeps running.
4000 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4001 // (pre 2.6.39).
4002 //
4003 // So calling this with 0 is an alternative.
4004 //
4005 void os::naked_short_sleep(jlong ms) {
4006 struct timespec req;
4007
4008 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4009 req.tv_sec = 0;
4010 if (ms > 0) {
4011 req.tv_nsec = (ms % 1000) * 1000000;
4012 } else {
4013 req.tv_nsec = 1;
4014 }
4015
4016 nanosleep(&req, NULL);
4017
4018 return;
4019 }
4020
4021 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4022 void os::infinite_sleep() {
4023 while (true) { // sleep forever ...
4024 ::sleep(100); // ... 100 seconds at a time
4025 }
4026 }
4027
4028 // Used to convert frequent JVM_Yield() to nops
4029 bool os::dont_yield() {
4030 return DontYieldALot;
4031 }
4032
4033 void os::naked_yield() {
4034 sched_yield();
4035 }
4036
4037 ////////////////////////////////////////////////////////////////////////////////
4038 // thread priority support
4039
4040 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4041 // only supports dynamic priority, static priority must be zero. For real-time
4042 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4043 // However, for large multi-threaded applications, SCHED_RR is not only slower
4044 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4045 // of 5 runs - Sep 2005).
4046 //
4047 // The following code actually changes the niceness of kernel-thread/LWP. It
4048 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4049 // not the entire user process, and user level threads are 1:1 mapped to kernel
4050 // threads. It has always been the case, but could change in the future. For
4051 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4052 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4053
4054 int os::java_to_os_priority[CriticalPriority + 1] = {
4055 19, // 0 Entry should never be used
4056
4057 4, // 1 MinPriority
4058 3, // 2
4059 2, // 3
4060
4061 1, // 4
4062 0, // 5 NormPriority
4063 -1, // 6
4064
4065 -2, // 7
4066 -3, // 8
4067 -4, // 9 NearMaxPriority
4068
4069 -5, // 10 MaxPriority
4070
4071 -5 // 11 CriticalPriority
4072 };
4073
4074 static int prio_init() {
4075 if (ThreadPriorityPolicy == 1) {
4076 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4077 // if effective uid is not root. Perhaps, a more elegant way of doing
4078 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4079 if (geteuid() != 0) {
4080 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4081 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4082 }
4083 ThreadPriorityPolicy = 0;
4084 }
4085 }
4086 if (UseCriticalJavaThreadPriority) {
4087 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4088 }
4089 return 0;
4090 }
4091
4092 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4093 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4094
4095 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4096 return (ret == 0) ? OS_OK : OS_ERR;
4097 }
4098
4099 OSReturn os::get_native_priority(const Thread* const thread,
4100 int *priority_ptr) {
4101 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4102 *priority_ptr = java_to_os_priority[NormPriority];
4103 return OS_OK;
4104 }
4105
4106 errno = 0;
4107 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4108 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4109 }
4110
4111 // Hint to the underlying OS that a task switch would not be good.
4112 // Void return because it's a hint and can fail.
4113 void os::hint_no_preempt() {}
4114
4115 ////////////////////////////////////////////////////////////////////////////////
4116 // suspend/resume support
4117
4118 // The low-level signal-based suspend/resume support is a remnant from the
4119 // old VM-suspension that used to be for java-suspension, safepoints etc,
4120 // within hotspot. Currently used by JFR's OSThreadSampler
4121 //
4122 // The remaining code is greatly simplified from the more general suspension
4123 // code that used to be used.
4124 //
4125 // The protocol is quite simple:
4126 // - suspend:
4127 // - sends a signal to the target thread
4128 // - polls the suspend state of the osthread using a yield loop
4129 // - target thread signal handler (SR_handler) sets suspend state
4130 // and blocks in sigsuspend until continued
4131 // - resume:
4132 // - sets target osthread state to continue
4133 // - sends signal to end the sigsuspend loop in the SR_handler
4134 //
4135 // Note that the SR_lock plays no role in this suspend/resume protocol,
4136 // but is checked for NULL in SR_handler as a thread termination indicator.
4137 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4138 //
4139 // Note that resume_clear_context() and suspend_save_context() are needed
4140 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4141 // which in part is used by:
4142 // - Forte Analyzer: AsyncGetCallTrace()
4143 // - StackBanging: get_frame_at_stack_banging_point()
4144
4145 static void resume_clear_context(OSThread *osthread) {
4146 osthread->set_ucontext(NULL);
4147 osthread->set_siginfo(NULL);
4148 }
4149
4150 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4151 ucontext_t* context) {
4152 osthread->set_ucontext(context);
4153 osthread->set_siginfo(siginfo);
4154 }
4155
4156 // Handler function invoked when a thread's execution is suspended or
4157 // resumed. We have to be careful that only async-safe functions are
4158 // called here (Note: most pthread functions are not async safe and
4159 // should be avoided.)
4160 //
4161 // Note: sigwait() is a more natural fit than sigsuspend() from an
4162 // interface point of view, but sigwait() prevents the signal hander
4163 // from being run. libpthread would get very confused by not having
4164 // its signal handlers run and prevents sigwait()'s use with the
4165 // mutex granting granting signal.
4166 //
4167 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4168 //
4169 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4170 // Save and restore errno to avoid confusing native code with EINTR
4171 // after sigsuspend.
4172 int old_errno = errno;
4173
4174 Thread* thread = Thread::current_or_null_safe();
4175 assert(thread != NULL, "Missing current thread in SR_handler");
4176
4177 // On some systems we have seen signal delivery get "stuck" until the signal
4178 // mask is changed as part of thread termination. Check that the current thread
4179 // has not already terminated (via SR_lock()) - else the following assertion
4180 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4181 // destructor has completed.
4182
4183 if (thread->SR_lock() == NULL) {
4184 return;
4185 }
4186
4187 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4188
4189 OSThread* osthread = thread->osthread();
4190
4191 os::SuspendResume::State current = osthread->sr.state();
4192 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4193 suspend_save_context(osthread, siginfo, context);
4194
4195 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4196 os::SuspendResume::State state = osthread->sr.suspended();
4197 if (state == os::SuspendResume::SR_SUSPENDED) {
4198 sigset_t suspend_set; // signals for sigsuspend()
4199 sigemptyset(&suspend_set);
4200 // get current set of blocked signals and unblock resume signal
4201 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4202 sigdelset(&suspend_set, SR_signum);
4203
4204 sr_semaphore.signal();
4205 // wait here until we are resumed
4206 while (1) {
4207 sigsuspend(&suspend_set);
4208
4209 os::SuspendResume::State result = osthread->sr.running();
4210 if (result == os::SuspendResume::SR_RUNNING) {
4211 sr_semaphore.signal();
4212 break;
4213 }
4214 }
4215
4216 } else if (state == os::SuspendResume::SR_RUNNING) {
4217 // request was cancelled, continue
4218 } else {
4219 ShouldNotReachHere();
4220 }
4221
4222 resume_clear_context(osthread);
4223 } else if (current == os::SuspendResume::SR_RUNNING) {
4224 // request was cancelled, continue
4225 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4226 // ignore
4227 } else {
4228 // ignore
4229 }
4230
4231 errno = old_errno;
4232 }
4233
4234 static int SR_initialize() {
4235 struct sigaction act;
4236 char *s;
4237
4238 // Get signal number to use for suspend/resume
4239 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4240 int sig = ::strtol(s, 0, 10);
4241 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4242 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4243 SR_signum = sig;
4244 } else {
4245 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4246 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4247 }
4248 }
4249
4250 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4251 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4252
4253 sigemptyset(&SR_sigset);
4254 sigaddset(&SR_sigset, SR_signum);
4255
4256 // Set up signal handler for suspend/resume
4257 act.sa_flags = SA_RESTART|SA_SIGINFO;
4258 act.sa_handler = (void (*)(int)) SR_handler;
4259
4260 // SR_signum is blocked by default.
4261 // 4528190 - We also need to block pthread restart signal (32 on all
4262 // supported Linux platforms). Note that LinuxThreads need to block
4263 // this signal for all threads to work properly. So we don't have
4264 // to use hard-coded signal number when setting up the mask.
4265 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4266
4267 if (sigaction(SR_signum, &act, 0) == -1) {
4268 return -1;
4269 }
4270
4271 // Save signal flag
4272 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4273 return 0;
4274 }
4275
4276 static int sr_notify(OSThread* osthread) {
4277 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4278 assert_status(status == 0, status, "pthread_kill");
4279 return status;
4280 }
4281
4282 // "Randomly" selected value for how long we want to spin
4283 // before bailing out on suspending a thread, also how often
4284 // we send a signal to a thread we want to resume
4285 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4286 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4287
4288 // returns true on success and false on error - really an error is fatal
4289 // but this seems the normal response to library errors
4290 static bool do_suspend(OSThread* osthread) {
4291 assert(osthread->sr.is_running(), "thread should be running");
4292 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4293
4294 // mark as suspended and send signal
4295 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4296 // failed to switch, state wasn't running?
4297 ShouldNotReachHere();
4298 return false;
4299 }
4300
4301 if (sr_notify(osthread) != 0) {
4302 ShouldNotReachHere();
4303 }
4304
4305 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4306 while (true) {
4307 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4308 break;
4309 } else {
4310 // timeout
4311 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4312 if (cancelled == os::SuspendResume::SR_RUNNING) {
4313 return false;
4314 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4315 // make sure that we consume the signal on the semaphore as well
4316 sr_semaphore.wait();
4317 break;
4318 } else {
4319 ShouldNotReachHere();
4320 return false;
4321 }
4322 }
4323 }
4324
4325 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4326 return true;
4327 }
4328
4329 static void do_resume(OSThread* osthread) {
4330 assert(osthread->sr.is_suspended(), "thread should be suspended");
4331 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4332
4333 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4334 // failed to switch to WAKEUP_REQUEST
4335 ShouldNotReachHere();
4336 return;
4337 }
4338
4339 while (true) {
4340 if (sr_notify(osthread) == 0) {
4341 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4342 if (osthread->sr.is_running()) {
4343 return;
4344 }
4345 }
4346 } else {
4347 ShouldNotReachHere();
4348 }
4349 }
4350
4351 guarantee(osthread->sr.is_running(), "Must be running!");
4352 }
4353
4354 ///////////////////////////////////////////////////////////////////////////////////
4355 // signal handling (except suspend/resume)
4356
4357 // This routine may be used by user applications as a "hook" to catch signals.
4358 // The user-defined signal handler must pass unrecognized signals to this
4359 // routine, and if it returns true (non-zero), then the signal handler must
4360 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4361 // routine will never retun false (zero), but instead will execute a VM panic
4362 // routine kill the process.
4363 //
4364 // If this routine returns false, it is OK to call it again. This allows
4365 // the user-defined signal handler to perform checks either before or after
4366 // the VM performs its own checks. Naturally, the user code would be making
4367 // a serious error if it tried to handle an exception (such as a null check
4368 // or breakpoint) that the VM was generating for its own correct operation.
4369 //
4370 // This routine may recognize any of the following kinds of signals:
4371 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4372 // It should be consulted by handlers for any of those signals.
4373 //
4374 // The caller of this routine must pass in the three arguments supplied
4375 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4376 // field of the structure passed to sigaction(). This routine assumes that
4377 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4378 //
4379 // Note that the VM will print warnings if it detects conflicting signal
4380 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4381 //
4382 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4383 siginfo_t* siginfo,
4384 void* ucontext,
4385 int abort_if_unrecognized);
4386
4387 void signalHandler(int sig, siginfo_t* info, void* uc) {
4388 assert(info != NULL && uc != NULL, "it must be old kernel");
4389 int orig_errno = errno; // Preserve errno value over signal handler.
4390 JVM_handle_linux_signal(sig, info, uc, true);
4391 errno = orig_errno;
4392 }
4393
4394
4395 // This boolean allows users to forward their own non-matching signals
4396 // to JVM_handle_linux_signal, harmlessly.
4397 bool os::Linux::signal_handlers_are_installed = false;
4398
4399 // For signal-chaining
4400 struct sigaction sigact[NSIG];
4401 uint64_t sigs = 0;
4402 #if (64 < NSIG-1)
4403 #error "Not all signals can be encoded in sigs. Adapt its type!"
4404 #endif
4405 bool os::Linux::libjsig_is_loaded = false;
4406 typedef struct sigaction *(*get_signal_t)(int);
4407 get_signal_t os::Linux::get_signal_action = NULL;
4408
4409 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4410 struct sigaction *actp = NULL;
4411
4412 if (libjsig_is_loaded) {
4413 // Retrieve the old signal handler from libjsig
4414 actp = (*get_signal_action)(sig);
4415 }
4416 if (actp == NULL) {
4417 // Retrieve the preinstalled signal handler from jvm
4418 actp = get_preinstalled_handler(sig);
4419 }
4420
4421 return actp;
4422 }
4423
4424 static bool call_chained_handler(struct sigaction *actp, int sig,
4425 siginfo_t *siginfo, void *context) {
4426 // Call the old signal handler
4427 if (actp->sa_handler == SIG_DFL) {
4428 // It's more reasonable to let jvm treat it as an unexpected exception
4429 // instead of taking the default action.
4430 return false;
4431 } else if (actp->sa_handler != SIG_IGN) {
4432 if ((actp->sa_flags & SA_NODEFER) == 0) {
4433 // automaticlly block the signal
4434 sigaddset(&(actp->sa_mask), sig);
4435 }
4436
4437 sa_handler_t hand = NULL;
4438 sa_sigaction_t sa = NULL;
4439 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4440 // retrieve the chained handler
4441 if (siginfo_flag_set) {
4442 sa = actp->sa_sigaction;
4443 } else {
4444 hand = actp->sa_handler;
4445 }
4446
4447 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4448 actp->sa_handler = SIG_DFL;
4449 }
4450
4451 // try to honor the signal mask
4452 sigset_t oset;
4453 sigemptyset(&oset);
4454 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4455
4456 // call into the chained handler
4457 if (siginfo_flag_set) {
4458 (*sa)(sig, siginfo, context);
4459 } else {
4460 (*hand)(sig);
4461 }
4462
4463 // restore the signal mask
4464 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4465 }
4466 // Tell jvm's signal handler the signal is taken care of.
4467 return true;
4468 }
4469
4470 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4471 bool chained = false;
4472 // signal-chaining
4473 if (UseSignalChaining) {
4474 struct sigaction *actp = get_chained_signal_action(sig);
4475 if (actp != NULL) {
4476 chained = call_chained_handler(actp, sig, siginfo, context);
4477 }
4478 }
4479 return chained;
4480 }
4481
4482 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4483 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4484 return &sigact[sig];
4485 }
4486 return NULL;
4487 }
4488
4489 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4490 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4491 sigact[sig] = oldAct;
4492 sigs |= (uint64_t)1 << (sig-1);
4493 }
4494
4495 // for diagnostic
4496 int sigflags[NSIG];
4497
4498 int os::Linux::get_our_sigflags(int sig) {
4499 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4500 return sigflags[sig];
4501 }
4502
4503 void os::Linux::set_our_sigflags(int sig, int flags) {
4504 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4505 if (sig > 0 && sig < NSIG) {
4506 sigflags[sig] = flags;
4507 }
4508 }
4509
4510 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4511 // Check for overwrite.
4512 struct sigaction oldAct;
4513 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4514
4515 void* oldhand = oldAct.sa_sigaction
4516 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4517 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4518 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4519 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4520 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4521 if (AllowUserSignalHandlers || !set_installed) {
4522 // Do not overwrite; user takes responsibility to forward to us.
4523 return;
4524 } else if (UseSignalChaining) {
4525 // save the old handler in jvm
4526 save_preinstalled_handler(sig, oldAct);
4527 // libjsig also interposes the sigaction() call below and saves the
4528 // old sigaction on it own.
4529 } else {
4530 fatal("Encountered unexpected pre-existing sigaction handler "
4531 "%#lx for signal %d.", (long)oldhand, sig);
4532 }
4533 }
4534
4535 struct sigaction sigAct;
4536 sigfillset(&(sigAct.sa_mask));
4537 sigAct.sa_handler = SIG_DFL;
4538 if (!set_installed) {
4539 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4540 } else {
4541 sigAct.sa_sigaction = signalHandler;
4542 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4543 }
4544 // Save flags, which are set by ours
4545 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4546 sigflags[sig] = sigAct.sa_flags;
4547
4548 int ret = sigaction(sig, &sigAct, &oldAct);
4549 assert(ret == 0, "check");
4550
4551 void* oldhand2 = oldAct.sa_sigaction
4552 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4553 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4554 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4555 }
4556
4557 // install signal handlers for signals that HotSpot needs to
4558 // handle in order to support Java-level exception handling.
4559
4560 void os::Linux::install_signal_handlers() {
4561 if (!signal_handlers_are_installed) {
4562 signal_handlers_are_installed = true;
4563
4564 // signal-chaining
4565 typedef void (*signal_setting_t)();
4566 signal_setting_t begin_signal_setting = NULL;
4567 signal_setting_t end_signal_setting = NULL;
4568 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4569 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4570 if (begin_signal_setting != NULL) {
4571 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4572 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4573 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4574 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4575 libjsig_is_loaded = true;
4576 assert(UseSignalChaining, "should enable signal-chaining");
4577 }
4578 if (libjsig_is_loaded) {
4579 // Tell libjsig jvm is setting signal handlers
4580 (*begin_signal_setting)();
4581 }
4582
4583 set_signal_handler(SIGSEGV, true);
4584 set_signal_handler(SIGPIPE, true);
4585 set_signal_handler(SIGBUS, true);
4586 set_signal_handler(SIGILL, true);
4587 set_signal_handler(SIGFPE, true);
4588 #if defined(PPC64)
4589 set_signal_handler(SIGTRAP, true);
4590 #endif
4591 set_signal_handler(SIGXFSZ, true);
4592
4593 if (libjsig_is_loaded) {
4594 // Tell libjsig jvm finishes setting signal handlers
4595 (*end_signal_setting)();
4596 }
4597
4598 // We don't activate signal checker if libjsig is in place, we trust ourselves
4599 // and if UserSignalHandler is installed all bets are off.
4600 // Log that signal checking is off only if -verbose:jni is specified.
4601 if (CheckJNICalls) {
4602 if (libjsig_is_loaded) {
4603 if (PrintJNIResolving) {
4604 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4605 }
4606 check_signals = false;
4607 }
4608 if (AllowUserSignalHandlers) {
4609 if (PrintJNIResolving) {
4610 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4611 }
4612 check_signals = false;
4613 }
4614 }
4615 }
4616 }
4617
4618 // This is the fastest way to get thread cpu time on Linux.
4619 // Returns cpu time (user+sys) for any thread, not only for current.
4620 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4621 // It might work on 2.6.10+ with a special kernel/glibc patch.
4622 // For reference, please, see IEEE Std 1003.1-2004:
4623 // http://www.unix.org/single_unix_specification
4624
4625 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4626 struct timespec tp;
4627 int rc = os::Linux::clock_gettime(clockid, &tp);
4628 assert(rc == 0, "clock_gettime is expected to return 0 code");
4629
4630 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4631 }
4632
4633 void os::Linux::initialize_os_info() {
4634 assert(_os_version == 0, "OS info already initialized");
4635
4636 struct utsname _uname;
4637
4638 uint32_t major;
4639 uint32_t minor;
4640 uint32_t fix;
4641
4642 int rc;
4643
4644 // Kernel version is unknown if
4645 // verification below fails.
4646 _os_version = 0x01000000;
4647
4648 rc = uname(&_uname);
4649 if (rc != -1) {
4650
4651 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4652 if (rc == 3) {
4653
4654 if (major < 256 && minor < 256 && fix < 256) {
4655 // Kernel version format is as expected,
4656 // set it overriding unknown state.
4657 _os_version = (major << 16) |
4658 (minor << 8 ) |
4659 (fix << 0 ) ;
4660 }
4661 }
4662 }
4663 }
4664
4665 uint32_t os::Linux::os_version() {
4666 assert(_os_version != 0, "not initialized");
4667 return _os_version & 0x00FFFFFF;
4668 }
4669
4670 bool os::Linux::os_version_is_known() {
4671 assert(_os_version != 0, "not initialized");
4672 return _os_version & 0x01000000 ? false : true;
4673 }
4674
4675 /////
4676 // glibc on Linux platform uses non-documented flag
4677 // to indicate, that some special sort of signal
4678 // trampoline is used.
4679 // We will never set this flag, and we should
4680 // ignore this flag in our diagnostic
4681 #ifdef SIGNIFICANT_SIGNAL_MASK
4682 #undef SIGNIFICANT_SIGNAL_MASK
4683 #endif
4684 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4685
4686 static const char* get_signal_handler_name(address handler,
4687 char* buf, int buflen) {
4688 int offset = 0;
4689 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4690 if (found) {
4691 // skip directory names
4692 const char *p1, *p2;
4693 p1 = buf;
4694 size_t len = strlen(os::file_separator());
4695 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4696 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4697 } else {
4698 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4699 }
4700 return buf;
4701 }
4702
4703 static void print_signal_handler(outputStream* st, int sig,
4704 char* buf, size_t buflen) {
4705 struct sigaction sa;
4706
4707 sigaction(sig, NULL, &sa);
4708
4709 // See comment for SIGNIFICANT_SIGNAL_MASK define
4710 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4711
4712 st->print("%s: ", os::exception_name(sig, buf, buflen));
4713
4714 address handler = (sa.sa_flags & SA_SIGINFO)
4715 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4716 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4717
4718 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4719 st->print("SIG_DFL");
4720 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4721 st->print("SIG_IGN");
4722 } else {
4723 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4724 }
4725
4726 st->print(", sa_mask[0]=");
4727 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4728
4729 address rh = VMError::get_resetted_sighandler(sig);
4730 // May be, handler was resetted by VMError?
4731 if (rh != NULL) {
4732 handler = rh;
4733 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4734 }
4735
4736 st->print(", sa_flags=");
4737 os::Posix::print_sa_flags(st, sa.sa_flags);
4738
4739 // Check: is it our handler?
4740 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4741 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4742 // It is our signal handler
4743 // check for flags, reset system-used one!
4744 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4745 st->print(
4746 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4747 os::Linux::get_our_sigflags(sig));
4748 }
4749 }
4750 st->cr();
4751 }
4752
4753
4754 #define DO_SIGNAL_CHECK(sig) \
4755 do { \
4756 if (!sigismember(&check_signal_done, sig)) { \
4757 os::Linux::check_signal_handler(sig); \
4758 } \
4759 } while (0)
4760
4761 // This method is a periodic task to check for misbehaving JNI applications
4762 // under CheckJNI, we can add any periodic checks here
4763
4764 void os::run_periodic_checks() {
4765 if (check_signals == false) return;
4766
4767 // SEGV and BUS if overridden could potentially prevent
4768 // generation of hs*.log in the event of a crash, debugging
4769 // such a case can be very challenging, so we absolutely
4770 // check the following for a good measure:
4771 DO_SIGNAL_CHECK(SIGSEGV);
4772 DO_SIGNAL_CHECK(SIGILL);
4773 DO_SIGNAL_CHECK(SIGFPE);
4774 DO_SIGNAL_CHECK(SIGBUS);
4775 DO_SIGNAL_CHECK(SIGPIPE);
4776 DO_SIGNAL_CHECK(SIGXFSZ);
4777 #if defined(PPC64)
4778 DO_SIGNAL_CHECK(SIGTRAP);
4779 #endif
4780
4781 // ReduceSignalUsage allows the user to override these handlers
4782 // see comments at the very top and jvm_md.h
4783 if (!ReduceSignalUsage) {
4784 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4785 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4786 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4787 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4788 }
4789
4790 DO_SIGNAL_CHECK(SR_signum);
4791 }
4792
4793 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4794
4795 static os_sigaction_t os_sigaction = NULL;
4796
4797 void os::Linux::check_signal_handler(int sig) {
4798 char buf[O_BUFLEN];
4799 address jvmHandler = NULL;
4800
4801
4802 struct sigaction act;
4803 if (os_sigaction == NULL) {
4804 // only trust the default sigaction, in case it has been interposed
4805 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4806 if (os_sigaction == NULL) return;
4807 }
4808
4809 os_sigaction(sig, (struct sigaction*)NULL, &act);
4810
4811
4812 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4813
4814 address thisHandler = (act.sa_flags & SA_SIGINFO)
4815 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4816 : CAST_FROM_FN_PTR(address, act.sa_handler);
4817
4818
4819 switch (sig) {
4820 case SIGSEGV:
4821 case SIGBUS:
4822 case SIGFPE:
4823 case SIGPIPE:
4824 case SIGILL:
4825 case SIGXFSZ:
4826 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4827 break;
4828
4829 case SHUTDOWN1_SIGNAL:
4830 case SHUTDOWN2_SIGNAL:
4831 case SHUTDOWN3_SIGNAL:
4832 case BREAK_SIGNAL:
4833 jvmHandler = (address)user_handler();
4834 break;
4835
4836 default:
4837 if (sig == SR_signum) {
4838 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4839 } else {
4840 return;
4841 }
4842 break;
4843 }
4844
4845 if (thisHandler != jvmHandler) {
4846 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4847 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4848 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4849 // No need to check this sig any longer
4850 sigaddset(&check_signal_done, sig);
4851 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4852 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4853 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4854 exception_name(sig, buf, O_BUFLEN));
4855 }
4856 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4857 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4858 tty->print("expected:");
4859 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4860 tty->cr();
4861 tty->print(" found:");
4862 os::Posix::print_sa_flags(tty, act.sa_flags);
4863 tty->cr();
4864 // No need to check this sig any longer
4865 sigaddset(&check_signal_done, sig);
4866 }
4867
4868 // Dump all the signal
4869 if (sigismember(&check_signal_done, sig)) {
4870 print_signal_handlers(tty, buf, O_BUFLEN);
4871 }
4872 }
4873
4874 extern void report_error(char* file_name, int line_no, char* title,
4875 char* format, ...);
4876
4877 // this is called _before_ most of the global arguments have been parsed
4878 void os::init(void) {
4879 char dummy; // used to get a guess on initial stack address
4880
4881 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4882
4883 init_random(1234567);
4884
4885 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4886 if (Linux::page_size() == -1) {
4887 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4888 os::strerror(errno));
4889 }
4890 init_page_sizes((size_t) Linux::page_size());
4891
4892 Linux::initialize_system_info();
4893
4894 Linux::initialize_os_info();
4895
4896 // _main_thread points to the thread that created/loaded the JVM.
4897 Linux::_main_thread = pthread_self();
4898
4899 Linux::clock_init();
4900 initial_time_count = javaTimeNanos();
4901
4902 // retrieve entry point for pthread_setname_np
4903 Linux::_pthread_setname_np =
4904 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4905
4906 os::Posix::init();
4907 }
4908
4909 // To install functions for atexit system call
4910 extern "C" {
4911 static void perfMemory_exit_helper() {
4912 perfMemory_exit();
4913 }
4914 }
4915
4916 void os::pd_init_container_support() {
4917 OSContainer::init();
4918 }
4919
4920 // this is called _after_ the global arguments have been parsed
4921 jint os::init_2(void) {
4922
4923 os::Posix::init_2();
4924
4925 Linux::fast_thread_clock_init();
4926
4927 // initialize suspend/resume support - must do this before signal_sets_init()
4928 if (SR_initialize() != 0) {
4929 perror("SR_initialize failed");
4930 return JNI_ERR;
4931 }
4932
4933 Linux::signal_sets_init();
4934 Linux::install_signal_handlers();
4935
4936 // Check and sets minimum stack sizes against command line options
4937 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4938 return JNI_ERR;
4939 }
4940 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4941
4942 #if defined(IA32)
4943 workaround_expand_exec_shield_cs_limit();
4944 #endif
4945
4946 Linux::libpthread_init();
4947 Linux::sched_getcpu_init();
4948 log_info(os)("HotSpot is running with %s, %s",
4949 Linux::glibc_version(), Linux::libpthread_version());
4950
4951 if (UseNUMA) {
4952 if (!Linux::libnuma_init()) {
4953 UseNUMA = false;
4954 } else {
4955 if ((Linux::numa_max_node() < 1)) {
4956 // There's only one node(they start from 0), disable NUMA.
4957 UseNUMA = false;
4958 }
4959 }
4960
4961 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4962 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4963 // we can make the adaptive lgrp chunk resizing work. If the user specified both
4964 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
4965 // and disable adaptive resizing.
4966 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4967 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
4968 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4969 UseAdaptiveSizePolicy = false;
4970 UseAdaptiveNUMAChunkSizing = false;
4971 }
4972 }
4973
4974 if (!UseNUMA && ForceNUMA) {
4975 UseNUMA = true;
4976 }
4977 }
4978
4979 if (MaxFDLimit) {
4980 // set the number of file descriptors to max. print out error
4981 // if getrlimit/setrlimit fails but continue regardless.
4982 struct rlimit nbr_files;
4983 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4984 if (status != 0) {
4985 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4986 } else {
4987 nbr_files.rlim_cur = nbr_files.rlim_max;
4988 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4989 if (status != 0) {
4990 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4991 }
4992 }
4993 }
4994
4995 // Initialize lock used to serialize thread creation (see os::create_thread)
4996 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4997
4998 // at-exit methods are called in the reverse order of their registration.
4999 // atexit functions are called on return from main or as a result of a
5000 // call to exit(3C). There can be only 32 of these functions registered
5001 // and atexit() does not set errno.
5002
5003 if (PerfAllowAtExitRegistration) {
5004 // only register atexit functions if PerfAllowAtExitRegistration is set.
5005 // atexit functions can be delayed until process exit time, which
5006 // can be problematic for embedded VM situations. Embedded VMs should
5007 // call DestroyJavaVM() to assure that VM resources are released.
5008
5009 // note: perfMemory_exit_helper atexit function may be removed in
5010 // the future if the appropriate cleanup code can be added to the
5011 // VM_Exit VMOperation's doit method.
5012 if (atexit(perfMemory_exit_helper) != 0) {
5013 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5014 }
5015 }
5016
5017 // initialize thread priority policy
5018 prio_init();
5019
5020 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5021 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5022 }
5023 return JNI_OK;
5024 }
5025
5026 // Mark the polling page as unreadable
5027 void os::make_polling_page_unreadable(void) {
5028 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5029 fatal("Could not disable polling page");
5030 }
5031 }
5032
5033 // Mark the polling page as readable
5034 void os::make_polling_page_readable(void) {
5035 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5036 fatal("Could not enable polling page");
5037 }
5038 }
5039
5040 // older glibc versions don't have this macro (which expands to
5041 // an optimized bit-counting function) so we have to roll our own
5042 #ifndef CPU_COUNT
5043
5044 static int _cpu_count(const cpu_set_t* cpus) {
5045 int count = 0;
5046 // only look up to the number of configured processors
5047 for (int i = 0; i < os::processor_count(); i++) {
5048 if (CPU_ISSET(i, cpus)) {
5049 count++;
5050 }
5051 }
5052 return count;
5053 }
5054
5055 #define CPU_COUNT(cpus) _cpu_count(cpus)
5056
5057 #endif // CPU_COUNT
5058
5059 // Get the current number of available processors for this process.
5060 // This value can change at any time during a process's lifetime.
5061 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5062 // If it appears there may be more than 1024 processors then we do a
5063 // dynamic check - see 6515172 for details.
5064 // If anything goes wrong we fallback to returning the number of online
5065 // processors - which can be greater than the number available to the process.
5066 int os::Linux::active_processor_count() {
5067 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5068 cpu_set_t* cpus_p = &cpus;
5069 int cpus_size = sizeof(cpu_set_t);
5070
5071 int configured_cpus = os::processor_count(); // upper bound on available cpus
5072 int cpu_count = 0;
5073
5074 // old build platforms may not support dynamic cpu sets
5075 #ifdef CPU_ALLOC
5076
5077 // To enable easy testing of the dynamic path on different platforms we
5078 // introduce a diagnostic flag: UseCpuAllocPath
5079 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5080 // kernel may use a mask bigger than cpu_set_t
5081 log_trace(os)("active_processor_count: using dynamic path %s"
5082 "- configured processors: %d",
5083 UseCpuAllocPath ? "(forced) " : "",
5084 configured_cpus);
5085 cpus_p = CPU_ALLOC(configured_cpus);
5086 if (cpus_p != NULL) {
5087 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5088 // zero it just to be safe
5089 CPU_ZERO_S(cpus_size, cpus_p);
5090 }
5091 else {
5092 // failed to allocate so fallback to online cpus
5093 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5094 log_trace(os)("active_processor_count: "
5095 "CPU_ALLOC failed (%s) - using "
5096 "online processor count: %d",
5097 os::strerror(errno), online_cpus);
5098 return online_cpus;
5099 }
5100 }
5101 else {
5102 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5103 configured_cpus);
5104 }
5105 #else // CPU_ALLOC
5106 // these stubs won't be executed
5107 #define CPU_COUNT_S(size, cpus) -1
5108 #define CPU_FREE(cpus)
5109
5110 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5111 configured_cpus);
5112 #endif // CPU_ALLOC
5113
5114 // pid 0 means the current thread - which we have to assume represents the process
5115 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5116 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5117 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5118 }
5119 else {
5120 cpu_count = CPU_COUNT(cpus_p);
5121 }
5122 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5123 }
5124 else {
5125 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5126 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5127 "which may exceed available processors", os::strerror(errno), cpu_count);
5128 }
5129
5130 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5131 CPU_FREE(cpus_p);
5132 }
5133
5134 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5135 return cpu_count;
5136 }
5137
5138 // Determine the active processor count from one of
5139 // three different sources:
5140 //
5141 // 1. User option -XX:ActiveProcessorCount
5142 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5143 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5144 //
5145 // Option 1, if specified, will always override.
5146 // If the cgroup subsystem is active and configured, we
5147 // will return the min of the cgroup and option 2 results.
5148 // This is required since tools, such as numactl, that
5149 // alter cpu affinity do not update cgroup subsystem
5150 // cpuset configuration files.
5151 int os::active_processor_count() {
5152 // User has overridden the number of active processors
5153 if (ActiveProcessorCount > 0) {
5154 log_trace(os)("active_processor_count: "
5155 "active processor count set by user : %d",
5156 ActiveProcessorCount);
5157 return ActiveProcessorCount;
5158 }
5159
5160 int active_cpus;
5161 if (OSContainer::is_containerized()) {
5162 active_cpus = OSContainer::active_processor_count();
5163 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5164 active_cpus);
5165 } else {
5166 active_cpus = os::Linux::active_processor_count();
5167 }
5168
5169 return active_cpus;
5170 }
5171
5172 void os::set_native_thread_name(const char *name) {
5173 if (Linux::_pthread_setname_np) {
5174 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5175 snprintf(buf, sizeof(buf), "%s", name);
5176 buf[sizeof(buf) - 1] = '\0';
5177 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5178 // ERANGE should not happen; all other errors should just be ignored.
5179 assert(rc != ERANGE, "pthread_setname_np failed");
5180 }
5181 }
5182
5183 bool os::distribute_processes(uint length, uint* distribution) {
5184 // Not yet implemented.
5185 return false;
5186 }
5187
5188 bool os::bind_to_processor(uint processor_id) {
5189 // Not yet implemented.
5190 return false;
5191 }
5192
5193 ///
5194
5195 void os::SuspendedThreadTask::internal_do_task() {
5196 if (do_suspend(_thread->osthread())) {
5197 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5198 do_task(context);
5199 do_resume(_thread->osthread());
5200 }
5201 }
5202
5203 ////////////////////////////////////////////////////////////////////////////////
5204 // debug support
5205
5206 bool os::find(address addr, outputStream* st) {
5207 Dl_info dlinfo;
5208 memset(&dlinfo, 0, sizeof(dlinfo));
5209 if (dladdr(addr, &dlinfo) != 0) {
5210 st->print(PTR_FORMAT ": ", p2i(addr));
5211 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5212 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5213 p2i(addr) - p2i(dlinfo.dli_saddr));
5214 } else if (dlinfo.dli_fbase != NULL) {
5215 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5216 } else {
5217 st->print("<absolute address>");
5218 }
5219 if (dlinfo.dli_fname != NULL) {
5220 st->print(" in %s", dlinfo.dli_fname);
5221 }
5222 if (dlinfo.dli_fbase != NULL) {
5223 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5224 }
5225 st->cr();
5226
5227 if (Verbose) {
5228 // decode some bytes around the PC
5229 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5230 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5231 address lowest = (address) dlinfo.dli_sname;
5232 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5233 if (begin < lowest) begin = lowest;
5234 Dl_info dlinfo2;
5235 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5236 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5237 end = (address) dlinfo2.dli_saddr;
5238 }
5239 Disassembler::decode(begin, end, st);
5240 }
5241 return true;
5242 }
5243 return false;
5244 }
5245
5246 ////////////////////////////////////////////////////////////////////////////////
5247 // misc
5248
5249 // This does not do anything on Linux. This is basically a hook for being
5250 // able to use structured exception handling (thread-local exception filters)
5251 // on, e.g., Win32.
5252 void
5253 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5254 JavaCallArguments* args, Thread* thread) {
5255 f(value, method, args, thread);
5256 }
5257
5258 void os::print_statistics() {
5259 }
5260
5261 bool os::message_box(const char* title, const char* message) {
5262 int i;
5263 fdStream err(defaultStream::error_fd());
5264 for (i = 0; i < 78; i++) err.print_raw("=");
5265 err.cr();
5266 err.print_raw_cr(title);
5267 for (i = 0; i < 78; i++) err.print_raw("-");
5268 err.cr();
5269 err.print_raw_cr(message);
5270 for (i = 0; i < 78; i++) err.print_raw("=");
5271 err.cr();
5272
5273 char buf[16];
5274 // Prevent process from exiting upon "read error" without consuming all CPU
5275 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5276
5277 return buf[0] == 'y' || buf[0] == 'Y';
5278 }
5279
5280 int os::stat(const char *path, struct stat *sbuf) {
5281 char pathbuf[MAX_PATH];
5282 if (strlen(path) > MAX_PATH - 1) {
5283 errno = ENAMETOOLONG;
5284 return -1;
5285 }
5286 os::native_path(strcpy(pathbuf, path));
5287 return ::stat(pathbuf, sbuf);
5288 }
5289
5290 // Is a (classpath) directory empty?
5291 bool os::dir_is_empty(const char* path) {
5292 DIR *dir = NULL;
5293 struct dirent *ptr;
5294
5295 dir = opendir(path);
5296 if (dir == NULL) return true;
5297
5298 // Scan the directory
5299 bool result = true;
5300 char buf[sizeof(struct dirent) + MAX_PATH];
5301 while (result && (ptr = ::readdir(dir)) != NULL) {
5302 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5303 result = false;
5304 }
5305 }
5306 closedir(dir);
5307 return result;
5308 }
5309
5310 // This code originates from JDK's sysOpen and open64_w
5311 // from src/solaris/hpi/src/system_md.c
5312
5313 int os::open(const char *path, int oflag, int mode) {
5314 if (strlen(path) > MAX_PATH - 1) {
5315 errno = ENAMETOOLONG;
5316 return -1;
5317 }
5318
5319 // All file descriptors that are opened in the Java process and not
5320 // specifically destined for a subprocess should have the close-on-exec
5321 // flag set. If we don't set it, then careless 3rd party native code
5322 // might fork and exec without closing all appropriate file descriptors
5323 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5324 // turn might:
5325 //
5326 // - cause end-of-file to fail to be detected on some file
5327 // descriptors, resulting in mysterious hangs, or
5328 //
5329 // - might cause an fopen in the subprocess to fail on a system
5330 // suffering from bug 1085341.
5331 //
5332 // (Yes, the default setting of the close-on-exec flag is a Unix
5333 // design flaw)
5334 //
5335 // See:
5336 // 1085341: 32-bit stdio routines should support file descriptors >255
5337 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5338 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5339 //
5340 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5341 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5342 // because it saves a system call and removes a small window where the flag
5343 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5344 // and we fall back to using FD_CLOEXEC (see below).
5345 #ifdef O_CLOEXEC
5346 oflag |= O_CLOEXEC;
5347 #endif
5348
5349 int fd = ::open64(path, oflag, mode);
5350 if (fd == -1) return -1;
5351
5352 //If the open succeeded, the file might still be a directory
5353 {
5354 struct stat64 buf64;
5355 int ret = ::fstat64(fd, &buf64);
5356 int st_mode = buf64.st_mode;
5357
5358 if (ret != -1) {
5359 if ((st_mode & S_IFMT) == S_IFDIR) {
5360 errno = EISDIR;
5361 ::close(fd);
5362 return -1;
5363 }
5364 } else {
5365 ::close(fd);
5366 return -1;
5367 }
5368 }
5369
5370 #ifdef FD_CLOEXEC
5371 // Validate that the use of the O_CLOEXEC flag on open above worked.
5372 // With recent kernels, we will perform this check exactly once.
5373 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5374 if (!O_CLOEXEC_is_known_to_work) {
5375 int flags = ::fcntl(fd, F_GETFD);
5376 if (flags != -1) {
5377 if ((flags & FD_CLOEXEC) != 0)
5378 O_CLOEXEC_is_known_to_work = 1;
5379 else
5380 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5381 }
5382 }
5383 #endif
5384
5385 return fd;
5386 }
5387
5388
5389 // create binary file, rewriting existing file if required
5390 int os::create_binary_file(const char* path, bool rewrite_existing) {
5391 int oflags = O_WRONLY | O_CREAT;
5392 if (!rewrite_existing) {
5393 oflags |= O_EXCL;
5394 }
5395 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5396 }
5397
5398 // return current position of file pointer
5399 jlong os::current_file_offset(int fd) {
5400 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5401 }
5402
5403 // move file pointer to the specified offset
5404 jlong os::seek_to_file_offset(int fd, jlong offset) {
5405 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5406 }
5407
5408 // This code originates from JDK's sysAvailable
5409 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5410
5411 int os::available(int fd, jlong *bytes) {
5412 jlong cur, end;
5413 int mode;
5414 struct stat64 buf64;
5415
5416 if (::fstat64(fd, &buf64) >= 0) {
5417 mode = buf64.st_mode;
5418 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5419 int n;
5420 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5421 *bytes = n;
5422 return 1;
5423 }
5424 }
5425 }
5426 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5427 return 0;
5428 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5429 return 0;
5430 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5431 return 0;
5432 }
5433 *bytes = end - cur;
5434 return 1;
5435 }
5436
5437 // Map a block of memory.
5438 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5439 char *addr, size_t bytes, bool read_only,
5440 bool allow_exec) {
5441 int prot;
5442 int flags = MAP_PRIVATE;
5443
5444 if (read_only) {
5445 prot = PROT_READ;
5446 } else {
5447 prot = PROT_READ | PROT_WRITE;
5448 }
5449
5450 if (allow_exec) {
5451 prot |= PROT_EXEC;
5452 }
5453
5454 if (addr != NULL) {
5455 flags |= MAP_FIXED;
5456 }
5457
5458 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5459 fd, file_offset);
5460 if (mapped_address == MAP_FAILED) {
5461 return NULL;
5462 }
5463 return mapped_address;
5464 }
5465
5466
5467 // Remap a block of memory.
5468 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5469 char *addr, size_t bytes, bool read_only,
5470 bool allow_exec) {
5471 // same as map_memory() on this OS
5472 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5473 allow_exec);
5474 }
5475
5476
5477 // Unmap a block of memory.
5478 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5479 return munmap(addr, bytes) == 0;
5480 }
5481
5482 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5483
5484 static clockid_t thread_cpu_clockid(Thread* thread) {
5485 pthread_t tid = thread->osthread()->pthread_id();
5486 clockid_t clockid;
5487
5488 // Get thread clockid
5489 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5490 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5491 return clockid;
5492 }
5493
5494 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5495 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5496 // of a thread.
5497 //
5498 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5499 // the fast estimate available on the platform.
5500
5501 jlong os::current_thread_cpu_time() {
5502 if (os::Linux::supports_fast_thread_cpu_time()) {
5503 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5504 } else {
5505 // return user + sys since the cost is the same
5506 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5507 }
5508 }
5509
5510 jlong os::thread_cpu_time(Thread* thread) {
5511 // consistent with what current_thread_cpu_time() returns
5512 if (os::Linux::supports_fast_thread_cpu_time()) {
5513 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5514 } else {
5515 return slow_thread_cpu_time(thread, true /* user + sys */);
5516 }
5517 }
5518
5519 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5520 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5521 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5522 } else {
5523 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5524 }
5525 }
5526
5527 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5528 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5529 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5530 } else {
5531 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5532 }
5533 }
5534
5535 // -1 on error.
5536 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5537 pid_t tid = thread->osthread()->thread_id();
5538 char *s;
5539 char stat[2048];
5540 int statlen;
5541 char proc_name[64];
5542 int count;
5543 long sys_time, user_time;
5544 char cdummy;
5545 int idummy;
5546 long ldummy;
5547 FILE *fp;
5548
5549 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5550 fp = fopen(proc_name, "r");
5551 if (fp == NULL) return -1;
5552 statlen = fread(stat, 1, 2047, fp);
5553 stat[statlen] = '\0';
5554 fclose(fp);
5555
5556 // Skip pid and the command string. Note that we could be dealing with
5557 // weird command names, e.g. user could decide to rename java launcher
5558 // to "java 1.4.2 :)", then the stat file would look like
5559 // 1234 (java 1.4.2 :)) R ... ...
5560 // We don't really need to know the command string, just find the last
5561 // occurrence of ")" and then start parsing from there. See bug 4726580.
5562 s = strrchr(stat, ')');
5563 if (s == NULL) return -1;
5564
5565 // Skip blank chars
5566 do { s++; } while (s && isspace(*s));
5567
5568 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5569 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5570 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5571 &user_time, &sys_time);
5572 if (count != 13) return -1;
5573 if (user_sys_cpu_time) {
5574 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5575 } else {
5576 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5577 }
5578 }
5579
5580 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5581 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5582 info_ptr->may_skip_backward = false; // elapsed time not wall time
5583 info_ptr->may_skip_forward = false; // elapsed time not wall time
5584 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5585 }
5586
5587 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5588 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5589 info_ptr->may_skip_backward = false; // elapsed time not wall time
5590 info_ptr->may_skip_forward = false; // elapsed time not wall time
5591 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5592 }
5593
5594 bool os::is_thread_cpu_time_supported() {
5595 return true;
5596 }
5597
5598 // System loadavg support. Returns -1 if load average cannot be obtained.
5599 // Linux doesn't yet have a (official) notion of processor sets,
5600 // so just return the system wide load average.
5601 int os::loadavg(double loadavg[], int nelem) {
5602 return ::getloadavg(loadavg, nelem);
5603 }
5604
5605 void os::pause() {
5606 char filename[MAX_PATH];
5607 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5608 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5609 } else {
5610 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5611 }
5612
5613 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5614 if (fd != -1) {
5615 struct stat buf;
5616 ::close(fd);
5617 while (::stat(filename, &buf) == 0) {
5618 (void)::poll(NULL, 0, 100);
5619 }
5620 } else {
5621 jio_fprintf(stderr,
5622 "Could not open pause file '%s', continuing immediately.\n", filename);
5623 }
5624 }
5625
5626 extern char** environ;
5627
5628 // Run the specified command in a separate process. Return its exit value,
5629 // or -1 on failure (e.g. can't fork a new process).
5630 // Unlike system(), this function can be called from signal handler. It
5631 // doesn't block SIGINT et al.
5632 int os::fork_and_exec(char* cmd) {
5633 const char * argv[4] = {"sh", "-c", cmd, NULL};
5634
5635 pid_t pid = fork();
5636
5637 if (pid < 0) {
5638 // fork failed
5639 return -1;
5640
5641 } else if (pid == 0) {
5642 // child process
5643
5644 execve("/bin/sh", (char* const*)argv, environ);
5645
5646 // execve failed
5647 _exit(-1);
5648
5649 } else {
5650 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5651 // care about the actual exit code, for now.
5652
5653 int status;
5654
5655 // Wait for the child process to exit. This returns immediately if
5656 // the child has already exited. */
5657 while (waitpid(pid, &status, 0) < 0) {
5658 switch (errno) {
5659 case ECHILD: return 0;
5660 case EINTR: break;
5661 default: return -1;
5662 }
5663 }
5664
5665 if (WIFEXITED(status)) {
5666 // The child exited normally; get its exit code.
5667 return WEXITSTATUS(status);
5668 } else if (WIFSIGNALED(status)) {
5669 // The child exited because of a signal
5670 // The best value to return is 0x80 + signal number,
5671 // because that is what all Unix shells do, and because
5672 // it allows callers to distinguish between process exit and
5673 // process death by signal.
5674 return 0x80 + WTERMSIG(status);
5675 } else {
5676 // Unknown exit code; pass it through
5677 return status;
5678 }
5679 }
5680 }
5681
5682 // Get the default path to the core file
5683 // Returns the length of the string
5684 int os::get_core_path(char* buffer, size_t bufferSize) {
5685 /*
5686 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5687 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5688 */
5689 const int core_pattern_len = 129;
5690 char core_pattern[core_pattern_len] = {0};
5691
5692 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5693 if (core_pattern_file == -1) {
5694 return -1;
5695 }
5696
5697 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5698 ::close(core_pattern_file);
5699 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5700 return -1;
5701 }
5702 if (core_pattern[ret-1] == '\n') {
5703 core_pattern[ret-1] = '\0';
5704 } else {
5705 core_pattern[ret] = '\0';
5706 }
5707
5708 char *pid_pos = strstr(core_pattern, "%p");
5709 int written;
5710
5711 if (core_pattern[0] == '/') {
5712 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5713 } else {
5714 char cwd[PATH_MAX];
5715
5716 const char* p = get_current_directory(cwd, PATH_MAX);
5717 if (p == NULL) {
5718 return -1;
5719 }
5720
5721 if (core_pattern[0] == '|') {
5722 written = jio_snprintf(buffer, bufferSize,
5723 "\"%s\" (or dumping to %s/core.%d)",
5724 &core_pattern[1], p, current_process_id());
5725 } else {
5726 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5727 }
5728 }
5729
5730 if (written < 0) {
5731 return -1;
5732 }
5733
5734 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5735 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5736
5737 if (core_uses_pid_file != -1) {
5738 char core_uses_pid = 0;
5739 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5740 ::close(core_uses_pid_file);
5741
5742 if (core_uses_pid == '1') {
5743 jio_snprintf(buffer + written, bufferSize - written,
5744 ".%d", current_process_id());
5745 }
5746 }
5747 }
5748
5749 return strlen(buffer);
5750 }
5751
5752 bool os::start_debugging(char *buf, int buflen) {
5753 int len = (int)strlen(buf);
5754 char *p = &buf[len];
5755
5756 jio_snprintf(p, buflen-len,
5757 "\n\n"
5758 "Do you want to debug the problem?\n\n"
5759 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5760 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5761 "Otherwise, press RETURN to abort...",
5762 os::current_process_id(), os::current_process_id(),
5763 os::current_thread_id(), os::current_thread_id());
5764
5765 bool yes = os::message_box("Unexpected Error", buf);
5766
5767 if (yes) {
5768 // yes, user asked VM to launch debugger
5769 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5770 os::current_process_id(), os::current_process_id());
5771
5772 os::fork_and_exec(buf);
5773 yes = false;
5774 }
5775 return yes;
5776 }
5777
5778
5779 // Java/Compiler thread:
5780 //
5781 // Low memory addresses
5782 // P0 +------------------------+
5783 // | |\ Java thread created by VM does not have glibc
5784 // | glibc guard page | - guard page, attached Java thread usually has
5785 // | |/ 1 glibc guard page.
5786 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5787 // | |\
5788 // | HotSpot Guard Pages | - red, yellow and reserved pages
5789 // | |/
5790 // +------------------------+ JavaThread::stack_reserved_zone_base()
5791 // | |\
5792 // | Normal Stack | -
5793 // | |/
5794 // P2 +------------------------+ Thread::stack_base()
5795 //
5796 // Non-Java thread:
5797 //
5798 // Low memory addresses
5799 // P0 +------------------------+
5800 // | |\
5801 // | glibc guard page | - usually 1 page
5802 // | |/
5803 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5804 // | |\
5805 // | Normal Stack | -
5806 // | |/
5807 // P2 +------------------------+ Thread::stack_base()
5808 //
5809 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5810 // returned from pthread_attr_getstack().
5811 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5812 // of the stack size given in pthread_attr. We work around this for
5813 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5814 //
5815 #ifndef ZERO
5816 static void current_stack_region(address * bottom, size_t * size) {
5817 if (os::is_primordial_thread()) {
5818 // primordial thread needs special handling because pthread_getattr_np()
5819 // may return bogus value.
5820 *bottom = os::Linux::initial_thread_stack_bottom();
5821 *size = os::Linux::initial_thread_stack_size();
5822 } else {
5823 pthread_attr_t attr;
5824
5825 int rslt = pthread_getattr_np(pthread_self(), &attr);
5826
5827 // JVM needs to know exact stack location, abort if it fails
5828 if (rslt != 0) {
5829 if (rslt == ENOMEM) {
5830 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5831 } else {
5832 fatal("pthread_getattr_np failed with error = %d", rslt);
5833 }
5834 }
5835
5836 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5837 fatal("Cannot locate current stack attributes!");
5838 }
5839
5840 // Work around NPTL stack guard error.
5841 size_t guard_size = 0;
5842 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5843 if (rslt != 0) {
5844 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5845 }
5846 *bottom += guard_size;
5847 *size -= guard_size;
5848
5849 pthread_attr_destroy(&attr);
5850
5851 }
5852 assert(os::current_stack_pointer() >= *bottom &&
5853 os::current_stack_pointer() < *bottom + *size, "just checking");
5854 }
5855
5856 address os::current_stack_base() {
5857 address bottom;
5858 size_t size;
5859 current_stack_region(&bottom, &size);
5860 return (bottom + size);
5861 }
5862
5863 size_t os::current_stack_size() {
5864 // This stack size includes the usable stack and HotSpot guard pages
5865 // (for the threads that have Hotspot guard pages).
5866 address bottom;
5867 size_t size;
5868 current_stack_region(&bottom, &size);
5869 return size;
5870 }
5871 #endif
5872
5873 static inline struct timespec get_mtime(const char* filename) {
5874 struct stat st;
5875 int ret = os::stat(filename, &st);
5876 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5877 return st.st_mtim;
5878 }
5879
5880 int os::compare_file_modified_times(const char* file1, const char* file2) {
5881 struct timespec filetime1 = get_mtime(file1);
5882 struct timespec filetime2 = get_mtime(file2);
5883 int diff = filetime1.tv_sec - filetime2.tv_sec;
5884 if (diff == 0) {
5885 return filetime1.tv_nsec - filetime2.tv_nsec;
5886 }
5887 return diff;
5888 }
5889
5890 /////////////// Unit tests ///////////////
5891
5892 #ifndef PRODUCT
5893
5894 #define test_log(...) \
5895 do { \
5896 if (VerboseInternalVMTests) { \
5897 tty->print_cr(__VA_ARGS__); \
5898 tty->flush(); \
5899 } \
5900 } while (false)
5901
5902 class TestReserveMemorySpecial : AllStatic {
5903 public:
5904 static void small_page_write(void* addr, size_t size) {
5905 size_t page_size = os::vm_page_size();
5906
5907 char* end = (char*)addr + size;
5908 for (char* p = (char*)addr; p < end; p += page_size) {
5909 *p = 1;
5910 }
5911 }
5912
5913 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5914 if (!UseHugeTLBFS) {
5915 return;
5916 }
5917
5918 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5919
5920 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5921
5922 if (addr != NULL) {
5923 small_page_write(addr, size);
5924
5925 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5926 }
5927 }
5928
5929 static void test_reserve_memory_special_huge_tlbfs_only() {
5930 if (!UseHugeTLBFS) {
5931 return;
5932 }
5933
5934 size_t lp = os::large_page_size();
5935
5936 for (size_t size = lp; size <= lp * 10; size += lp) {
5937 test_reserve_memory_special_huge_tlbfs_only(size);
5938 }
5939 }
5940
5941 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5942 size_t lp = os::large_page_size();
5943 size_t ag = os::vm_allocation_granularity();
5944
5945 // sizes to test
5946 const size_t sizes[] = {
5947 lp, lp + ag, lp + lp / 2, lp * 2,
5948 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5949 lp * 10, lp * 10 + lp / 2
5950 };
5951 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5952
5953 // For each size/alignment combination, we test three scenarios:
5954 // 1) with req_addr == NULL
5955 // 2) with a non-null req_addr at which we expect to successfully allocate
5956 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5957 // expect the allocation to either fail or to ignore req_addr
5958
5959 // Pre-allocate two areas; they shall be as large as the largest allocation
5960 // and aligned to the largest alignment we will be testing.
5961 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5962 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5963 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5964 -1, 0);
5965 assert(mapping1 != MAP_FAILED, "should work");
5966
5967 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5968 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5969 -1, 0);
5970 assert(mapping2 != MAP_FAILED, "should work");
5971
5972 // Unmap the first mapping, but leave the second mapping intact: the first
5973 // mapping will serve as a value for a "good" req_addr (case 2). The second
5974 // mapping, still intact, as "bad" req_addr (case 3).
5975 ::munmap(mapping1, mapping_size);
5976
5977 // Case 1
5978 test_log("%s, req_addr NULL:", __FUNCTION__);
5979 test_log("size align result");
5980
5981 for (int i = 0; i < num_sizes; i++) {
5982 const size_t size = sizes[i];
5983 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5984 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5985 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
5986 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
5987 if (p != NULL) {
5988 assert(is_aligned(p, alignment), "must be");
5989 small_page_write(p, size);
5990 os::Linux::release_memory_special_huge_tlbfs(p, size);
5991 }
5992 }
5993 }
5994
5995 // Case 2
5996 test_log("%s, req_addr non-NULL:", __FUNCTION__);
5997 test_log("size align req_addr result");
5998
5999 for (int i = 0; i < num_sizes; i++) {
6000 const size_t size = sizes[i];
6001 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6002 char* const req_addr = align_up(mapping1, alignment);
6003 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6004 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6005 size, alignment, p2i(req_addr), p2i(p),
6006 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6007 if (p != NULL) {
6008 assert(p == req_addr, "must be");
6009 small_page_write(p, size);
6010 os::Linux::release_memory_special_huge_tlbfs(p, size);
6011 }
6012 }
6013 }
6014
6015 // Case 3
6016 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6017 test_log("size align req_addr result");
6018
6019 for (int i = 0; i < num_sizes; i++) {
6020 const size_t size = sizes[i];
6021 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6022 char* const req_addr = align_up(mapping2, alignment);
6023 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6024 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6025 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6026 // as the area around req_addr contains already existing mappings, the API should always
6027 // return NULL (as per contract, it cannot return another address)
6028 assert(p == NULL, "must be");
6029 }
6030 }
6031
6032 ::munmap(mapping2, mapping_size);
6033
6034 }
6035
6036 static void test_reserve_memory_special_huge_tlbfs() {
6037 if (!UseHugeTLBFS) {
6038 return;
6039 }
6040
6041 test_reserve_memory_special_huge_tlbfs_only();
6042 test_reserve_memory_special_huge_tlbfs_mixed();
6043 }
6044
6045 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6046 if (!UseSHM) {
6047 return;
6048 }
6049
6050 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6051
6052 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6053
6054 if (addr != NULL) {
6055 assert(is_aligned(addr, alignment), "Check");
6056 assert(is_aligned(addr, os::large_page_size()), "Check");
6057
6058 small_page_write(addr, size);
6059
6060 os::Linux::release_memory_special_shm(addr, size);
6061 }
6062 }
6063
6064 static void test_reserve_memory_special_shm() {
6065 size_t lp = os::large_page_size();
6066 size_t ag = os::vm_allocation_granularity();
6067
6068 for (size_t size = ag; size < lp * 3; size += ag) {
6069 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6070 test_reserve_memory_special_shm(size, alignment);
6071 }
6072 }
6073 }
6074
6075 static void test() {
6076 test_reserve_memory_special_huge_tlbfs();
6077 test_reserve_memory_special_shm();
6078 }
6079 };
6080
6081 void TestReserveMemorySpecial_test() {
6082 TestReserveMemorySpecial::test();
6083 }
6084
6085 #endif