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.inline.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.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 enum CoredumpFilterBit {
132 FILE_BACKED_PVT_BIT = 1 << 2,
133 LARGEPAGES_BIT = 1 << 6,
134 DAX_SHARED_BIT = 1 << 8
135 };
136
137 ////////////////////////////////////////////////////////////////////////////////
138 // global variables
139 julong os::Linux::_physical_memory = 0;
140
141 address os::Linux::_initial_thread_stack_bottom = NULL;
142 uintptr_t os::Linux::_initial_thread_stack_size = 0;
143
144 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
145 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
146 Mutex* os::Linux::_createThread_lock = NULL;
147 pthread_t os::Linux::_main_thread;
148 int os::Linux::_page_size = -1;
149 bool os::Linux::_supports_fast_thread_cpu_time = false;
150 uint32_t os::Linux::_os_version = 0;
151 const char * os::Linux::_glibc_version = NULL;
152 const char * os::Linux::_libpthread_version = NULL;
153
154 static jlong initial_time_count=0;
155
156 static int clock_tics_per_sec = 100;
157
158 // If the VM might have been created on the primordial thread, we need to resolve the
159 // primordial thread stack bounds and check if the current thread might be the
160 // primordial thread in places. If we know that the primordial thread is never used,
161 // such as when the VM was created by one of the standard java launchers, we can
162 // avoid this
163 static bool suppress_primordial_thread_resolution = false;
164
165 // For diagnostics to print a message once. see run_periodic_checks
166 static sigset_t check_signal_done;
167 static bool check_signals = true;
168
169 // Signal number used to suspend/resume a thread
170
171 // do not use any signal number less than SIGSEGV, see 4355769
172 static int SR_signum = SIGUSR2;
173 sigset_t SR_sigset;
174
175 // utility functions
176
177 static int SR_initialize();
178
179 julong os::available_memory() {
180 return Linux::available_memory();
181 }
182
183 julong os::Linux::available_memory() {
184 // values in struct sysinfo are "unsigned long"
185 struct sysinfo si;
186 julong avail_mem;
187
188 if (OSContainer::is_containerized()) {
189 jlong mem_limit, mem_usage;
190 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
191 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
192 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
193 }
194 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
195 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
196 }
197 if (mem_limit > 0 && mem_usage > 0 ) {
198 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
199 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
200 return avail_mem;
201 }
202 }
203
204 sysinfo(&si);
205 avail_mem = (julong)si.freeram * si.mem_unit;
206 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
207 return avail_mem;
208 }
209
210 julong os::physical_memory() {
211 jlong phys_mem = 0;
212 if (OSContainer::is_containerized()) {
213 jlong mem_limit;
214 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
215 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
216 return mem_limit;
217 }
218 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
219 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
220 }
221
222 phys_mem = Linux::physical_memory();
223 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
224 return phys_mem;
225 }
226
227 // Return true if user is running as root.
228
229 bool os::have_special_privileges() {
230 static bool init = false;
231 static bool privileges = false;
232 if (!init) {
233 privileges = (getuid() != geteuid()) || (getgid() != getegid());
234 init = true;
235 }
236 return privileges;
237 }
238
239
240 #ifndef SYS_gettid
241 // i386: 224, ia64: 1105, amd64: 186, sparc 143
242 #ifdef __ia64__
243 #define SYS_gettid 1105
244 #else
245 #ifdef __i386__
246 #define SYS_gettid 224
247 #else
248 #ifdef __amd64__
249 #define SYS_gettid 186
250 #else
251 #ifdef __sparc__
252 #define SYS_gettid 143
253 #else
254 #error define gettid for the arch
255 #endif
256 #endif
257 #endif
258 #endif
259 #endif
260
261
262 // pid_t gettid()
263 //
264 // Returns the kernel thread id of the currently running thread. Kernel
265 // thread id is used to access /proc.
266 pid_t os::Linux::gettid() {
267 int rslt = syscall(SYS_gettid);
268 assert(rslt != -1, "must be."); // old linuxthreads implementation?
269 return (pid_t)rslt;
270 }
271
272 // Most versions of linux have a bug where the number of processors are
273 // determined by looking at the /proc file system. In a chroot environment,
274 // the system call returns 1.
275 static bool unsafe_chroot_detected = false;
276 static const char *unstable_chroot_error = "/proc file system not found.\n"
277 "Java may be unstable running multithreaded in a chroot "
278 "environment on Linux when /proc filesystem is not mounted.";
279
280 void os::Linux::initialize_system_info() {
281 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
282 if (processor_count() == 1) {
283 pid_t pid = os::Linux::gettid();
284 char fname[32];
285 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
286 FILE *fp = fopen(fname, "r");
287 if (fp == NULL) {
288 unsafe_chroot_detected = true;
289 } else {
290 fclose(fp);
291 }
292 }
293 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
294 assert(processor_count() > 0, "linux error");
295 }
296
297 void os::init_system_properties_values() {
298 // The next steps are taken in the product version:
299 //
300 // Obtain the JAVA_HOME value from the location of libjvm.so.
301 // This library should be located at:
302 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
303 //
304 // If "/jre/lib/" appears at the right place in the path, then we
305 // assume libjvm.so is installed in a JDK and we use this path.
306 //
307 // Otherwise exit with message: "Could not create the Java virtual machine."
308 //
309 // The following extra steps are taken in the debugging version:
310 //
311 // If "/jre/lib/" does NOT appear at the right place in the path
312 // instead of exit check for $JAVA_HOME environment variable.
313 //
314 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
315 // then we append a fake suffix "hotspot/libjvm.so" to this path so
316 // it looks like libjvm.so is installed there
317 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
318 //
319 // Otherwise exit.
320 //
321 // Important note: if the location of libjvm.so changes this
322 // code needs to be changed accordingly.
323
324 // See ld(1):
325 // The linker uses the following search paths to locate required
326 // shared libraries:
327 // 1: ...
328 // ...
329 // 7: The default directories, normally /lib and /usr/lib.
330 #ifndef OVERRIDE_LIBPATH
331 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
332 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
333 #else
334 #define DEFAULT_LIBPATH "/lib:/usr/lib"
335 #endif
336 #else
337 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
338 #endif
339
340 // Base path of extensions installed on the system.
341 #define SYS_EXT_DIR "/usr/java/packages"
342 #define EXTENSIONS_DIR "/lib/ext"
343
344 // Buffer that fits several sprintfs.
345 // Note that the space for the colon and the trailing null are provided
346 // by the nulls included by the sizeof operator.
347 const size_t bufsize =
348 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
349 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
350 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
351
352 // sysclasspath, java_home, dll_dir
353 {
354 char *pslash;
355 os::jvm_path(buf, bufsize);
356
357 // Found the full path to libjvm.so.
358 // Now cut the path to <java_home>/jre if we can.
359 pslash = strrchr(buf, '/');
360 if (pslash != NULL) {
361 *pslash = '\0'; // Get rid of /libjvm.so.
362 }
363 pslash = strrchr(buf, '/');
364 if (pslash != NULL) {
365 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
366 }
367 Arguments::set_dll_dir(buf);
368
369 if (pslash != NULL) {
370 pslash = strrchr(buf, '/');
371 if (pslash != NULL) {
372 *pslash = '\0'; // Get rid of /lib.
373 }
374 }
375 Arguments::set_java_home(buf);
376 if (!set_boot_path('/', ':')) {
377 vm_exit_during_initialization("Failed setting boot class path.", NULL);
378 }
379 }
380
381 // Where to look for native libraries.
382 //
383 // Note: Due to a legacy implementation, most of the library path
384 // is set in the launcher. This was to accomodate linking restrictions
385 // on legacy Linux implementations (which are no longer supported).
386 // Eventually, all the library path setting will be done here.
387 //
388 // However, to prevent the proliferation of improperly built native
389 // libraries, the new path component /usr/java/packages is added here.
390 // Eventually, all the library path setting will be done here.
391 {
392 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
393 // should always exist (until the legacy problem cited above is
394 // addressed).
395 const char *v = ::getenv("LD_LIBRARY_PATH");
396 const char *v_colon = ":";
397 if (v == NULL) { v = ""; v_colon = ""; }
398 // That's +1 for the colon and +1 for the trailing '\0'.
399 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
400 strlen(v) + 1 +
401 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
402 mtInternal);
403 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
404 Arguments::set_library_path(ld_library_path);
405 FREE_C_HEAP_ARRAY(char, ld_library_path);
406 }
407
408 // Extensions directories.
409 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
410 Arguments::set_ext_dirs(buf);
411
412 FREE_C_HEAP_ARRAY(char, buf);
413
414 #undef DEFAULT_LIBPATH
415 #undef SYS_EXT_DIR
416 #undef EXTENSIONS_DIR
417 }
418
419 ////////////////////////////////////////////////////////////////////////////////
420 // breakpoint support
421
422 void os::breakpoint() {
423 BREAKPOINT;
424 }
425
426 extern "C" void breakpoint() {
427 // use debugger to set breakpoint here
428 }
429
430 ////////////////////////////////////////////////////////////////////////////////
431 // signal support
432
433 debug_only(static bool signal_sets_initialized = false);
434 static sigset_t unblocked_sigs, vm_sigs;
435
436 void os::Linux::signal_sets_init() {
437 // Should also have an assertion stating we are still single-threaded.
438 assert(!signal_sets_initialized, "Already initialized");
439 // Fill in signals that are necessarily unblocked for all threads in
440 // the VM. Currently, we unblock the following signals:
441 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
442 // by -Xrs (=ReduceSignalUsage));
443 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
444 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
445 // the dispositions or masks wrt these signals.
446 // Programs embedding the VM that want to use the above signals for their
447 // own purposes must, at this time, use the "-Xrs" option to prevent
448 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
449 // (See bug 4345157, and other related bugs).
450 // In reality, though, unblocking these signals is really a nop, since
451 // these signals are not blocked by default.
452 sigemptyset(&unblocked_sigs);
453 sigaddset(&unblocked_sigs, SIGILL);
454 sigaddset(&unblocked_sigs, SIGSEGV);
455 sigaddset(&unblocked_sigs, SIGBUS);
456 sigaddset(&unblocked_sigs, SIGFPE);
457 #if defined(PPC64)
458 sigaddset(&unblocked_sigs, SIGTRAP);
459 #endif
460 sigaddset(&unblocked_sigs, SR_signum);
461
462 if (!ReduceSignalUsage) {
463 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
464 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
465 }
466 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
468 }
469 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
470 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
471 }
472 }
473 // Fill in signals that are blocked by all but the VM thread.
474 sigemptyset(&vm_sigs);
475 if (!ReduceSignalUsage) {
476 sigaddset(&vm_sigs, BREAK_SIGNAL);
477 }
478 debug_only(signal_sets_initialized = true);
479
480 }
481
482 // These are signals that are unblocked while a thread is running Java.
483 // (For some reason, they get blocked by default.)
484 sigset_t* os::Linux::unblocked_signals() {
485 assert(signal_sets_initialized, "Not initialized");
486 return &unblocked_sigs;
487 }
488
489 // These are the signals that are blocked while a (non-VM) thread is
490 // running Java. Only the VM thread handles these signals.
491 sigset_t* os::Linux::vm_signals() {
492 assert(signal_sets_initialized, "Not initialized");
493 return &vm_sigs;
494 }
495
496 void os::Linux::hotspot_sigmask(Thread* thread) {
497
498 //Save caller's signal mask before setting VM signal mask
499 sigset_t caller_sigmask;
500 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
501
502 OSThread* osthread = thread->osthread();
503 osthread->set_caller_sigmask(caller_sigmask);
504
505 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
506
507 if (!ReduceSignalUsage) {
508 if (thread->is_VM_thread()) {
509 // Only the VM thread handles BREAK_SIGNAL ...
510 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
511 } else {
512 // ... all other threads block BREAK_SIGNAL
513 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
514 }
515 }
516 }
517
518 //////////////////////////////////////////////////////////////////////////////
519 // detecting pthread library
520
521 void os::Linux::libpthread_init() {
522 // Save glibc and pthread version strings.
523 #if !defined(_CS_GNU_LIBC_VERSION) || \
524 !defined(_CS_GNU_LIBPTHREAD_VERSION)
525 #error "glibc too old (< 2.3.2)"
526 #endif
527
528 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
529 assert(n > 0, "cannot retrieve glibc version");
530 char *str = (char *)malloc(n, mtInternal);
531 confstr(_CS_GNU_LIBC_VERSION, str, n);
532 os::Linux::set_glibc_version(str);
533
534 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
535 assert(n > 0, "cannot retrieve pthread version");
536 str = (char *)malloc(n, mtInternal);
537 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
538 os::Linux::set_libpthread_version(str);
539 }
540
541 /////////////////////////////////////////////////////////////////////////////
542 // thread stack expansion
543
544 // os::Linux::manually_expand_stack() takes care of expanding the thread
545 // stack. Note that this is normally not needed: pthread stacks allocate
546 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
547 // committed. Therefore it is not necessary to expand the stack manually.
548 //
549 // Manually expanding the stack was historically needed on LinuxThreads
550 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
551 // it is kept to deal with very rare corner cases:
552 //
553 // For one, user may run the VM on an own implementation of threads
554 // whose stacks are - like the old LinuxThreads - implemented using
555 // mmap(MAP_GROWSDOWN).
556 //
557 // Also, this coding may be needed if the VM is running on the primordial
558 // thread. Normally we avoid running on the primordial thread; however,
559 // user may still invoke the VM on the primordial thread.
560 //
561 // The following historical comment describes the details about running
562 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
563
564
565 // Force Linux kernel to expand current thread stack. If "bottom" is close
566 // to the stack guard, caller should block all signals.
567 //
568 // MAP_GROWSDOWN:
569 // A special mmap() flag that is used to implement thread stacks. It tells
570 // kernel that the memory region should extend downwards when needed. This
571 // allows early versions of LinuxThreads to only mmap the first few pages
572 // when creating a new thread. Linux kernel will automatically expand thread
573 // stack as needed (on page faults).
574 //
575 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
576 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
577 // region, it's hard to tell if the fault is due to a legitimate stack
578 // access or because of reading/writing non-exist memory (e.g. buffer
579 // overrun). As a rule, if the fault happens below current stack pointer,
580 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
581 // application (see Linux kernel fault.c).
582 //
583 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
584 // stack overflow detection.
585 //
586 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
587 // not use MAP_GROWSDOWN.
588 //
589 // To get around the problem and allow stack banging on Linux, we need to
590 // manually expand thread stack after receiving the SIGSEGV.
591 //
592 // There are two ways to expand thread stack to address "bottom", we used
593 // both of them in JVM before 1.5:
594 // 1. adjust stack pointer first so that it is below "bottom", and then
595 // touch "bottom"
596 // 2. mmap() the page in question
597 //
598 // Now alternate signal stack is gone, it's harder to use 2. For instance,
599 // if current sp is already near the lower end of page 101, and we need to
600 // call mmap() to map page 100, it is possible that part of the mmap() frame
601 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
602 // That will destroy the mmap() frame and cause VM to crash.
603 //
604 // The following code works by adjusting sp first, then accessing the "bottom"
605 // page to force a page fault. Linux kernel will then automatically expand the
606 // stack mapping.
607 //
608 // _expand_stack_to() assumes its frame size is less than page size, which
609 // should always be true if the function is not inlined.
610
611 static void NOINLINE _expand_stack_to(address bottom) {
612 address sp;
613 size_t size;
614 volatile char *p;
615
616 // Adjust bottom to point to the largest address within the same page, it
617 // gives us a one-page buffer if alloca() allocates slightly more memory.
618 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
619 bottom += os::Linux::page_size() - 1;
620
621 // sp might be slightly above current stack pointer; if that's the case, we
622 // will alloca() a little more space than necessary, which is OK. Don't use
623 // os::current_stack_pointer(), as its result can be slightly below current
624 // stack pointer, causing us to not alloca enough to reach "bottom".
625 sp = (address)&sp;
626
627 if (sp > bottom) {
628 size = sp - bottom;
629 p = (volatile char *)alloca(size);
630 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
631 p[0] = '\0';
632 }
633 }
634
635 void os::Linux::expand_stack_to(address bottom) {
636 _expand_stack_to(bottom);
637 }
638
639 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
640 assert(t!=NULL, "just checking");
641 assert(t->osthread()->expanding_stack(), "expand should be set");
642 assert(t->stack_base() != NULL, "stack_base was not initialized");
643
644 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
645 sigset_t mask_all, old_sigset;
646 sigfillset(&mask_all);
647 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
648 _expand_stack_to(addr);
649 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
650 return true;
651 }
652 return false;
653 }
654
655 //////////////////////////////////////////////////////////////////////////////
656 // create new thread
657
658 // Thread start routine for all newly created threads
659 static void *thread_native_entry(Thread *thread) {
660
661 thread->record_stack_base_and_size();
662
663 // Try to randomize the cache line index of hot stack frames.
664 // This helps when threads of the same stack traces evict each other's
665 // cache lines. The threads can be either from the same JVM instance, or
666 // from different JVM instances. The benefit is especially true for
667 // processors with hyperthreading technology.
668 static int counter = 0;
669 int pid = os::current_process_id();
670 alloca(((pid ^ counter++) & 7) * 128);
671
672 thread->initialize_thread_current();
673
674 OSThread* osthread = thread->osthread();
675 Monitor* sync = osthread->startThread_lock();
676
677 osthread->set_thread_id(os::current_thread_id());
678
679 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
680 os::current_thread_id(), (uintx) pthread_self());
681
682 if (UseNUMA) {
683 int lgrp_id = os::numa_get_group_id();
684 if (lgrp_id != -1) {
685 thread->set_lgrp_id(lgrp_id);
686 }
687 }
688 // initialize signal mask for this thread
689 os::Linux::hotspot_sigmask(thread);
690
691 // initialize floating point control register
692 os::Linux::init_thread_fpu_state();
693
694 // handshaking with parent thread
695 {
696 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
697
698 // notify parent thread
699 osthread->set_state(INITIALIZED);
700 sync->notify_all();
701
702 // wait until os::start_thread()
703 while (osthread->get_state() == INITIALIZED) {
704 sync->wait(Mutex::_no_safepoint_check_flag);
705 }
706 }
707
708 // call one more level start routine
709 thread->call_run();
710
711 // Note: at this point the thread object may already have deleted itself.
712 // Prevent dereferencing it from here on out.
713 thread = NULL;
714
715 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
716 os::current_thread_id(), (uintx) pthread_self());
717
718 return 0;
719 }
720
721 bool os::create_thread(Thread* thread, ThreadType thr_type,
722 size_t req_stack_size) {
723 assert(thread->osthread() == NULL, "caller responsible");
724
725 // Allocate the OSThread object
726 OSThread* osthread = new OSThread(NULL, NULL);
727 if (osthread == NULL) {
728 return false;
729 }
730
731 // set the correct thread state
732 osthread->set_thread_type(thr_type);
733
734 // Initial state is ALLOCATED but not INITIALIZED
735 osthread->set_state(ALLOCATED);
736
737 thread->set_osthread(osthread);
738
739 // init thread attributes
740 pthread_attr_t attr;
741 pthread_attr_init(&attr);
742 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
743
744 // Calculate stack size if it's not specified by caller.
745 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
746 // In the Linux NPTL pthread implementation the guard size mechanism
747 // is not implemented properly. The posix standard requires adding
748 // the size of the guard pages to the stack size, instead Linux
749 // takes the space out of 'stacksize'. Thus we adapt the requested
750 // stack_size by the size of the guard pages to mimick proper
751 // behaviour. However, be careful not to end up with a size
752 // of zero due to overflow. Don't add the guard page in that case.
753 size_t guard_size = os::Linux::default_guard_size(thr_type);
754 if (stack_size <= SIZE_MAX - guard_size) {
755 stack_size += guard_size;
756 }
757 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
758
759 int status = pthread_attr_setstacksize(&attr, stack_size);
760 assert_status(status == 0, status, "pthread_attr_setstacksize");
761
762 // Configure glibc guard page.
763 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
764
765 ThreadState state;
766
767 {
768 pthread_t tid;
769 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
770
771 char buf[64];
772 if (ret == 0) {
773 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
774 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
775 } else {
776 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
777 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
778 }
779
780 pthread_attr_destroy(&attr);
781
782 if (ret != 0) {
783 // Need to clean up stuff we've allocated so far
784 thread->set_osthread(NULL);
785 delete osthread;
786 return false;
787 }
788
789 // Store pthread info into the OSThread
790 osthread->set_pthread_id(tid);
791
792 // Wait until child thread is either initialized or aborted
793 {
794 Monitor* sync_with_child = osthread->startThread_lock();
795 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
796 while ((state = osthread->get_state()) == ALLOCATED) {
797 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
798 }
799 }
800 }
801
802 // Aborted due to thread limit being reached
803 if (state == ZOMBIE) {
804 thread->set_osthread(NULL);
805 delete osthread;
806 return false;
807 }
808
809 // The thread is returned suspended (in state INITIALIZED),
810 // and is started higher up in the call chain
811 assert(state == INITIALIZED, "race condition");
812 return true;
813 }
814
815 /////////////////////////////////////////////////////////////////////////////
816 // attach existing thread
817
818 // bootstrap the main thread
819 bool os::create_main_thread(JavaThread* thread) {
820 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
821 return create_attached_thread(thread);
822 }
823
824 bool os::create_attached_thread(JavaThread* thread) {
825 #ifdef ASSERT
826 thread->verify_not_published();
827 #endif
828
829 // Allocate the OSThread object
830 OSThread* osthread = new OSThread(NULL, NULL);
831
832 if (osthread == NULL) {
833 return false;
834 }
835
836 // Store pthread info into the OSThread
837 osthread->set_thread_id(os::Linux::gettid());
838 osthread->set_pthread_id(::pthread_self());
839
840 // initialize floating point control register
841 os::Linux::init_thread_fpu_state();
842
843 // Initial thread state is RUNNABLE
844 osthread->set_state(RUNNABLE);
845
846 thread->set_osthread(osthread);
847
848 if (UseNUMA) {
849 int lgrp_id = os::numa_get_group_id();
850 if (lgrp_id != -1) {
851 thread->set_lgrp_id(lgrp_id);
852 }
853 }
854
855 if (os::is_primordial_thread()) {
856 // If current thread is primordial thread, its stack is mapped on demand,
857 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
858 // the entire stack region to avoid SEGV in stack banging.
859 // It is also useful to get around the heap-stack-gap problem on SuSE
860 // kernel (see 4821821 for details). We first expand stack to the top
861 // of yellow zone, then enable stack yellow zone (order is significant,
862 // enabling yellow zone first will crash JVM on SuSE Linux), so there
863 // is no gap between the last two virtual memory regions.
864
865 JavaThread *jt = (JavaThread *)thread;
866 address addr = jt->stack_reserved_zone_base();
867 assert(addr != NULL, "initialization problem?");
868 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
869
870 osthread->set_expanding_stack();
871 os::Linux::manually_expand_stack(jt, addr);
872 osthread->clear_expanding_stack();
873 }
874
875 // initialize signal mask for this thread
876 // and save the caller's signal mask
877 os::Linux::hotspot_sigmask(thread);
878
879 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
880 os::current_thread_id(), (uintx) pthread_self());
881
882 return true;
883 }
884
885 void os::pd_start_thread(Thread* thread) {
886 OSThread * osthread = thread->osthread();
887 assert(osthread->get_state() != INITIALIZED, "just checking");
888 Monitor* sync_with_child = osthread->startThread_lock();
889 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
890 sync_with_child->notify();
891 }
892
893 // Free Linux resources related to the OSThread
894 void os::free_thread(OSThread* osthread) {
895 assert(osthread != NULL, "osthread not set");
896
897 // We are told to free resources of the argument thread,
898 // but we can only really operate on the current thread.
899 assert(Thread::current()->osthread() == osthread,
900 "os::free_thread but not current thread");
901
902 #ifdef ASSERT
903 sigset_t current;
904 sigemptyset(¤t);
905 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
906 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
907 #endif
908
909 // Restore caller's signal mask
910 sigset_t sigmask = osthread->caller_sigmask();
911 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
912
913 delete osthread;
914 }
915
916 //////////////////////////////////////////////////////////////////////////////
917 // primordial thread
918
919 // Check if current thread is the primordial thread, similar to Solaris thr_main.
920 bool os::is_primordial_thread(void) {
921 if (suppress_primordial_thread_resolution) {
922 return false;
923 }
924 char dummy;
925 // If called before init complete, thread stack bottom will be null.
926 // Can be called if fatal error occurs before initialization.
927 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
928 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
929 os::Linux::initial_thread_stack_size() != 0,
930 "os::init did not locate primordial thread's stack region");
931 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
932 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
933 os::Linux::initial_thread_stack_size()) {
934 return true;
935 } else {
936 return false;
937 }
938 }
939
940 // Find the virtual memory area that contains addr
941 static bool find_vma(address addr, address* vma_low, address* vma_high) {
942 FILE *fp = fopen("/proc/self/maps", "r");
943 if (fp) {
944 address low, high;
945 while (!feof(fp)) {
946 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
947 if (low <= addr && addr < high) {
948 if (vma_low) *vma_low = low;
949 if (vma_high) *vma_high = high;
950 fclose(fp);
951 return true;
952 }
953 }
954 for (;;) {
955 int ch = fgetc(fp);
956 if (ch == EOF || ch == (int)'\n') break;
957 }
958 }
959 fclose(fp);
960 }
961 return false;
962 }
963
964 // Locate primordial thread stack. This special handling of primordial thread stack
965 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
966 // bogus value for the primordial process thread. While the launcher has created
967 // the VM in a new thread since JDK 6, we still have to allow for the use of the
968 // JNI invocation API from a primordial thread.
969 void os::Linux::capture_initial_stack(size_t max_size) {
970
971 // max_size is either 0 (which means accept OS default for thread stacks) or
972 // a user-specified value known to be at least the minimum needed. If we
973 // are actually on the primordial thread we can make it appear that we have a
974 // smaller max_size stack by inserting the guard pages at that location. But we
975 // cannot do anything to emulate a larger stack than what has been provided by
976 // the OS or threading library. In fact if we try to use a stack greater than
977 // what is set by rlimit then we will crash the hosting process.
978
979 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
980 // If this is "unlimited" then it will be a huge value.
981 struct rlimit rlim;
982 getrlimit(RLIMIT_STACK, &rlim);
983 size_t stack_size = rlim.rlim_cur;
984
985 // 6308388: a bug in ld.so will relocate its own .data section to the
986 // lower end of primordial stack; reduce ulimit -s value a little bit
987 // so we won't install guard page on ld.so's data section.
988 // But ensure we don't underflow the stack size - allow 1 page spare
989 if (stack_size >= (size_t)(3 * page_size())) {
990 stack_size -= 2 * page_size();
991 }
992
993 // Try to figure out where the stack base (top) is. This is harder.
994 //
995 // When an application is started, glibc saves the initial stack pointer in
996 // a global variable "__libc_stack_end", which is then used by system
997 // libraries. __libc_stack_end should be pretty close to stack top. The
998 // variable is available since the very early days. However, because it is
999 // a private interface, it could disappear in the future.
1000 //
1001 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1002 // to __libc_stack_end, it is very close to stack top, but isn't the real
1003 // stack top. Note that /proc may not exist if VM is running as a chroot
1004 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1005 // /proc/<pid>/stat could change in the future (though unlikely).
1006 //
1007 // We try __libc_stack_end first. If that doesn't work, look for
1008 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1009 // as a hint, which should work well in most cases.
1010
1011 uintptr_t stack_start;
1012
1013 // try __libc_stack_end first
1014 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1015 if (p && *p) {
1016 stack_start = *p;
1017 } else {
1018 // see if we can get the start_stack field from /proc/self/stat
1019 FILE *fp;
1020 int pid;
1021 char state;
1022 int ppid;
1023 int pgrp;
1024 int session;
1025 int nr;
1026 int tpgrp;
1027 unsigned long flags;
1028 unsigned long minflt;
1029 unsigned long cminflt;
1030 unsigned long majflt;
1031 unsigned long cmajflt;
1032 unsigned long utime;
1033 unsigned long stime;
1034 long cutime;
1035 long cstime;
1036 long prio;
1037 long nice;
1038 long junk;
1039 long it_real;
1040 uintptr_t start;
1041 uintptr_t vsize;
1042 intptr_t rss;
1043 uintptr_t rsslim;
1044 uintptr_t scodes;
1045 uintptr_t ecode;
1046 int i;
1047
1048 // Figure what the primordial thread stack base is. Code is inspired
1049 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1050 // followed by command name surrounded by parentheses, state, etc.
1051 char stat[2048];
1052 int statlen;
1053
1054 fp = fopen("/proc/self/stat", "r");
1055 if (fp) {
1056 statlen = fread(stat, 1, 2047, fp);
1057 stat[statlen] = '\0';
1058 fclose(fp);
1059
1060 // Skip pid and the command string. Note that we could be dealing with
1061 // weird command names, e.g. user could decide to rename java launcher
1062 // to "java 1.4.2 :)", then the stat file would look like
1063 // 1234 (java 1.4.2 :)) R ... ...
1064 // We don't really need to know the command string, just find the last
1065 // occurrence of ")" and then start parsing from there. See bug 4726580.
1066 char * s = strrchr(stat, ')');
1067
1068 i = 0;
1069 if (s) {
1070 // Skip blank chars
1071 do { s++; } while (s && isspace(*s));
1072
1073 #define _UFM UINTX_FORMAT
1074 #define _DFM INTX_FORMAT
1075
1076 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1077 // 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
1078 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,
1079 &state, // 3 %c
1080 &ppid, // 4 %d
1081 &pgrp, // 5 %d
1082 &session, // 6 %d
1083 &nr, // 7 %d
1084 &tpgrp, // 8 %d
1085 &flags, // 9 %lu
1086 &minflt, // 10 %lu
1087 &cminflt, // 11 %lu
1088 &majflt, // 12 %lu
1089 &cmajflt, // 13 %lu
1090 &utime, // 14 %lu
1091 &stime, // 15 %lu
1092 &cutime, // 16 %ld
1093 &cstime, // 17 %ld
1094 &prio, // 18 %ld
1095 &nice, // 19 %ld
1096 &junk, // 20 %ld
1097 &it_real, // 21 %ld
1098 &start, // 22 UINTX_FORMAT
1099 &vsize, // 23 UINTX_FORMAT
1100 &rss, // 24 INTX_FORMAT
1101 &rsslim, // 25 UINTX_FORMAT
1102 &scodes, // 26 UINTX_FORMAT
1103 &ecode, // 27 UINTX_FORMAT
1104 &stack_start); // 28 UINTX_FORMAT
1105 }
1106
1107 #undef _UFM
1108 #undef _DFM
1109
1110 if (i != 28 - 2) {
1111 assert(false, "Bad conversion from /proc/self/stat");
1112 // product mode - assume we are the primordial thread, good luck in the
1113 // embedded case.
1114 warning("Can't detect primordial thread stack location - bad conversion");
1115 stack_start = (uintptr_t) &rlim;
1116 }
1117 } else {
1118 // For some reason we can't open /proc/self/stat (for example, running on
1119 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1120 // most cases, so don't abort:
1121 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1122 stack_start = (uintptr_t) &rlim;
1123 }
1124 }
1125
1126 // Now we have a pointer (stack_start) very close to the stack top, the
1127 // next thing to do is to figure out the exact location of stack top. We
1128 // can find out the virtual memory area that contains stack_start by
1129 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1130 // and its upper limit is the real stack top. (again, this would fail if
1131 // running inside chroot, because /proc may not exist.)
1132
1133 uintptr_t stack_top;
1134 address low, high;
1135 if (find_vma((address)stack_start, &low, &high)) {
1136 // success, "high" is the true stack top. (ignore "low", because initial
1137 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1138 stack_top = (uintptr_t)high;
1139 } else {
1140 // failed, likely because /proc/self/maps does not exist
1141 warning("Can't detect primordial thread stack location - find_vma failed");
1142 // best effort: stack_start is normally within a few pages below the real
1143 // stack top, use it as stack top, and reduce stack size so we won't put
1144 // guard page outside stack.
1145 stack_top = stack_start;
1146 stack_size -= 16 * page_size();
1147 }
1148
1149 // stack_top could be partially down the page so align it
1150 stack_top = align_up(stack_top, page_size());
1151
1152 // Allowed stack value is minimum of max_size and what we derived from rlimit
1153 if (max_size > 0) {
1154 _initial_thread_stack_size = MIN2(max_size, stack_size);
1155 } else {
1156 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1157 // clamp it at 8MB as we do on Solaris
1158 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1159 }
1160 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1161 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1162
1163 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1164
1165 if (log_is_enabled(Info, os, thread)) {
1166 // See if we seem to be on primordial process thread
1167 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1168 uintptr_t(&rlim) < stack_top;
1169
1170 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1171 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1172 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1173 stack_top, intptr_t(_initial_thread_stack_bottom));
1174 }
1175 }
1176
1177 ////////////////////////////////////////////////////////////////////////////////
1178 // time support
1179
1180 #ifndef SUPPORTS_CLOCK_MONOTONIC
1181 #error "Build platform doesn't support clock_gettime and related functionality"
1182 #endif
1183
1184 // Time since start-up in seconds to a fine granularity.
1185 // Used by VMSelfDestructTimer and the MemProfiler.
1186 double os::elapsedTime() {
1187
1188 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1189 }
1190
1191 jlong os::elapsed_counter() {
1192 return javaTimeNanos() - initial_time_count;
1193 }
1194
1195 jlong os::elapsed_frequency() {
1196 return NANOSECS_PER_SEC; // nanosecond resolution
1197 }
1198
1199 bool os::supports_vtime() { return true; }
1200 bool os::enable_vtime() { return false; }
1201 bool os::vtime_enabled() { return false; }
1202
1203 double os::elapsedVTime() {
1204 struct rusage usage;
1205 int retval = getrusage(RUSAGE_THREAD, &usage);
1206 if (retval == 0) {
1207 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);
1208 } else {
1209 // better than nothing, but not much
1210 return elapsedTime();
1211 }
1212 }
1213
1214 jlong os::javaTimeMillis() {
1215 timeval time;
1216 int status = gettimeofday(&time, NULL);
1217 assert(status != -1, "linux error");
1218 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1219 }
1220
1221 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1222 timeval time;
1223 int status = gettimeofday(&time, NULL);
1224 assert(status != -1, "linux error");
1225 seconds = jlong(time.tv_sec);
1226 nanos = jlong(time.tv_usec) * 1000;
1227 }
1228
1229 void os::Linux::fast_thread_clock_init() {
1230 if (!UseLinuxPosixThreadCPUClocks) {
1231 return;
1232 }
1233 clockid_t clockid;
1234 struct timespec tp;
1235 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1236 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1237
1238 // Switch to using fast clocks for thread cpu time if
1239 // the clock_getres() returns 0 error code.
1240 // Note, that some kernels may support the current thread
1241 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1242 // returned by the pthread_getcpuclockid().
1243 // If the fast Posix clocks are supported then the clock_getres()
1244 // must return at least tp.tv_sec == 0 which means a resolution
1245 // better than 1 sec. This is extra check for reliability.
1246
1247 if (pthread_getcpuclockid_func &&
1248 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1249 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1250 _supports_fast_thread_cpu_time = true;
1251 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1252 }
1253 }
1254
1255 jlong os::javaTimeNanos() {
1256 if (os::supports_monotonic_clock()) {
1257 struct timespec tp;
1258 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1259 assert(status == 0, "gettime error");
1260 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1261 return result;
1262 } else {
1263 timeval time;
1264 int status = gettimeofday(&time, NULL);
1265 assert(status != -1, "linux error");
1266 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1267 return 1000 * usecs;
1268 }
1269 }
1270
1271 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1272 if (os::supports_monotonic_clock()) {
1273 info_ptr->max_value = ALL_64_BITS;
1274
1275 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1276 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1277 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1278 } else {
1279 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1280 info_ptr->max_value = ALL_64_BITS;
1281
1282 // gettimeofday is a real time clock so it skips
1283 info_ptr->may_skip_backward = true;
1284 info_ptr->may_skip_forward = true;
1285 }
1286
1287 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1288 }
1289
1290 // Return the real, user, and system times in seconds from an
1291 // arbitrary fixed point in the past.
1292 bool os::getTimesSecs(double* process_real_time,
1293 double* process_user_time,
1294 double* process_system_time) {
1295 struct tms ticks;
1296 clock_t real_ticks = times(&ticks);
1297
1298 if (real_ticks == (clock_t) (-1)) {
1299 return false;
1300 } else {
1301 double ticks_per_second = (double) clock_tics_per_sec;
1302 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1303 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1304 *process_real_time = ((double) real_ticks) / ticks_per_second;
1305
1306 return true;
1307 }
1308 }
1309
1310
1311 char * os::local_time_string(char *buf, size_t buflen) {
1312 struct tm t;
1313 time_t long_time;
1314 time(&long_time);
1315 localtime_r(&long_time, &t);
1316 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1317 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1318 t.tm_hour, t.tm_min, t.tm_sec);
1319 return buf;
1320 }
1321
1322 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1323 return localtime_r(clock, res);
1324 }
1325
1326 ////////////////////////////////////////////////////////////////////////////////
1327 // runtime exit support
1328
1329 // Note: os::shutdown() might be called very early during initialization, or
1330 // called from signal handler. Before adding something to os::shutdown(), make
1331 // sure it is async-safe and can handle partially initialized VM.
1332 void os::shutdown() {
1333
1334 // allow PerfMemory to attempt cleanup of any persistent resources
1335 perfMemory_exit();
1336
1337 // needs to remove object in file system
1338 AttachListener::abort();
1339
1340 // flush buffered output, finish log files
1341 ostream_abort();
1342
1343 // Check for abort hook
1344 abort_hook_t abort_hook = Arguments::abort_hook();
1345 if (abort_hook != NULL) {
1346 abort_hook();
1347 }
1348
1349 }
1350
1351 // Note: os::abort() might be called very early during initialization, or
1352 // called from signal handler. Before adding something to os::abort(), make
1353 // sure it is async-safe and can handle partially initialized VM.
1354 void os::abort(bool dump_core, void* siginfo, const void* context) {
1355 os::shutdown();
1356 if (dump_core) {
1357 #if INCLUDE_CDS
1358 if (UseSharedSpaces && DumpPrivateMappingsInCore) {
1359 ClassLoader::close_jrt_image();
1360 }
1361 #endif
1362 #ifndef PRODUCT
1363 fdStream out(defaultStream::output_fd());
1364 out.print_raw("Current thread is ");
1365 char buf[16];
1366 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1367 out.print_raw_cr(buf);
1368 out.print_raw_cr("Dumping core ...");
1369 #endif
1370 ::abort(); // dump core
1371 }
1372
1373 ::exit(1);
1374 }
1375
1376 // Die immediately, no exit hook, no abort hook, no cleanup.
1377 void os::die() {
1378 ::abort();
1379 }
1380
1381 // thread_id is kernel thread id (similar to Solaris LWP id)
1382 intx os::current_thread_id() { return os::Linux::gettid(); }
1383 int os::current_process_id() {
1384 return ::getpid();
1385 }
1386
1387 // DLL functions
1388
1389 const char* os::dll_file_extension() { return ".so"; }
1390
1391 // This must be hard coded because it's the system's temporary
1392 // directory not the java application's temp directory, ala java.io.tmpdir.
1393 const char* os::get_temp_directory() { return "/tmp"; }
1394
1395 static bool file_exists(const char* filename) {
1396 struct stat statbuf;
1397 if (filename == NULL || strlen(filename) == 0) {
1398 return false;
1399 }
1400 return os::stat(filename, &statbuf) == 0;
1401 }
1402
1403 // check if addr is inside libjvm.so
1404 bool os::address_is_in_vm(address addr) {
1405 static address libjvm_base_addr;
1406 Dl_info dlinfo;
1407
1408 if (libjvm_base_addr == NULL) {
1409 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1410 libjvm_base_addr = (address)dlinfo.dli_fbase;
1411 }
1412 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1413 }
1414
1415 if (dladdr((void *)addr, &dlinfo) != 0) {
1416 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1417 }
1418
1419 return false;
1420 }
1421
1422 bool os::dll_address_to_function_name(address addr, char *buf,
1423 int buflen, int *offset,
1424 bool demangle) {
1425 // buf is not optional, but offset is optional
1426 assert(buf != NULL, "sanity check");
1427
1428 Dl_info dlinfo;
1429
1430 if (dladdr((void*)addr, &dlinfo) != 0) {
1431 // see if we have a matching symbol
1432 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1433 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1434 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1435 }
1436 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1437 return true;
1438 }
1439 // no matching symbol so try for just file info
1440 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1441 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1442 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1443 return true;
1444 }
1445 }
1446 }
1447
1448 buf[0] = '\0';
1449 if (offset != NULL) *offset = -1;
1450 return false;
1451 }
1452
1453 struct _address_to_library_name {
1454 address addr; // input : memory address
1455 size_t buflen; // size of fname
1456 char* fname; // output: library name
1457 address base; // library base addr
1458 };
1459
1460 static int address_to_library_name_callback(struct dl_phdr_info *info,
1461 size_t size, void *data) {
1462 int i;
1463 bool found = false;
1464 address libbase = NULL;
1465 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1466
1467 // iterate through all loadable segments
1468 for (i = 0; i < info->dlpi_phnum; i++) {
1469 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1470 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1471 // base address of a library is the lowest address of its loaded
1472 // segments.
1473 if (libbase == NULL || libbase > segbase) {
1474 libbase = segbase;
1475 }
1476 // see if 'addr' is within current segment
1477 if (segbase <= d->addr &&
1478 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1479 found = true;
1480 }
1481 }
1482 }
1483
1484 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1485 // so dll_address_to_library_name() can fall through to use dladdr() which
1486 // can figure out executable name from argv[0].
1487 if (found && info->dlpi_name && info->dlpi_name[0]) {
1488 d->base = libbase;
1489 if (d->fname) {
1490 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1491 }
1492 return 1;
1493 }
1494 return 0;
1495 }
1496
1497 bool os::dll_address_to_library_name(address addr, char* buf,
1498 int buflen, int* offset) {
1499 // buf is not optional, but offset is optional
1500 assert(buf != NULL, "sanity check");
1501
1502 Dl_info dlinfo;
1503 struct _address_to_library_name data;
1504
1505 // There is a bug in old glibc dladdr() implementation that it could resolve
1506 // to wrong library name if the .so file has a base address != NULL. Here
1507 // we iterate through the program headers of all loaded libraries to find
1508 // out which library 'addr' really belongs to. This workaround can be
1509 // removed once the minimum requirement for glibc is moved to 2.3.x.
1510 data.addr = addr;
1511 data.fname = buf;
1512 data.buflen = buflen;
1513 data.base = NULL;
1514 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1515
1516 if (rslt) {
1517 // buf already contains library name
1518 if (offset) *offset = addr - data.base;
1519 return true;
1520 }
1521 if (dladdr((void*)addr, &dlinfo) != 0) {
1522 if (dlinfo.dli_fname != NULL) {
1523 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1524 }
1525 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1526 *offset = addr - (address)dlinfo.dli_fbase;
1527 }
1528 return true;
1529 }
1530
1531 buf[0] = '\0';
1532 if (offset) *offset = -1;
1533 return false;
1534 }
1535
1536 // Loads .dll/.so and
1537 // in case of error it checks if .dll/.so was built for the
1538 // same architecture as Hotspot is running on
1539
1540
1541 // Remember the stack's state. The Linux dynamic linker will change
1542 // the stack to 'executable' at most once, so we must safepoint only once.
1543 bool os::Linux::_stack_is_executable = false;
1544
1545 // VM operation that loads a library. This is necessary if stack protection
1546 // of the Java stacks can be lost during loading the library. If we
1547 // do not stop the Java threads, they can stack overflow before the stacks
1548 // are protected again.
1549 class VM_LinuxDllLoad: public VM_Operation {
1550 private:
1551 const char *_filename;
1552 char *_ebuf;
1553 int _ebuflen;
1554 void *_lib;
1555 public:
1556 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1557 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1558 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1559 void doit() {
1560 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1561 os::Linux::_stack_is_executable = true;
1562 }
1563 void* loaded_library() { return _lib; }
1564 };
1565
1566 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1567 void * result = NULL;
1568 bool load_attempted = false;
1569
1570 // Check whether the library to load might change execution rights
1571 // of the stack. If they are changed, the protection of the stack
1572 // guard pages will be lost. We need a safepoint to fix this.
1573 //
1574 // See Linux man page execstack(8) for more info.
1575 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1576 if (!ElfFile::specifies_noexecstack(filename)) {
1577 if (!is_init_completed()) {
1578 os::Linux::_stack_is_executable = true;
1579 // This is OK - No Java threads have been created yet, and hence no
1580 // stack guard pages to fix.
1581 //
1582 // Dynamic loader will make all stacks executable after
1583 // this function returns, and will not do that again.
1584 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1585 } else {
1586 warning("You have loaded library %s which might have disabled stack guard. "
1587 "The VM will try to fix the stack guard now.\n"
1588 "It's highly recommended that you fix the library with "
1589 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1590 filename);
1591
1592 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1593 JavaThread *jt = JavaThread::current();
1594 if (jt->thread_state() != _thread_in_native) {
1595 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1596 // that requires ExecStack. Cannot enter safe point. Let's give up.
1597 warning("Unable to fix stack guard. Giving up.");
1598 } else {
1599 if (!LoadExecStackDllInVMThread) {
1600 // This is for the case where the DLL has an static
1601 // constructor function that executes JNI code. We cannot
1602 // load such DLLs in the VMThread.
1603 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1604 }
1605
1606 ThreadInVMfromNative tiv(jt);
1607 debug_only(VMNativeEntryWrapper vew;)
1608
1609 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1610 VMThread::execute(&op);
1611 if (LoadExecStackDllInVMThread) {
1612 result = op.loaded_library();
1613 }
1614 load_attempted = true;
1615 }
1616 }
1617 }
1618 }
1619
1620 if (!load_attempted) {
1621 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1622 }
1623
1624 if (result != NULL) {
1625 // Successful loading
1626 return result;
1627 }
1628
1629 Elf32_Ehdr elf_head;
1630 int diag_msg_max_length=ebuflen-strlen(ebuf);
1631 char* diag_msg_buf=ebuf+strlen(ebuf);
1632
1633 if (diag_msg_max_length==0) {
1634 // No more space in ebuf for additional diagnostics message
1635 return NULL;
1636 }
1637
1638
1639 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1640
1641 if (file_descriptor < 0) {
1642 // Can't open library, report dlerror() message
1643 return NULL;
1644 }
1645
1646 bool failed_to_read_elf_head=
1647 (sizeof(elf_head)!=
1648 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1649
1650 ::close(file_descriptor);
1651 if (failed_to_read_elf_head) {
1652 // file i/o error - report dlerror() msg
1653 return NULL;
1654 }
1655
1656 typedef struct {
1657 Elf32_Half code; // Actual value as defined in elf.h
1658 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1659 unsigned char elf_class; // 32 or 64 bit
1660 unsigned char endianess; // MSB or LSB
1661 char* name; // String representation
1662 } arch_t;
1663
1664 #ifndef EM_486
1665 #define EM_486 6 /* Intel 80486 */
1666 #endif
1667 #ifndef EM_AARCH64
1668 #define EM_AARCH64 183 /* ARM AARCH64 */
1669 #endif
1670
1671 static const arch_t arch_array[]={
1672 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1673 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1674 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1675 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1676 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1677 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1678 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1679 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1680 #if defined(VM_LITTLE_ENDIAN)
1681 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1682 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1683 #else
1684 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1685 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1686 #endif
1687 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1688 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1689 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1690 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1691 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1692 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1693 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1694 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1695 };
1696
1697 #if (defined IA32)
1698 static Elf32_Half running_arch_code=EM_386;
1699 #elif (defined AMD64) || (defined X32)
1700 static Elf32_Half running_arch_code=EM_X86_64;
1701 #elif (defined IA64)
1702 static Elf32_Half running_arch_code=EM_IA_64;
1703 #elif (defined __sparc) && (defined _LP64)
1704 static Elf32_Half running_arch_code=EM_SPARCV9;
1705 #elif (defined __sparc) && (!defined _LP64)
1706 static Elf32_Half running_arch_code=EM_SPARC;
1707 #elif (defined __powerpc64__)
1708 static Elf32_Half running_arch_code=EM_PPC64;
1709 #elif (defined __powerpc__)
1710 static Elf32_Half running_arch_code=EM_PPC;
1711 #elif (defined AARCH64)
1712 static Elf32_Half running_arch_code=EM_AARCH64;
1713 #elif (defined ARM)
1714 static Elf32_Half running_arch_code=EM_ARM;
1715 #elif (defined S390)
1716 static Elf32_Half running_arch_code=EM_S390;
1717 #elif (defined ALPHA)
1718 static Elf32_Half running_arch_code=EM_ALPHA;
1719 #elif (defined MIPSEL)
1720 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1721 #elif (defined PARISC)
1722 static Elf32_Half running_arch_code=EM_PARISC;
1723 #elif (defined MIPS)
1724 static Elf32_Half running_arch_code=EM_MIPS;
1725 #elif (defined M68K)
1726 static Elf32_Half running_arch_code=EM_68K;
1727 #elif (defined SH)
1728 static Elf32_Half running_arch_code=EM_SH;
1729 #else
1730 #error Method os::dll_load requires that one of following is defined:\
1731 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1732 #endif
1733
1734 // Identify compatability class for VM's architecture and library's architecture
1735 // Obtain string descriptions for architectures
1736
1737 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1738 int running_arch_index=-1;
1739
1740 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1741 if (running_arch_code == arch_array[i].code) {
1742 running_arch_index = i;
1743 }
1744 if (lib_arch.code == arch_array[i].code) {
1745 lib_arch.compat_class = arch_array[i].compat_class;
1746 lib_arch.name = arch_array[i].name;
1747 }
1748 }
1749
1750 assert(running_arch_index != -1,
1751 "Didn't find running architecture code (running_arch_code) in arch_array");
1752 if (running_arch_index == -1) {
1753 // Even though running architecture detection failed
1754 // we may still continue with reporting dlerror() message
1755 return NULL;
1756 }
1757
1758 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1759 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1760 return NULL;
1761 }
1762
1763 #ifndef S390
1764 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1765 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1766 return NULL;
1767 }
1768 #endif // !S390
1769
1770 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1771 if (lib_arch.name!=NULL) {
1772 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1773 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1774 lib_arch.name, arch_array[running_arch_index].name);
1775 } else {
1776 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1777 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1778 lib_arch.code,
1779 arch_array[running_arch_index].name);
1780 }
1781 }
1782
1783 return NULL;
1784 }
1785
1786 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1787 int ebuflen) {
1788 void * result = ::dlopen(filename, RTLD_LAZY);
1789 if (result == NULL) {
1790 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1791 ebuf[ebuflen-1] = '\0';
1792 }
1793 return result;
1794 }
1795
1796 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1797 int ebuflen) {
1798 void * result = NULL;
1799 if (LoadExecStackDllInVMThread) {
1800 result = dlopen_helper(filename, ebuf, ebuflen);
1801 }
1802
1803 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1804 // library that requires an executable stack, or which does not have this
1805 // stack attribute set, dlopen changes the stack attribute to executable. The
1806 // read protection of the guard pages gets lost.
1807 //
1808 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1809 // may have been queued at the same time.
1810
1811 if (!_stack_is_executable) {
1812 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1813 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1814 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1815 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1816 warning("Attempt to reguard stack yellow zone failed.");
1817 }
1818 }
1819 }
1820 }
1821
1822 return result;
1823 }
1824
1825 void* os::dll_lookup(void* handle, const char* name) {
1826 void* res = dlsym(handle, name);
1827 return res;
1828 }
1829
1830 void* os::get_default_process_handle() {
1831 return (void*)::dlopen(NULL, RTLD_LAZY);
1832 }
1833
1834 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1835 int fd = ::open(filename, O_RDONLY);
1836 if (fd == -1) {
1837 return false;
1838 }
1839
1840 if (hdr != NULL) {
1841 st->print_cr("%s", hdr);
1842 }
1843
1844 char buf[33];
1845 int bytes;
1846 buf[32] = '\0';
1847 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1848 st->print_raw(buf, bytes);
1849 }
1850
1851 ::close(fd);
1852
1853 return true;
1854 }
1855
1856 void os::print_dll_info(outputStream *st) {
1857 st->print_cr("Dynamic libraries:");
1858
1859 char fname[32];
1860 pid_t pid = os::Linux::gettid();
1861
1862 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1863
1864 if (!_print_ascii_file(fname, st)) {
1865 st->print("Can not get library information for pid = %d\n", pid);
1866 }
1867 }
1868
1869 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1870 FILE *procmapsFile = NULL;
1871
1872 // Open the procfs maps file for the current process
1873 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1874 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1875 char line[PATH_MAX + 100];
1876
1877 // Read line by line from 'file'
1878 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1879 u8 base, top, offset, inode;
1880 char permissions[5];
1881 char device[6];
1882 char name[PATH_MAX + 1];
1883
1884 // Parse fields from line
1885 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1886 &base, &top, permissions, &offset, device, &inode, name);
1887
1888 // Filter by device id '00:00' so that we only get file system mapped files.
1889 if (strcmp(device, "00:00") != 0) {
1890
1891 // Call callback with the fields of interest
1892 if(callback(name, (address)base, (address)top, param)) {
1893 // Oops abort, callback aborted
1894 fclose(procmapsFile);
1895 return 1;
1896 }
1897 }
1898 }
1899 fclose(procmapsFile);
1900 }
1901 return 0;
1902 }
1903
1904 void os::print_os_info_brief(outputStream* st) {
1905 os::Linux::print_distro_info(st);
1906
1907 os::Posix::print_uname_info(st);
1908
1909 os::Linux::print_libversion_info(st);
1910
1911 }
1912
1913 void os::print_os_info(outputStream* st) {
1914 st->print("OS:");
1915
1916 os::Linux::print_distro_info(st);
1917
1918 os::Posix::print_uname_info(st);
1919
1920 // Print warning if unsafe chroot environment detected
1921 if (unsafe_chroot_detected) {
1922 st->print("WARNING!! ");
1923 st->print_cr("%s", unstable_chroot_error);
1924 }
1925
1926 os::Linux::print_libversion_info(st);
1927
1928 os::Posix::print_rlimit_info(st);
1929
1930 os::Posix::print_load_average(st);
1931
1932 os::Linux::print_full_memory_info(st);
1933
1934 os::Linux::print_proc_sys_info(st);
1935
1936 os::Linux::print_ld_preload_file(st);
1937
1938 os::Linux::print_container_info(st);
1939 }
1940
1941 // Try to identify popular distros.
1942 // Most Linux distributions have a /etc/XXX-release file, which contains
1943 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1944 // file that also contains the OS version string. Some have more than one
1945 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1946 // /etc/redhat-release.), so the order is important.
1947 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1948 // their own specific XXX-release file as well as a redhat-release file.
1949 // Because of this the XXX-release file needs to be searched for before the
1950 // redhat-release file.
1951 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1952 // search for redhat-release / SuSE-release needs to be before lsb-release.
1953 // Since the lsb-release file is the new standard it needs to be searched
1954 // before the older style release files.
1955 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1956 // next to last resort. The os-release file is a new standard that contains
1957 // distribution information and the system-release file seems to be an old
1958 // standard that has been replaced by the lsb-release and os-release files.
1959 // Searching for the debian_version file is the last resort. It contains
1960 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1961 // "Debian " is printed before the contents of the debian_version file.
1962
1963 const char* distro_files[] = {
1964 "/etc/oracle-release",
1965 "/etc/mandriva-release",
1966 "/etc/mandrake-release",
1967 "/etc/sun-release",
1968 "/etc/redhat-release",
1969 "/etc/SuSE-release",
1970 "/etc/lsb-release",
1971 "/etc/turbolinux-release",
1972 "/etc/gentoo-release",
1973 "/etc/ltib-release",
1974 "/etc/angstrom-version",
1975 "/etc/system-release",
1976 "/etc/os-release",
1977 NULL };
1978
1979 void os::Linux::print_distro_info(outputStream* st) {
1980 for (int i = 0;; i++) {
1981 const char* file = distro_files[i];
1982 if (file == NULL) {
1983 break; // done
1984 }
1985 // If file prints, we found it.
1986 if (_print_ascii_file(file, st)) {
1987 return;
1988 }
1989 }
1990
1991 if (file_exists("/etc/debian_version")) {
1992 st->print("Debian ");
1993 _print_ascii_file("/etc/debian_version", st);
1994 } else {
1995 st->print("Linux");
1996 }
1997 st->cr();
1998 }
1999
2000 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2001 char buf[256];
2002 while (fgets(buf, sizeof(buf), fp)) {
2003 // Edit out extra stuff in expected format
2004 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2005 char* ptr = strstr(buf, "\""); // the name is in quotes
2006 if (ptr != NULL) {
2007 ptr++; // go beyond first quote
2008 char* nl = strchr(ptr, '\"');
2009 if (nl != NULL) *nl = '\0';
2010 strncpy(distro, ptr, length);
2011 } else {
2012 ptr = strstr(buf, "=");
2013 ptr++; // go beyond equals then
2014 char* nl = strchr(ptr, '\n');
2015 if (nl != NULL) *nl = '\0';
2016 strncpy(distro, ptr, length);
2017 }
2018 return;
2019 } else if (get_first_line) {
2020 char* nl = strchr(buf, '\n');
2021 if (nl != NULL) *nl = '\0';
2022 strncpy(distro, buf, length);
2023 return;
2024 }
2025 }
2026 // print last line and close
2027 char* nl = strchr(buf, '\n');
2028 if (nl != NULL) *nl = '\0';
2029 strncpy(distro, buf, length);
2030 }
2031
2032 static void parse_os_info(char* distro, size_t length, const char* file) {
2033 FILE* fp = fopen(file, "r");
2034 if (fp != NULL) {
2035 // if suse format, print out first line
2036 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2037 parse_os_info_helper(fp, distro, length, get_first_line);
2038 fclose(fp);
2039 }
2040 }
2041
2042 void os::get_summary_os_info(char* buf, size_t buflen) {
2043 for (int i = 0;; i++) {
2044 const char* file = distro_files[i];
2045 if (file == NULL) {
2046 break; // ran out of distro_files
2047 }
2048 if (file_exists(file)) {
2049 parse_os_info(buf, buflen, file);
2050 return;
2051 }
2052 }
2053 // special case for debian
2054 if (file_exists("/etc/debian_version")) {
2055 strncpy(buf, "Debian ", buflen);
2056 if (buflen > 7) {
2057 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2058 }
2059 } else {
2060 strncpy(buf, "Linux", buflen);
2061 }
2062 }
2063
2064 void os::Linux::print_libversion_info(outputStream* st) {
2065 // libc, pthread
2066 st->print("libc:");
2067 st->print("%s ", os::Linux::glibc_version());
2068 st->print("%s ", os::Linux::libpthread_version());
2069 st->cr();
2070 }
2071
2072 void os::Linux::print_proc_sys_info(outputStream* st) {
2073 st->cr();
2074 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2075 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2076 st->cr();
2077 st->cr();
2078
2079 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2080 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2081 st->cr();
2082 st->cr();
2083
2084 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2085 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2086 st->cr();
2087 st->cr();
2088 }
2089
2090 void os::Linux::print_full_memory_info(outputStream* st) {
2091 st->print("\n/proc/meminfo:\n");
2092 _print_ascii_file("/proc/meminfo", st);
2093 st->cr();
2094 }
2095
2096 void os::Linux::print_ld_preload_file(outputStream* st) {
2097 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2098 st->cr();
2099 }
2100
2101 void os::Linux::print_container_info(outputStream* st) {
2102 if (!OSContainer::is_containerized()) {
2103 return;
2104 }
2105
2106 st->print("container (cgroup) information:\n");
2107
2108 const char *p_ct = OSContainer::container_type();
2109 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2110
2111 char *p = OSContainer::cpu_cpuset_cpus();
2112 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2113 free(p);
2114
2115 p = OSContainer::cpu_cpuset_memory_nodes();
2116 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2117 free(p);
2118
2119 int i = OSContainer::active_processor_count();
2120 if (i > 0) {
2121 st->print("active_processor_count: %d\n", i);
2122 } else {
2123 st->print("active_processor_count: failed\n");
2124 }
2125
2126 i = OSContainer::cpu_quota();
2127 st->print("cpu_quota: %d\n", i);
2128
2129 i = OSContainer::cpu_period();
2130 st->print("cpu_period: %d\n", i);
2131
2132 i = OSContainer::cpu_shares();
2133 st->print("cpu_shares: %d\n", i);
2134
2135 jlong j = OSContainer::memory_limit_in_bytes();
2136 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2137
2138 j = OSContainer::memory_and_swap_limit_in_bytes();
2139 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2140
2141 j = OSContainer::memory_soft_limit_in_bytes();
2142 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2143
2144 j = OSContainer::OSContainer::memory_usage_in_bytes();
2145 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2146
2147 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2148 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2149 st->cr();
2150 }
2151
2152 void os::print_memory_info(outputStream* st) {
2153
2154 st->print("Memory:");
2155 st->print(" %dk page", os::vm_page_size()>>10);
2156
2157 // values in struct sysinfo are "unsigned long"
2158 struct sysinfo si;
2159 sysinfo(&si);
2160
2161 st->print(", physical " UINT64_FORMAT "k",
2162 os::physical_memory() >> 10);
2163 st->print("(" UINT64_FORMAT "k free)",
2164 os::available_memory() >> 10);
2165 st->print(", swap " UINT64_FORMAT "k",
2166 ((jlong)si.totalswap * si.mem_unit) >> 10);
2167 st->print("(" UINT64_FORMAT "k free)",
2168 ((jlong)si.freeswap * si.mem_unit) >> 10);
2169 st->cr();
2170 }
2171
2172 // Print the first "model name" line and the first "flags" line
2173 // that we find and nothing more. We assume "model name" comes
2174 // before "flags" so if we find a second "model name", then the
2175 // "flags" field is considered missing.
2176 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2177 #if defined(IA32) || defined(AMD64)
2178 // Other platforms have less repetitive cpuinfo files
2179 FILE *fp = fopen("/proc/cpuinfo", "r");
2180 if (fp) {
2181 while (!feof(fp)) {
2182 if (fgets(buf, buflen, fp)) {
2183 // Assume model name comes before flags
2184 bool model_name_printed = false;
2185 if (strstr(buf, "model name") != NULL) {
2186 if (!model_name_printed) {
2187 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2188 st->print_raw(buf);
2189 model_name_printed = true;
2190 } else {
2191 // model name printed but not flags? Odd, just return
2192 fclose(fp);
2193 return true;
2194 }
2195 }
2196 // print the flags line too
2197 if (strstr(buf, "flags") != NULL) {
2198 st->print_raw(buf);
2199 fclose(fp);
2200 return true;
2201 }
2202 }
2203 }
2204 fclose(fp);
2205 }
2206 #endif // x86 platforms
2207 return false;
2208 }
2209
2210 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2211 // Only print the model name if the platform provides this as a summary
2212 if (!print_model_name_and_flags(st, buf, buflen)) {
2213 st->print("\n/proc/cpuinfo:\n");
2214 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2215 st->print_cr(" <Not Available>");
2216 }
2217 }
2218 }
2219
2220 #if defined(AMD64) || defined(IA32) || defined(X32)
2221 const char* search_string = "model name";
2222 #elif defined(M68K)
2223 const char* search_string = "CPU";
2224 #elif defined(PPC64)
2225 const char* search_string = "cpu";
2226 #elif defined(S390)
2227 const char* search_string = "processor";
2228 #elif defined(SPARC)
2229 const char* search_string = "cpu";
2230 #else
2231 const char* search_string = "Processor";
2232 #endif
2233
2234 // Parses the cpuinfo file for string representing the model name.
2235 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2236 FILE* fp = fopen("/proc/cpuinfo", "r");
2237 if (fp != NULL) {
2238 while (!feof(fp)) {
2239 char buf[256];
2240 if (fgets(buf, sizeof(buf), fp)) {
2241 char* start = strstr(buf, search_string);
2242 if (start != NULL) {
2243 char *ptr = start + strlen(search_string);
2244 char *end = buf + strlen(buf);
2245 while (ptr != end) {
2246 // skip whitespace and colon for the rest of the name.
2247 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2248 break;
2249 }
2250 ptr++;
2251 }
2252 if (ptr != end) {
2253 // reasonable string, get rid of newline and keep the rest
2254 char* nl = strchr(buf, '\n');
2255 if (nl != NULL) *nl = '\0';
2256 strncpy(cpuinfo, ptr, length);
2257 fclose(fp);
2258 return;
2259 }
2260 }
2261 }
2262 }
2263 fclose(fp);
2264 }
2265 // cpuinfo not found or parsing failed, just print generic string. The entire
2266 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2267 #if defined(AARCH64)
2268 strncpy(cpuinfo, "AArch64", length);
2269 #elif defined(AMD64)
2270 strncpy(cpuinfo, "x86_64", length);
2271 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2272 strncpy(cpuinfo, "ARM", length);
2273 #elif defined(IA32)
2274 strncpy(cpuinfo, "x86_32", length);
2275 #elif defined(IA64)
2276 strncpy(cpuinfo, "IA64", length);
2277 #elif defined(PPC)
2278 strncpy(cpuinfo, "PPC64", length);
2279 #elif defined(S390)
2280 strncpy(cpuinfo, "S390", length);
2281 #elif defined(SPARC)
2282 strncpy(cpuinfo, "sparcv9", length);
2283 #elif defined(ZERO_LIBARCH)
2284 strncpy(cpuinfo, ZERO_LIBARCH, length);
2285 #else
2286 strncpy(cpuinfo, "unknown", length);
2287 #endif
2288 }
2289
2290 static void print_signal_handler(outputStream* st, int sig,
2291 char* buf, size_t buflen);
2292
2293 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2294 st->print_cr("Signal Handlers:");
2295 print_signal_handler(st, SIGSEGV, buf, buflen);
2296 print_signal_handler(st, SIGBUS , buf, buflen);
2297 print_signal_handler(st, SIGFPE , buf, buflen);
2298 print_signal_handler(st, SIGPIPE, buf, buflen);
2299 print_signal_handler(st, SIGXFSZ, buf, buflen);
2300 print_signal_handler(st, SIGILL , buf, buflen);
2301 print_signal_handler(st, SR_signum, buf, buflen);
2302 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2303 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2304 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2305 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2306 #if defined(PPC64)
2307 print_signal_handler(st, SIGTRAP, buf, buflen);
2308 #endif
2309 }
2310
2311 static char saved_jvm_path[MAXPATHLEN] = {0};
2312
2313 // Find the full path to the current module, libjvm.so
2314 void os::jvm_path(char *buf, jint buflen) {
2315 // Error checking.
2316 if (buflen < MAXPATHLEN) {
2317 assert(false, "must use a large-enough buffer");
2318 buf[0] = '\0';
2319 return;
2320 }
2321 // Lazy resolve the path to current module.
2322 if (saved_jvm_path[0] != 0) {
2323 strcpy(buf, saved_jvm_path);
2324 return;
2325 }
2326
2327 char dli_fname[MAXPATHLEN];
2328 bool ret = dll_address_to_library_name(
2329 CAST_FROM_FN_PTR(address, os::jvm_path),
2330 dli_fname, sizeof(dli_fname), NULL);
2331 assert(ret, "cannot locate libjvm");
2332 char *rp = NULL;
2333 if (ret && dli_fname[0] != '\0') {
2334 rp = os::Posix::realpath(dli_fname, buf, buflen);
2335 }
2336 if (rp == NULL) {
2337 return;
2338 }
2339
2340 if (Arguments::sun_java_launcher_is_altjvm()) {
2341 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2342 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2343 // If "/jre/lib/" appears at the right place in the string, then
2344 // assume we are installed in a JDK and we're done. Otherwise, check
2345 // for a JAVA_HOME environment variable and fix up the path so it
2346 // looks like libjvm.so is installed there (append a fake suffix
2347 // hotspot/libjvm.so).
2348 const char *p = buf + strlen(buf) - 1;
2349 for (int count = 0; p > buf && count < 5; ++count) {
2350 for (--p; p > buf && *p != '/'; --p)
2351 /* empty */ ;
2352 }
2353
2354 if (strncmp(p, "/jre/lib/", 9) != 0) {
2355 // Look for JAVA_HOME in the environment.
2356 char* java_home_var = ::getenv("JAVA_HOME");
2357 if (java_home_var != NULL && java_home_var[0] != 0) {
2358 char* jrelib_p;
2359 int len;
2360
2361 // Check the current module name "libjvm.so".
2362 p = strrchr(buf, '/');
2363 if (p == NULL) {
2364 return;
2365 }
2366 assert(strstr(p, "/libjvm") == p, "invalid library name");
2367
2368 rp = os::Posix::realpath(java_home_var, buf, buflen);
2369 if (rp == NULL) {
2370 return;
2371 }
2372
2373 // determine if this is a legacy image or modules image
2374 // modules image doesn't have "jre" subdirectory
2375 len = strlen(buf);
2376 assert(len < buflen, "Ran out of buffer room");
2377 jrelib_p = buf + len;
2378 snprintf(jrelib_p, buflen-len, "/jre/lib");
2379 if (0 != access(buf, F_OK)) {
2380 snprintf(jrelib_p, buflen-len, "/lib");
2381 }
2382
2383 if (0 == access(buf, F_OK)) {
2384 // Use current module name "libjvm.so"
2385 len = strlen(buf);
2386 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2387 } else {
2388 // Go back to path of .so
2389 rp = os::Posix::realpath(dli_fname, buf, buflen);
2390 if (rp == NULL) {
2391 return;
2392 }
2393 }
2394 }
2395 }
2396 }
2397
2398 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2399 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2400 }
2401
2402 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2403 // no prefix required, not even "_"
2404 }
2405
2406 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2407 // no suffix required
2408 }
2409
2410 ////////////////////////////////////////////////////////////////////////////////
2411 // sun.misc.Signal support
2412
2413 static volatile jint sigint_count = 0;
2414
2415 static void UserHandler(int sig, void *siginfo, void *context) {
2416 // 4511530 - sem_post is serialized and handled by the manager thread. When
2417 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2418 // don't want to flood the manager thread with sem_post requests.
2419 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2420 return;
2421 }
2422
2423 // Ctrl-C is pressed during error reporting, likely because the error
2424 // handler fails to abort. Let VM die immediately.
2425 if (sig == SIGINT && VMError::is_error_reported()) {
2426 os::die();
2427 }
2428
2429 os::signal_notify(sig);
2430 }
2431
2432 void* os::user_handler() {
2433 return CAST_FROM_FN_PTR(void*, UserHandler);
2434 }
2435
2436 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2437 struct timespec ts;
2438 // Semaphore's are always associated with CLOCK_REALTIME
2439 os::Posix::clock_gettime(CLOCK_REALTIME, &ts);
2440 // see os_posix.cpp for discussion on overflow checking
2441 if (sec >= MAX_SECS) {
2442 ts.tv_sec += MAX_SECS;
2443 ts.tv_nsec = 0;
2444 } else {
2445 ts.tv_sec += sec;
2446 ts.tv_nsec += nsec;
2447 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2448 ts.tv_nsec -= NANOSECS_PER_SEC;
2449 ++ts.tv_sec; // note: this must be <= max_secs
2450 }
2451 }
2452
2453 return ts;
2454 }
2455
2456 extern "C" {
2457 typedef void (*sa_handler_t)(int);
2458 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2459 }
2460
2461 void* os::signal(int signal_number, void* handler) {
2462 struct sigaction sigAct, oldSigAct;
2463
2464 sigfillset(&(sigAct.sa_mask));
2465 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2466 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2467
2468 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2469 // -1 means registration failed
2470 return (void *)-1;
2471 }
2472
2473 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2474 }
2475
2476 void os::signal_raise(int signal_number) {
2477 ::raise(signal_number);
2478 }
2479
2480 // The following code is moved from os.cpp for making this
2481 // code platform specific, which it is by its very nature.
2482
2483 // Will be modified when max signal is changed to be dynamic
2484 int os::sigexitnum_pd() {
2485 return NSIG;
2486 }
2487
2488 // a counter for each possible signal value
2489 static volatile jint pending_signals[NSIG+1] = { 0 };
2490
2491 // Linux(POSIX) specific hand shaking semaphore.
2492 static Semaphore* sig_sem = NULL;
2493 static PosixSemaphore sr_semaphore;
2494
2495 static void jdk_misc_signal_init() {
2496 // Initialize signal structures
2497 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2498
2499 // Initialize signal semaphore
2500 sig_sem = new Semaphore();
2501 }
2502
2503 void os::signal_notify(int sig) {
2504 if (sig_sem != NULL) {
2505 Atomic::inc(&pending_signals[sig]);
2506 sig_sem->signal();
2507 } else {
2508 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2509 // initialization isn't called.
2510 assert(ReduceSignalUsage, "signal semaphore should be created");
2511 }
2512 }
2513
2514 static int check_pending_signals() {
2515 Atomic::store(0, &sigint_count);
2516 for (;;) {
2517 for (int i = 0; i < NSIG + 1; i++) {
2518 jint n = pending_signals[i];
2519 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2520 return i;
2521 }
2522 }
2523 JavaThread *thread = JavaThread::current();
2524 ThreadBlockInVM tbivm(thread);
2525
2526 bool threadIsSuspended;
2527 do {
2528 thread->set_suspend_equivalent();
2529 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2530 sig_sem->wait();
2531
2532 // were we externally suspended while we were waiting?
2533 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2534 if (threadIsSuspended) {
2535 // The semaphore has been incremented, but while we were waiting
2536 // another thread suspended us. We don't want to continue running
2537 // while suspended because that would surprise the thread that
2538 // suspended us.
2539 sig_sem->signal();
2540
2541 thread->java_suspend_self();
2542 }
2543 } while (threadIsSuspended);
2544 }
2545 }
2546
2547 int os::signal_wait() {
2548 return check_pending_signals();
2549 }
2550
2551 ////////////////////////////////////////////////////////////////////////////////
2552 // Virtual Memory
2553
2554 int os::vm_page_size() {
2555 // Seems redundant as all get out
2556 assert(os::Linux::page_size() != -1, "must call os::init");
2557 return os::Linux::page_size();
2558 }
2559
2560 // Solaris allocates memory by pages.
2561 int os::vm_allocation_granularity() {
2562 assert(os::Linux::page_size() != -1, "must call os::init");
2563 return os::Linux::page_size();
2564 }
2565
2566 // Rationale behind this function:
2567 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2568 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2569 // samples for JITted code. Here we create private executable mapping over the code cache
2570 // and then we can use standard (well, almost, as mapping can change) way to provide
2571 // info for the reporting script by storing timestamp and location of symbol
2572 void linux_wrap_code(char* base, size_t size) {
2573 static volatile jint cnt = 0;
2574
2575 if (!UseOprofile) {
2576 return;
2577 }
2578
2579 char buf[PATH_MAX+1];
2580 int num = Atomic::add(1, &cnt);
2581
2582 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2583 os::get_temp_directory(), os::current_process_id(), num);
2584 unlink(buf);
2585
2586 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2587
2588 if (fd != -1) {
2589 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2590 if (rv != (off_t)-1) {
2591 if (::write(fd, "", 1) == 1) {
2592 mmap(base, size,
2593 PROT_READ|PROT_WRITE|PROT_EXEC,
2594 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2595 }
2596 }
2597 ::close(fd);
2598 unlink(buf);
2599 }
2600 }
2601
2602 static bool recoverable_mmap_error(int err) {
2603 // See if the error is one we can let the caller handle. This
2604 // list of errno values comes from JBS-6843484. I can't find a
2605 // Linux man page that documents this specific set of errno
2606 // values so while this list currently matches Solaris, it may
2607 // change as we gain experience with this failure mode.
2608 switch (err) {
2609 case EBADF:
2610 case EINVAL:
2611 case ENOTSUP:
2612 // let the caller deal with these errors
2613 return true;
2614
2615 default:
2616 // Any remaining errors on this OS can cause our reserved mapping
2617 // to be lost. That can cause confusion where different data
2618 // structures think they have the same memory mapped. The worst
2619 // scenario is if both the VM and a library think they have the
2620 // same memory mapped.
2621 return false;
2622 }
2623 }
2624
2625 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2626 int err) {
2627 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2628 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2629 os::strerror(err), err);
2630 }
2631
2632 static void warn_fail_commit_memory(char* addr, size_t size,
2633 size_t alignment_hint, bool exec,
2634 int err) {
2635 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2636 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2637 alignment_hint, exec, os::strerror(err), err);
2638 }
2639
2640 // NOTE: Linux kernel does not really reserve the pages for us.
2641 // All it does is to check if there are enough free pages
2642 // left at the time of mmap(). This could be a potential
2643 // problem.
2644 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2645 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2646 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2647 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2648 if (res != (uintptr_t) MAP_FAILED) {
2649 if (UseNUMAInterleaving) {
2650 numa_make_global(addr, size);
2651 }
2652 return 0;
2653 }
2654
2655 int err = errno; // save errno from mmap() call above
2656
2657 if (!recoverable_mmap_error(err)) {
2658 warn_fail_commit_memory(addr, size, exec, err);
2659 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2660 }
2661
2662 return err;
2663 }
2664
2665 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2666 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2667 }
2668
2669 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2670 const char* mesg) {
2671 assert(mesg != NULL, "mesg must be specified");
2672 int err = os::Linux::commit_memory_impl(addr, size, exec);
2673 if (err != 0) {
2674 // the caller wants all commit errors to exit with the specified mesg:
2675 warn_fail_commit_memory(addr, size, exec, err);
2676 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2677 }
2678 }
2679
2680 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2681 #ifndef MAP_HUGETLB
2682 #define MAP_HUGETLB 0x40000
2683 #endif
2684
2685 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2686 #ifndef MADV_HUGEPAGE
2687 #define MADV_HUGEPAGE 14
2688 #endif
2689
2690 int os::Linux::commit_memory_impl(char* addr, size_t size,
2691 size_t alignment_hint, bool exec) {
2692 int err = os::Linux::commit_memory_impl(addr, size, exec);
2693 if (err == 0) {
2694 realign_memory(addr, size, alignment_hint);
2695 }
2696 return err;
2697 }
2698
2699 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2700 bool exec) {
2701 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2702 }
2703
2704 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2705 size_t alignment_hint, bool exec,
2706 const char* mesg) {
2707 assert(mesg != NULL, "mesg must be specified");
2708 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2709 if (err != 0) {
2710 // the caller wants all commit errors to exit with the specified mesg:
2711 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2712 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2713 }
2714 }
2715
2716 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2717 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2718 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2719 // be supported or the memory may already be backed by huge pages.
2720 ::madvise(addr, bytes, MADV_HUGEPAGE);
2721 }
2722 }
2723
2724 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2725 // This method works by doing an mmap over an existing mmaping and effectively discarding
2726 // the existing pages. However it won't work for SHM-based large pages that cannot be
2727 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2728 // small pages on top of the SHM segment. This method always works for small pages, so we
2729 // allow that in any case.
2730 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2731 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2732 }
2733 }
2734
2735 void os::numa_make_global(char *addr, size_t bytes) {
2736 Linux::numa_interleave_memory(addr, bytes);
2737 }
2738
2739 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2740 // bind policy to MPOL_PREFERRED for the current thread.
2741 #define USE_MPOL_PREFERRED 0
2742
2743 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2744 // To make NUMA and large pages more robust when both enabled, we need to ease
2745 // the requirements on where the memory should be allocated. MPOL_BIND is the
2746 // default policy and it will force memory to be allocated on the specified
2747 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2748 // the specified node, but will not force it. Using this policy will prevent
2749 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2750 // free large pages.
2751 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2752 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2753 }
2754
2755 bool os::numa_topology_changed() { return false; }
2756
2757 size_t os::numa_get_groups_num() {
2758 // Return just the number of nodes in which it's possible to allocate memory
2759 // (in numa terminology, configured nodes).
2760 return Linux::numa_num_configured_nodes();
2761 }
2762
2763 int os::numa_get_group_id() {
2764 int cpu_id = Linux::sched_getcpu();
2765 if (cpu_id != -1) {
2766 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2767 if (lgrp_id != -1) {
2768 return lgrp_id;
2769 }
2770 }
2771 return 0;
2772 }
2773
2774 int os::Linux::get_existing_num_nodes() {
2775 int node;
2776 int highest_node_number = Linux::numa_max_node();
2777 int num_nodes = 0;
2778
2779 // Get the total number of nodes in the system including nodes without memory.
2780 for (node = 0; node <= highest_node_number; node++) {
2781 if (isnode_in_existing_nodes(node)) {
2782 num_nodes++;
2783 }
2784 }
2785 return num_nodes;
2786 }
2787
2788 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2789 int highest_node_number = Linux::numa_max_node();
2790 size_t i = 0;
2791
2792 // Map all node ids in which it is possible to allocate memory. Also nodes are
2793 // not always consecutively available, i.e. available from 0 to the highest
2794 // node number. If the nodes have been bound explicitly using numactl membind,
2795 // then allocate memory from those nodes only.
2796 for (int node = 0; node <= highest_node_number; node++) {
2797 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2798 ids[i++] = node;
2799 }
2800 }
2801 return i;
2802 }
2803
2804 bool os::get_page_info(char *start, page_info* info) {
2805 return false;
2806 }
2807
2808 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2809 page_info* page_found) {
2810 return end;
2811 }
2812
2813
2814 int os::Linux::sched_getcpu_syscall(void) {
2815 unsigned int cpu = 0;
2816 int retval = -1;
2817
2818 #if defined(IA32)
2819 #ifndef SYS_getcpu
2820 #define SYS_getcpu 318
2821 #endif
2822 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2823 #elif defined(AMD64)
2824 // Unfortunately we have to bring all these macros here from vsyscall.h
2825 // to be able to compile on old linuxes.
2826 #define __NR_vgetcpu 2
2827 #define VSYSCALL_START (-10UL << 20)
2828 #define VSYSCALL_SIZE 1024
2829 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2830 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2831 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2832 retval = vgetcpu(&cpu, NULL, NULL);
2833 #endif
2834
2835 return (retval == -1) ? retval : cpu;
2836 }
2837
2838 void os::Linux::sched_getcpu_init() {
2839 // sched_getcpu() should be in libc.
2840 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2841 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2842
2843 // If it's not, try a direct syscall.
2844 if (sched_getcpu() == -1) {
2845 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2846 (void*)&sched_getcpu_syscall));
2847 }
2848
2849 if (sched_getcpu() == -1) {
2850 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2851 }
2852 }
2853
2854 // Something to do with the numa-aware allocator needs these symbols
2855 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2856 extern "C" JNIEXPORT void numa_error(char *where) { }
2857
2858 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2859 // load symbol from base version instead.
2860 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2861 void *f = dlvsym(handle, name, "libnuma_1.1");
2862 if (f == NULL) {
2863 f = dlsym(handle, name);
2864 }
2865 return f;
2866 }
2867
2868 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2869 // Return NULL if the symbol is not defined in this particular version.
2870 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2871 return dlvsym(handle, name, "libnuma_1.2");
2872 }
2873
2874 bool os::Linux::libnuma_init() {
2875 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2876 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2877 if (handle != NULL) {
2878 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2879 libnuma_dlsym(handle, "numa_node_to_cpus")));
2880 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2881 libnuma_dlsym(handle, "numa_max_node")));
2882 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2883 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2884 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2885 libnuma_dlsym(handle, "numa_available")));
2886 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2887 libnuma_dlsym(handle, "numa_tonode_memory")));
2888 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2889 libnuma_dlsym(handle, "numa_interleave_memory")));
2890 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2891 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2892 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2893 libnuma_dlsym(handle, "numa_set_bind_policy")));
2894 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2895 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2896 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2897 libnuma_dlsym(handle, "numa_distance")));
2898 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2899 libnuma_v2_dlsym(handle, "numa_get_membind")));
2900
2901 if (numa_available() != -1) {
2902 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2903 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2904 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2905 // Create an index -> node mapping, since nodes are not always consecutive
2906 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2907 rebuild_nindex_to_node_map();
2908 // Create a cpu -> node mapping
2909 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2910 rebuild_cpu_to_node_map();
2911 return true;
2912 }
2913 }
2914 }
2915 return false;
2916 }
2917
2918 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2919 // Creating guard page is very expensive. Java thread has HotSpot
2920 // guard pages, only enable glibc guard page for non-Java threads.
2921 // (Remember: compiler thread is a Java thread, too!)
2922 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2923 }
2924
2925 void os::Linux::rebuild_nindex_to_node_map() {
2926 int highest_node_number = Linux::numa_max_node();
2927
2928 nindex_to_node()->clear();
2929 for (int node = 0; node <= highest_node_number; node++) {
2930 if (Linux::isnode_in_existing_nodes(node)) {
2931 nindex_to_node()->append(node);
2932 }
2933 }
2934 }
2935
2936 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2937 // The table is later used in get_node_by_cpu().
2938 void os::Linux::rebuild_cpu_to_node_map() {
2939 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2940 // in libnuma (possible values are starting from 16,
2941 // and continuing up with every other power of 2, but less
2942 // than the maximum number of CPUs supported by kernel), and
2943 // is a subject to change (in libnuma version 2 the requirements
2944 // are more reasonable) we'll just hardcode the number they use
2945 // in the library.
2946 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2947
2948 size_t cpu_num = processor_count();
2949 size_t cpu_map_size = NCPUS / BitsPerCLong;
2950 size_t cpu_map_valid_size =
2951 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2952
2953 cpu_to_node()->clear();
2954 cpu_to_node()->at_grow(cpu_num - 1);
2955
2956 size_t node_num = get_existing_num_nodes();
2957
2958 int distance = 0;
2959 int closest_distance = INT_MAX;
2960 int closest_node = 0;
2961 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2962 for (size_t i = 0; i < node_num; i++) {
2963 // Check if node is configured (not a memory-less node). If it is not, find
2964 // the closest configured node. Check also if node is bound, i.e. it's allowed
2965 // to allocate memory from the node. If it's not allowed, map cpus in that node
2966 // to the closest node from which memory allocation is allowed.
2967 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
2968 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
2969 closest_distance = INT_MAX;
2970 // Check distance from all remaining nodes in the system. Ignore distance
2971 // from itself, from another non-configured node, and from another non-bound
2972 // node.
2973 for (size_t m = 0; m < node_num; m++) {
2974 if (m != i &&
2975 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
2976 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
2977 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2978 // If a closest node is found, update. There is always at least one
2979 // configured and bound node in the system so there is always at least
2980 // one node close.
2981 if (distance != 0 && distance < closest_distance) {
2982 closest_distance = distance;
2983 closest_node = nindex_to_node()->at(m);
2984 }
2985 }
2986 }
2987 } else {
2988 // Current node is already a configured node.
2989 closest_node = nindex_to_node()->at(i);
2990 }
2991
2992 // Get cpus from the original node and map them to the closest node. If node
2993 // is a configured node (not a memory-less node), then original node and
2994 // closest node are the same.
2995 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2996 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2997 if (cpu_map[j] != 0) {
2998 for (size_t k = 0; k < BitsPerCLong; k++) {
2999 if (cpu_map[j] & (1UL << k)) {
3000 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3001 }
3002 }
3003 }
3004 }
3005 }
3006 }
3007 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3008 }
3009
3010 int os::Linux::get_node_by_cpu(int cpu_id) {
3011 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3012 return cpu_to_node()->at(cpu_id);
3013 }
3014 return -1;
3015 }
3016
3017 GrowableArray<int>* os::Linux::_cpu_to_node;
3018 GrowableArray<int>* os::Linux::_nindex_to_node;
3019 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3020 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3021 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3022 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3023 os::Linux::numa_available_func_t os::Linux::_numa_available;
3024 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3025 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3026 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3027 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3028 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3029 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3030 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3031 unsigned long* os::Linux::_numa_all_nodes;
3032 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3033 struct bitmask* os::Linux::_numa_nodes_ptr;
3034
3035 bool os::pd_uncommit_memory(char* addr, size_t size) {
3036 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3037 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3038 return res != (uintptr_t) MAP_FAILED;
3039 }
3040
3041 static address get_stack_commited_bottom(address bottom, size_t size) {
3042 address nbot = bottom;
3043 address ntop = bottom + size;
3044
3045 size_t page_sz = os::vm_page_size();
3046 unsigned pages = size / page_sz;
3047
3048 unsigned char vec[1];
3049 unsigned imin = 1, imax = pages + 1, imid;
3050 int mincore_return_value = 0;
3051
3052 assert(imin <= imax, "Unexpected page size");
3053
3054 while (imin < imax) {
3055 imid = (imax + imin) / 2;
3056 nbot = ntop - (imid * page_sz);
3057
3058 // Use a trick with mincore to check whether the page is mapped or not.
3059 // mincore sets vec to 1 if page resides in memory and to 0 if page
3060 // is swapped output but if page we are asking for is unmapped
3061 // it returns -1,ENOMEM
3062 mincore_return_value = mincore(nbot, page_sz, vec);
3063
3064 if (mincore_return_value == -1) {
3065 // Page is not mapped go up
3066 // to find first mapped page
3067 if (errno != EAGAIN) {
3068 assert(errno == ENOMEM, "Unexpected mincore errno");
3069 imax = imid;
3070 }
3071 } else {
3072 // Page is mapped go down
3073 // to find first not mapped page
3074 imin = imid + 1;
3075 }
3076 }
3077
3078 nbot = nbot + page_sz;
3079
3080 // Adjust stack bottom one page up if last checked page is not mapped
3081 if (mincore_return_value == -1) {
3082 nbot = nbot + page_sz;
3083 }
3084
3085 return nbot;
3086 }
3087
3088 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3089 int mincore_return_value;
3090 const size_t stripe = 1024; // query this many pages each time
3091 unsigned char vec[stripe + 1];
3092 // set a guard
3093 vec[stripe] = 'X';
3094
3095 const size_t page_sz = os::vm_page_size();
3096 size_t pages = size / page_sz;
3097
3098 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3099 assert(is_aligned(size, page_sz), "Size must be page aligned");
3100
3101 committed_start = NULL;
3102
3103 int loops = (pages + stripe - 1) / stripe;
3104 int committed_pages = 0;
3105 address loop_base = start;
3106 bool found_range = false;
3107
3108 for (int index = 0; index < loops && !found_range; index ++) {
3109 assert(pages > 0, "Nothing to do");
3110 int pages_to_query = (pages >= stripe) ? stripe : pages;
3111 pages -= pages_to_query;
3112
3113 // Get stable read
3114 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3115
3116 // During shutdown, some memory goes away without properly notifying NMT,
3117 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3118 // Bailout and return as not committed for now.
3119 if (mincore_return_value == -1 && errno == ENOMEM) {
3120 return false;
3121 }
3122
3123 assert(vec[stripe] == 'X', "overflow guard");
3124 assert(mincore_return_value == 0, "Range must be valid");
3125 // Process this stripe
3126 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3127 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3128 // End of current contiguous region
3129 if (committed_start != NULL) {
3130 found_range = true;
3131 break;
3132 }
3133 } else { // committed
3134 // Start of region
3135 if (committed_start == NULL) {
3136 committed_start = loop_base + page_sz * vecIdx;
3137 }
3138 committed_pages ++;
3139 }
3140 }
3141
3142 loop_base += pages_to_query * page_sz;
3143 }
3144
3145 if (committed_start != NULL) {
3146 assert(committed_pages > 0, "Must have committed region");
3147 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3148 assert(committed_start >= start && committed_start < start + size, "Out of range");
3149 committed_size = page_sz * committed_pages;
3150 return true;
3151 } else {
3152 assert(committed_pages == 0, "Should not have committed region");
3153 return false;
3154 }
3155 }
3156
3157
3158 // Linux uses a growable mapping for the stack, and if the mapping for
3159 // the stack guard pages is not removed when we detach a thread the
3160 // stack cannot grow beyond the pages where the stack guard was
3161 // mapped. If at some point later in the process the stack expands to
3162 // that point, the Linux kernel cannot expand the stack any further
3163 // because the guard pages are in the way, and a segfault occurs.
3164 //
3165 // However, it's essential not to split the stack region by unmapping
3166 // a region (leaving a hole) that's already part of the stack mapping,
3167 // so if the stack mapping has already grown beyond the guard pages at
3168 // the time we create them, we have to truncate the stack mapping.
3169 // So, we need to know the extent of the stack mapping when
3170 // create_stack_guard_pages() is called.
3171
3172 // We only need this for stacks that are growable: at the time of
3173 // writing thread stacks don't use growable mappings (i.e. those
3174 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3175 // only applies to the main thread.
3176
3177 // If the (growable) stack mapping already extends beyond the point
3178 // where we're going to put our guard pages, truncate the mapping at
3179 // that point by munmap()ping it. This ensures that when we later
3180 // munmap() the guard pages we don't leave a hole in the stack
3181 // mapping. This only affects the main/primordial thread
3182
3183 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3184 if (os::is_primordial_thread()) {
3185 // As we manually grow stack up to bottom inside create_attached_thread(),
3186 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3187 // we don't need to do anything special.
3188 // Check it first, before calling heavy function.
3189 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3190 unsigned char vec[1];
3191
3192 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3193 // Fallback to slow path on all errors, including EAGAIN
3194 stack_extent = (uintptr_t) get_stack_commited_bottom(
3195 os::Linux::initial_thread_stack_bottom(),
3196 (size_t)addr - stack_extent);
3197 }
3198
3199 if (stack_extent < (uintptr_t)addr) {
3200 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3201 }
3202 }
3203
3204 return os::commit_memory(addr, size, !ExecMem);
3205 }
3206
3207 // If this is a growable mapping, remove the guard pages entirely by
3208 // munmap()ping them. If not, just call uncommit_memory(). This only
3209 // affects the main/primordial thread, but guard against future OS changes.
3210 // It's safe to always unmap guard pages for primordial thread because we
3211 // always place it right after end of the mapped region.
3212
3213 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3214 uintptr_t stack_extent, stack_base;
3215
3216 if (os::is_primordial_thread()) {
3217 return ::munmap(addr, size) == 0;
3218 }
3219
3220 return os::uncommit_memory(addr, size);
3221 }
3222
3223 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3224 // at 'requested_addr'. If there are existing memory mappings at the same
3225 // location, however, they will be overwritten. If 'fixed' is false,
3226 // 'requested_addr' is only treated as a hint, the return value may or
3227 // may not start from the requested address. Unlike Linux mmap(), this
3228 // function returns NULL to indicate failure.
3229 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3230 char * addr;
3231 int flags;
3232
3233 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3234 if (fixed) {
3235 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3236 flags |= MAP_FIXED;
3237 }
3238
3239 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3240 // touch an uncommitted page. Otherwise, the read/write might
3241 // succeed if we have enough swap space to back the physical page.
3242 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3243 flags, -1, 0);
3244
3245 return addr == MAP_FAILED ? NULL : addr;
3246 }
3247
3248 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3249 // (req_addr != NULL) or with a given alignment.
3250 // - bytes shall be a multiple of alignment.
3251 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3252 // - alignment sets the alignment at which memory shall be allocated.
3253 // It must be a multiple of allocation granularity.
3254 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3255 // req_addr or NULL.
3256 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3257
3258 size_t extra_size = bytes;
3259 if (req_addr == NULL && alignment > 0) {
3260 extra_size += alignment;
3261 }
3262
3263 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3264 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3265 -1, 0);
3266 if (start == MAP_FAILED) {
3267 start = NULL;
3268 } else {
3269 if (req_addr != NULL) {
3270 if (start != req_addr) {
3271 ::munmap(start, extra_size);
3272 start = NULL;
3273 }
3274 } else {
3275 char* const start_aligned = align_up(start, alignment);
3276 char* const end_aligned = start_aligned + bytes;
3277 char* const end = start + extra_size;
3278 if (start_aligned > start) {
3279 ::munmap(start, start_aligned - start);
3280 }
3281 if (end_aligned < end) {
3282 ::munmap(end_aligned, end - end_aligned);
3283 }
3284 start = start_aligned;
3285 }
3286 }
3287 return start;
3288 }
3289
3290 static int anon_munmap(char * addr, size_t size) {
3291 return ::munmap(addr, size) == 0;
3292 }
3293
3294 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3295 size_t alignment_hint) {
3296 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3297 }
3298
3299 bool os::pd_release_memory(char* addr, size_t size) {
3300 return anon_munmap(addr, size);
3301 }
3302
3303 static bool linux_mprotect(char* addr, size_t size, int prot) {
3304 // Linux wants the mprotect address argument to be page aligned.
3305 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3306
3307 // According to SUSv3, mprotect() should only be used with mappings
3308 // established by mmap(), and mmap() always maps whole pages. Unaligned
3309 // 'addr' likely indicates problem in the VM (e.g. trying to change
3310 // protection of malloc'ed or statically allocated memory). Check the
3311 // caller if you hit this assert.
3312 assert(addr == bottom, "sanity check");
3313
3314 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3315 return ::mprotect(bottom, size, prot) == 0;
3316 }
3317
3318 // Set protections specified
3319 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3320 bool is_committed) {
3321 unsigned int p = 0;
3322 switch (prot) {
3323 case MEM_PROT_NONE: p = PROT_NONE; break;
3324 case MEM_PROT_READ: p = PROT_READ; break;
3325 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3326 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3327 default:
3328 ShouldNotReachHere();
3329 }
3330 // is_committed is unused.
3331 return linux_mprotect(addr, bytes, p);
3332 }
3333
3334 bool os::guard_memory(char* addr, size_t size) {
3335 return linux_mprotect(addr, size, PROT_NONE);
3336 }
3337
3338 bool os::unguard_memory(char* addr, size_t size) {
3339 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3340 }
3341
3342 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3343 size_t page_size) {
3344 bool result = false;
3345 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3346 MAP_ANONYMOUS|MAP_PRIVATE,
3347 -1, 0);
3348 if (p != MAP_FAILED) {
3349 void *aligned_p = align_up(p, page_size);
3350
3351 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3352
3353 munmap(p, page_size * 2);
3354 }
3355
3356 if (warn && !result) {
3357 warning("TransparentHugePages is not supported by the operating system.");
3358 }
3359
3360 return result;
3361 }
3362
3363 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3364 bool result = false;
3365 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3366 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3367 -1, 0);
3368
3369 if (p != MAP_FAILED) {
3370 // We don't know if this really is a huge page or not.
3371 FILE *fp = fopen("/proc/self/maps", "r");
3372 if (fp) {
3373 while (!feof(fp)) {
3374 char chars[257];
3375 long x = 0;
3376 if (fgets(chars, sizeof(chars), fp)) {
3377 if (sscanf(chars, "%lx-%*x", &x) == 1
3378 && x == (long)p) {
3379 if (strstr (chars, "hugepage")) {
3380 result = true;
3381 break;
3382 }
3383 }
3384 }
3385 }
3386 fclose(fp);
3387 }
3388 munmap(p, page_size);
3389 }
3390
3391 if (warn && !result) {
3392 warning("HugeTLBFS is not supported by the operating system.");
3393 }
3394
3395 return result;
3396 }
3397
3398 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3399 //
3400 // From the coredump_filter documentation:
3401 //
3402 // - (bit 0) anonymous private memory
3403 // - (bit 1) anonymous shared memory
3404 // - (bit 2) file-backed private memory
3405 // - (bit 3) file-backed shared memory
3406 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3407 // effective only if the bit 2 is cleared)
3408 // - (bit 5) hugetlb private memory
3409 // - (bit 6) hugetlb shared memory
3410 // - (bit 7) dax private memory
3411 // - (bit 8) dax shared memory
3412 //
3413 static void set_coredump_filter(CoredumpFilterBit bit) {
3414 FILE *f;
3415 long cdm;
3416
3417 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3418 return;
3419 }
3420
3421 if (fscanf(f, "%lx", &cdm) != 1) {
3422 fclose(f);
3423 return;
3424 }
3425
3426 long saved_cdm = cdm;
3427 rewind(f);
3428 cdm |= bit;
3429
3430 if (cdm != saved_cdm) {
3431 fprintf(f, "%#lx", cdm);
3432 }
3433
3434 fclose(f);
3435 }
3436
3437 // Large page support
3438
3439 static size_t _large_page_size = 0;
3440
3441 size_t os::Linux::find_large_page_size() {
3442 size_t large_page_size = 0;
3443
3444 // large_page_size on Linux is used to round up heap size. x86 uses either
3445 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3446 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3447 // page as large as 256M.
3448 //
3449 // Here we try to figure out page size by parsing /proc/meminfo and looking
3450 // for a line with the following format:
3451 // Hugepagesize: 2048 kB
3452 //
3453 // If we can't determine the value (e.g. /proc is not mounted, or the text
3454 // format has been changed), we'll use the largest page size supported by
3455 // the processor.
3456
3457 #ifndef ZERO
3458 large_page_size =
3459 AARCH64_ONLY(2 * M)
3460 AMD64_ONLY(2 * M)
3461 ARM32_ONLY(2 * M)
3462 IA32_ONLY(4 * M)
3463 IA64_ONLY(256 * M)
3464 PPC_ONLY(4 * M)
3465 S390_ONLY(1 * M)
3466 SPARC_ONLY(4 * M);
3467 #endif // ZERO
3468
3469 FILE *fp = fopen("/proc/meminfo", "r");
3470 if (fp) {
3471 while (!feof(fp)) {
3472 int x = 0;
3473 char buf[16];
3474 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3475 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3476 large_page_size = x * K;
3477 break;
3478 }
3479 } else {
3480 // skip to next line
3481 for (;;) {
3482 int ch = fgetc(fp);
3483 if (ch == EOF || ch == (int)'\n') break;
3484 }
3485 }
3486 }
3487 fclose(fp);
3488 }
3489
3490 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3491 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3492 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3493 proper_unit_for_byte_size(large_page_size));
3494 }
3495
3496 return large_page_size;
3497 }
3498
3499 size_t os::Linux::setup_large_page_size() {
3500 _large_page_size = Linux::find_large_page_size();
3501 const size_t default_page_size = (size_t)Linux::page_size();
3502 if (_large_page_size > default_page_size) {
3503 _page_sizes[0] = _large_page_size;
3504 _page_sizes[1] = default_page_size;
3505 _page_sizes[2] = 0;
3506 }
3507
3508 return _large_page_size;
3509 }
3510
3511 bool os::Linux::setup_large_page_type(size_t page_size) {
3512 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3513 FLAG_IS_DEFAULT(UseSHM) &&
3514 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3515
3516 // The type of large pages has not been specified by the user.
3517
3518 // Try UseHugeTLBFS and then UseSHM.
3519 UseHugeTLBFS = UseSHM = true;
3520
3521 // Don't try UseTransparentHugePages since there are known
3522 // performance issues with it turned on. This might change in the future.
3523 UseTransparentHugePages = false;
3524 }
3525
3526 if (UseTransparentHugePages) {
3527 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3528 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3529 UseHugeTLBFS = false;
3530 UseSHM = false;
3531 return true;
3532 }
3533 UseTransparentHugePages = false;
3534 }
3535
3536 if (UseHugeTLBFS) {
3537 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3538 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3539 UseSHM = false;
3540 return true;
3541 }
3542 UseHugeTLBFS = false;
3543 }
3544
3545 return UseSHM;
3546 }
3547
3548 void os::large_page_init() {
3549 if (!UseLargePages &&
3550 !UseTransparentHugePages &&
3551 !UseHugeTLBFS &&
3552 !UseSHM) {
3553 // Not using large pages.
3554 return;
3555 }
3556
3557 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3558 // The user explicitly turned off large pages.
3559 // Ignore the rest of the large pages flags.
3560 UseTransparentHugePages = false;
3561 UseHugeTLBFS = false;
3562 UseSHM = false;
3563 return;
3564 }
3565
3566 size_t large_page_size = Linux::setup_large_page_size();
3567 UseLargePages = Linux::setup_large_page_type(large_page_size);
3568
3569 set_coredump_filter(LARGEPAGES_BIT);
3570 }
3571
3572 #ifndef SHM_HUGETLB
3573 #define SHM_HUGETLB 04000
3574 #endif
3575
3576 #define shm_warning_format(format, ...) \
3577 do { \
3578 if (UseLargePages && \
3579 (!FLAG_IS_DEFAULT(UseLargePages) || \
3580 !FLAG_IS_DEFAULT(UseSHM) || \
3581 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3582 warning(format, __VA_ARGS__); \
3583 } \
3584 } while (0)
3585
3586 #define shm_warning(str) shm_warning_format("%s", str)
3587
3588 #define shm_warning_with_errno(str) \
3589 do { \
3590 int err = errno; \
3591 shm_warning_format(str " (error = %d)", err); \
3592 } while (0)
3593
3594 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3595 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3596
3597 if (!is_aligned(alignment, SHMLBA)) {
3598 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3599 return NULL;
3600 }
3601
3602 // To ensure that we get 'alignment' aligned memory from shmat,
3603 // we pre-reserve aligned virtual memory and then attach to that.
3604
3605 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3606 if (pre_reserved_addr == NULL) {
3607 // Couldn't pre-reserve aligned memory.
3608 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3609 return NULL;
3610 }
3611
3612 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3613 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3614
3615 if ((intptr_t)addr == -1) {
3616 int err = errno;
3617 shm_warning_with_errno("Failed to attach shared memory.");
3618
3619 assert(err != EACCES, "Unexpected error");
3620 assert(err != EIDRM, "Unexpected error");
3621 assert(err != EINVAL, "Unexpected error");
3622
3623 // Since we don't know if the kernel unmapped the pre-reserved memory area
3624 // we can't unmap it, since that would potentially unmap memory that was
3625 // mapped from other threads.
3626 return NULL;
3627 }
3628
3629 return addr;
3630 }
3631
3632 static char* shmat_at_address(int shmid, char* req_addr) {
3633 if (!is_aligned(req_addr, SHMLBA)) {
3634 assert(false, "Requested address needs to be SHMLBA aligned");
3635 return NULL;
3636 }
3637
3638 char* addr = (char*)shmat(shmid, req_addr, 0);
3639
3640 if ((intptr_t)addr == -1) {
3641 shm_warning_with_errno("Failed to attach shared memory.");
3642 return NULL;
3643 }
3644
3645 return addr;
3646 }
3647
3648 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3649 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3650 if (req_addr != NULL) {
3651 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3652 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3653 return shmat_at_address(shmid, req_addr);
3654 }
3655
3656 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3657 // return large page size aligned memory addresses when req_addr == NULL.
3658 // However, if the alignment is larger than the large page size, we have
3659 // to manually ensure that the memory returned is 'alignment' aligned.
3660 if (alignment > os::large_page_size()) {
3661 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3662 return shmat_with_alignment(shmid, bytes, alignment);
3663 } else {
3664 return shmat_at_address(shmid, NULL);
3665 }
3666 }
3667
3668 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3669 char* req_addr, bool exec) {
3670 // "exec" is passed in but not used. Creating the shared image for
3671 // the code cache doesn't have an SHM_X executable permission to check.
3672 assert(UseLargePages && UseSHM, "only for SHM large pages");
3673 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3674 assert(is_aligned(req_addr, alignment), "Unaligned address");
3675
3676 if (!is_aligned(bytes, os::large_page_size())) {
3677 return NULL; // Fallback to small pages.
3678 }
3679
3680 // Create a large shared memory region to attach to based on size.
3681 // Currently, size is the total size of the heap.
3682 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3683 if (shmid == -1) {
3684 // Possible reasons for shmget failure:
3685 // 1. shmmax is too small for Java heap.
3686 // > check shmmax value: cat /proc/sys/kernel/shmmax
3687 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3688 // 2. not enough large page memory.
3689 // > check available large pages: cat /proc/meminfo
3690 // > increase amount of large pages:
3691 // echo new_value > /proc/sys/vm/nr_hugepages
3692 // Note 1: different Linux may use different name for this property,
3693 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3694 // Note 2: it's possible there's enough physical memory available but
3695 // they are so fragmented after a long run that they can't
3696 // coalesce into large pages. Try to reserve large pages when
3697 // the system is still "fresh".
3698 shm_warning_with_errno("Failed to reserve shared memory.");
3699 return NULL;
3700 }
3701
3702 // Attach to the region.
3703 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3704
3705 // Remove shmid. If shmat() is successful, the actual shared memory segment
3706 // will be deleted when it's detached by shmdt() or when the process
3707 // terminates. If shmat() is not successful this will remove the shared
3708 // segment immediately.
3709 shmctl(shmid, IPC_RMID, NULL);
3710
3711 return addr;
3712 }
3713
3714 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3715 int error) {
3716 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3717
3718 bool warn_on_failure = UseLargePages &&
3719 (!FLAG_IS_DEFAULT(UseLargePages) ||
3720 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3721 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3722
3723 if (warn_on_failure) {
3724 char msg[128];
3725 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3726 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3727 warning("%s", msg);
3728 }
3729 }
3730
3731 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3732 char* req_addr,
3733 bool exec) {
3734 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3735 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3736 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3737
3738 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3739 char* addr = (char*)::mmap(req_addr, bytes, prot,
3740 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3741 -1, 0);
3742
3743 if (addr == MAP_FAILED) {
3744 warn_on_large_pages_failure(req_addr, bytes, errno);
3745 return NULL;
3746 }
3747
3748 assert(is_aligned(addr, os::large_page_size()), "Must be");
3749
3750 return addr;
3751 }
3752
3753 // Reserve memory using mmap(MAP_HUGETLB).
3754 // - bytes shall be a multiple of alignment.
3755 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3756 // - alignment sets the alignment at which memory shall be allocated.
3757 // It must be a multiple of allocation granularity.
3758 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3759 // req_addr or NULL.
3760 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3761 size_t alignment,
3762 char* req_addr,
3763 bool exec) {
3764 size_t large_page_size = os::large_page_size();
3765 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3766
3767 assert(is_aligned(req_addr, alignment), "Must be");
3768 assert(is_aligned(bytes, alignment), "Must be");
3769
3770 // First reserve - but not commit - the address range in small pages.
3771 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3772
3773 if (start == NULL) {
3774 return NULL;
3775 }
3776
3777 assert(is_aligned(start, alignment), "Must be");
3778
3779 char* end = start + bytes;
3780
3781 // Find the regions of the allocated chunk that can be promoted to large pages.
3782 char* lp_start = align_up(start, large_page_size);
3783 char* lp_end = align_down(end, large_page_size);
3784
3785 size_t lp_bytes = lp_end - lp_start;
3786
3787 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3788
3789 if (lp_bytes == 0) {
3790 // The mapped region doesn't even span the start and the end of a large page.
3791 // Fall back to allocate a non-special area.
3792 ::munmap(start, end - start);
3793 return NULL;
3794 }
3795
3796 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3797
3798 void* result;
3799
3800 // Commit small-paged leading area.
3801 if (start != lp_start) {
3802 result = ::mmap(start, lp_start - start, prot,
3803 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3804 -1, 0);
3805 if (result == MAP_FAILED) {
3806 ::munmap(lp_start, end - lp_start);
3807 return NULL;
3808 }
3809 }
3810
3811 // Commit large-paged area.
3812 result = ::mmap(lp_start, lp_bytes, prot,
3813 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3814 -1, 0);
3815 if (result == MAP_FAILED) {
3816 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3817 // If the mmap above fails, the large pages region will be unmapped and we
3818 // have regions before and after with small pages. Release these regions.
3819 //
3820 // | mapped | unmapped | mapped |
3821 // ^ ^ ^ ^
3822 // start lp_start lp_end end
3823 //
3824 ::munmap(start, lp_start - start);
3825 ::munmap(lp_end, end - lp_end);
3826 return NULL;
3827 }
3828
3829 // Commit small-paged trailing area.
3830 if (lp_end != end) {
3831 result = ::mmap(lp_end, end - lp_end, prot,
3832 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3833 -1, 0);
3834 if (result == MAP_FAILED) {
3835 ::munmap(start, lp_end - start);
3836 return NULL;
3837 }
3838 }
3839
3840 return start;
3841 }
3842
3843 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3844 size_t alignment,
3845 char* req_addr,
3846 bool exec) {
3847 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3848 assert(is_aligned(req_addr, alignment), "Must be");
3849 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3850 assert(is_power_of_2(os::large_page_size()), "Must be");
3851 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3852
3853 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3854 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3855 } else {
3856 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3857 }
3858 }
3859
3860 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3861 char* req_addr, bool exec) {
3862 assert(UseLargePages, "only for large pages");
3863
3864 char* addr;
3865 if (UseSHM) {
3866 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3867 } else {
3868 assert(UseHugeTLBFS, "must be");
3869 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3870 }
3871
3872 if (addr != NULL) {
3873 if (UseNUMAInterleaving) {
3874 numa_make_global(addr, bytes);
3875 }
3876
3877 // The memory is committed
3878 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3879 }
3880
3881 return addr;
3882 }
3883
3884 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3885 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3886 return shmdt(base) == 0;
3887 }
3888
3889 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3890 return pd_release_memory(base, bytes);
3891 }
3892
3893 bool os::release_memory_special(char* base, size_t bytes) {
3894 bool res;
3895 if (MemTracker::tracking_level() > NMT_minimal) {
3896 Tracker tkr(Tracker::release);
3897 res = os::Linux::release_memory_special_impl(base, bytes);
3898 if (res) {
3899 tkr.record((address)base, bytes);
3900 }
3901
3902 } else {
3903 res = os::Linux::release_memory_special_impl(base, bytes);
3904 }
3905 return res;
3906 }
3907
3908 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3909 assert(UseLargePages, "only for large pages");
3910 bool res;
3911
3912 if (UseSHM) {
3913 res = os::Linux::release_memory_special_shm(base, bytes);
3914 } else {
3915 assert(UseHugeTLBFS, "must be");
3916 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3917 }
3918 return res;
3919 }
3920
3921 size_t os::large_page_size() {
3922 return _large_page_size;
3923 }
3924
3925 // With SysV SHM the entire memory region must be allocated as shared
3926 // memory.
3927 // HugeTLBFS allows application to commit large page memory on demand.
3928 // However, when committing memory with HugeTLBFS fails, the region
3929 // that was supposed to be committed will lose the old reservation
3930 // and allow other threads to steal that memory region. Because of this
3931 // behavior we can't commit HugeTLBFS memory.
3932 bool os::can_commit_large_page_memory() {
3933 return UseTransparentHugePages;
3934 }
3935
3936 bool os::can_execute_large_page_memory() {
3937 return UseTransparentHugePages || UseHugeTLBFS;
3938 }
3939
3940 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3941 assert(file_desc >= 0, "file_desc is not valid");
3942 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3943 if (result != NULL) {
3944 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3945 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3946 }
3947 }
3948 return result;
3949 }
3950
3951 // Reserve memory at an arbitrary address, only if that area is
3952 // available (and not reserved for something else).
3953
3954 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3955 const int max_tries = 10;
3956 char* base[max_tries];
3957 size_t size[max_tries];
3958 const size_t gap = 0x000000;
3959
3960 // Assert only that the size is a multiple of the page size, since
3961 // that's all that mmap requires, and since that's all we really know
3962 // about at this low abstraction level. If we need higher alignment,
3963 // we can either pass an alignment to this method or verify alignment
3964 // in one of the methods further up the call chain. See bug 5044738.
3965 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3966
3967 // Repeatedly allocate blocks until the block is allocated at the
3968 // right spot.
3969
3970 // Linux mmap allows caller to pass an address as hint; give it a try first,
3971 // if kernel honors the hint then we can return immediately.
3972 char * addr = anon_mmap(requested_addr, bytes, false);
3973 if (addr == requested_addr) {
3974 return requested_addr;
3975 }
3976
3977 if (addr != NULL) {
3978 // mmap() is successful but it fails to reserve at the requested address
3979 anon_munmap(addr, bytes);
3980 }
3981
3982 int i;
3983 for (i = 0; i < max_tries; ++i) {
3984 base[i] = reserve_memory(bytes);
3985
3986 if (base[i] != NULL) {
3987 // Is this the block we wanted?
3988 if (base[i] == requested_addr) {
3989 size[i] = bytes;
3990 break;
3991 }
3992
3993 // Does this overlap the block we wanted? Give back the overlapped
3994 // parts and try again.
3995
3996 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3997 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3998 unmap_memory(base[i], top_overlap);
3999 base[i] += top_overlap;
4000 size[i] = bytes - top_overlap;
4001 } else {
4002 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4003 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4004 unmap_memory(requested_addr, bottom_overlap);
4005 size[i] = bytes - bottom_overlap;
4006 } else {
4007 size[i] = bytes;
4008 }
4009 }
4010 }
4011 }
4012
4013 // Give back the unused reserved pieces.
4014
4015 for (int j = 0; j < i; ++j) {
4016 if (base[j] != NULL) {
4017 unmap_memory(base[j], size[j]);
4018 }
4019 }
4020
4021 if (i < max_tries) {
4022 return requested_addr;
4023 } else {
4024 return NULL;
4025 }
4026 }
4027
4028 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4029 return ::read(fd, buf, nBytes);
4030 }
4031
4032 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4033 return ::pread(fd, buf, nBytes, offset);
4034 }
4035
4036 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4037 void os::infinite_sleep() {
4038 while (true) { // sleep forever ...
4039 ::sleep(100); // ... 100 seconds at a time
4040 }
4041 }
4042
4043 // Used to convert frequent JVM_Yield() to nops
4044 bool os::dont_yield() {
4045 return DontYieldALot;
4046 }
4047
4048 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4049 // actually give up the CPU. Since skip buddy (v2.6.28):
4050 //
4051 // * Sets the yielding task as skip buddy for current CPU's run queue.
4052 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4053 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4054 //
4055 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4056 // getting re-scheduled immediately.
4057 //
4058 void os::naked_yield() {
4059 sched_yield();
4060 }
4061
4062 ////////////////////////////////////////////////////////////////////////////////
4063 // thread priority support
4064
4065 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4066 // only supports dynamic priority, static priority must be zero. For real-time
4067 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4068 // However, for large multi-threaded applications, SCHED_RR is not only slower
4069 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4070 // of 5 runs - Sep 2005).
4071 //
4072 // The following code actually changes the niceness of kernel-thread/LWP. It
4073 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4074 // not the entire user process, and user level threads are 1:1 mapped to kernel
4075 // threads. It has always been the case, but could change in the future. For
4076 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4077 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4078
4079 int os::java_to_os_priority[CriticalPriority + 1] = {
4080 19, // 0 Entry should never be used
4081
4082 4, // 1 MinPriority
4083 3, // 2
4084 2, // 3
4085
4086 1, // 4
4087 0, // 5 NormPriority
4088 -1, // 6
4089
4090 -2, // 7
4091 -3, // 8
4092 -4, // 9 NearMaxPriority
4093
4094 -5, // 10 MaxPriority
4095
4096 -5 // 11 CriticalPriority
4097 };
4098
4099 static int prio_init() {
4100 if (ThreadPriorityPolicy == 1) {
4101 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4102 // if effective uid is not root. Perhaps, a more elegant way of doing
4103 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4104 if (geteuid() != 0) {
4105 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4106 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4107 }
4108 ThreadPriorityPolicy = 0;
4109 }
4110 }
4111 if (UseCriticalJavaThreadPriority) {
4112 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4113 }
4114 return 0;
4115 }
4116
4117 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4118 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4119
4120 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4121 return (ret == 0) ? OS_OK : OS_ERR;
4122 }
4123
4124 OSReturn os::get_native_priority(const Thread* const thread,
4125 int *priority_ptr) {
4126 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4127 *priority_ptr = java_to_os_priority[NormPriority];
4128 return OS_OK;
4129 }
4130
4131 errno = 0;
4132 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4133 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4134 }
4135
4136 ////////////////////////////////////////////////////////////////////////////////
4137 // suspend/resume support
4138
4139 // The low-level signal-based suspend/resume support is a remnant from the
4140 // old VM-suspension that used to be for java-suspension, safepoints etc,
4141 // within hotspot. Currently used by JFR's OSThreadSampler
4142 //
4143 // The remaining code is greatly simplified from the more general suspension
4144 // code that used to be used.
4145 //
4146 // The protocol is quite simple:
4147 // - suspend:
4148 // - sends a signal to the target thread
4149 // - polls the suspend state of the osthread using a yield loop
4150 // - target thread signal handler (SR_handler) sets suspend state
4151 // and blocks in sigsuspend until continued
4152 // - resume:
4153 // - sets target osthread state to continue
4154 // - sends signal to end the sigsuspend loop in the SR_handler
4155 //
4156 // Note that the SR_lock plays no role in this suspend/resume protocol,
4157 // but is checked for NULL in SR_handler as a thread termination indicator.
4158 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4159 //
4160 // Note that resume_clear_context() and suspend_save_context() are needed
4161 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4162 // which in part is used by:
4163 // - Forte Analyzer: AsyncGetCallTrace()
4164 // - StackBanging: get_frame_at_stack_banging_point()
4165
4166 static void resume_clear_context(OSThread *osthread) {
4167 osthread->set_ucontext(NULL);
4168 osthread->set_siginfo(NULL);
4169 }
4170
4171 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4172 ucontext_t* context) {
4173 osthread->set_ucontext(context);
4174 osthread->set_siginfo(siginfo);
4175 }
4176
4177 // Handler function invoked when a thread's execution is suspended or
4178 // resumed. We have to be careful that only async-safe functions are
4179 // called here (Note: most pthread functions are not async safe and
4180 // should be avoided.)
4181 //
4182 // Note: sigwait() is a more natural fit than sigsuspend() from an
4183 // interface point of view, but sigwait() prevents the signal hander
4184 // from being run. libpthread would get very confused by not having
4185 // its signal handlers run and prevents sigwait()'s use with the
4186 // mutex granting granting signal.
4187 //
4188 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4189 //
4190 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4191 // Save and restore errno to avoid confusing native code with EINTR
4192 // after sigsuspend.
4193 int old_errno = errno;
4194
4195 Thread* thread = Thread::current_or_null_safe();
4196 assert(thread != NULL, "Missing current thread in SR_handler");
4197
4198 // On some systems we have seen signal delivery get "stuck" until the signal
4199 // mask is changed as part of thread termination. Check that the current thread
4200 // has not already terminated (via SR_lock()) - else the following assertion
4201 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4202 // destructor has completed.
4203
4204 if (thread->SR_lock() == NULL) {
4205 return;
4206 }
4207
4208 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4209
4210 OSThread* osthread = thread->osthread();
4211
4212 os::SuspendResume::State current = osthread->sr.state();
4213 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4214 suspend_save_context(osthread, siginfo, context);
4215
4216 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4217 os::SuspendResume::State state = osthread->sr.suspended();
4218 if (state == os::SuspendResume::SR_SUSPENDED) {
4219 sigset_t suspend_set; // signals for sigsuspend()
4220 sigemptyset(&suspend_set);
4221 // get current set of blocked signals and unblock resume signal
4222 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4223 sigdelset(&suspend_set, SR_signum);
4224
4225 sr_semaphore.signal();
4226 // wait here until we are resumed
4227 while (1) {
4228 sigsuspend(&suspend_set);
4229
4230 os::SuspendResume::State result = osthread->sr.running();
4231 if (result == os::SuspendResume::SR_RUNNING) {
4232 sr_semaphore.signal();
4233 break;
4234 }
4235 }
4236
4237 } else if (state == os::SuspendResume::SR_RUNNING) {
4238 // request was cancelled, continue
4239 } else {
4240 ShouldNotReachHere();
4241 }
4242
4243 resume_clear_context(osthread);
4244 } else if (current == os::SuspendResume::SR_RUNNING) {
4245 // request was cancelled, continue
4246 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4247 // ignore
4248 } else {
4249 // ignore
4250 }
4251
4252 errno = old_errno;
4253 }
4254
4255 static int SR_initialize() {
4256 struct sigaction act;
4257 char *s;
4258
4259 // Get signal number to use for suspend/resume
4260 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4261 int sig = ::strtol(s, 0, 10);
4262 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4263 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4264 SR_signum = sig;
4265 } else {
4266 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4267 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4268 }
4269 }
4270
4271 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4272 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4273
4274 sigemptyset(&SR_sigset);
4275 sigaddset(&SR_sigset, SR_signum);
4276
4277 // Set up signal handler for suspend/resume
4278 act.sa_flags = SA_RESTART|SA_SIGINFO;
4279 act.sa_handler = (void (*)(int)) SR_handler;
4280
4281 // SR_signum is blocked by default.
4282 // 4528190 - We also need to block pthread restart signal (32 on all
4283 // supported Linux platforms). Note that LinuxThreads need to block
4284 // this signal for all threads to work properly. So we don't have
4285 // to use hard-coded signal number when setting up the mask.
4286 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4287
4288 if (sigaction(SR_signum, &act, 0) == -1) {
4289 return -1;
4290 }
4291
4292 // Save signal flag
4293 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4294 return 0;
4295 }
4296
4297 static int sr_notify(OSThread* osthread) {
4298 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4299 assert_status(status == 0, status, "pthread_kill");
4300 return status;
4301 }
4302
4303 // "Randomly" selected value for how long we want to spin
4304 // before bailing out on suspending a thread, also how often
4305 // we send a signal to a thread we want to resume
4306 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4307 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4308
4309 // returns true on success and false on error - really an error is fatal
4310 // but this seems the normal response to library errors
4311 static bool do_suspend(OSThread* osthread) {
4312 assert(osthread->sr.is_running(), "thread should be running");
4313 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4314
4315 // mark as suspended and send signal
4316 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4317 // failed to switch, state wasn't running?
4318 ShouldNotReachHere();
4319 return false;
4320 }
4321
4322 if (sr_notify(osthread) != 0) {
4323 ShouldNotReachHere();
4324 }
4325
4326 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4327 while (true) {
4328 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4329 break;
4330 } else {
4331 // timeout
4332 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4333 if (cancelled == os::SuspendResume::SR_RUNNING) {
4334 return false;
4335 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4336 // make sure that we consume the signal on the semaphore as well
4337 sr_semaphore.wait();
4338 break;
4339 } else {
4340 ShouldNotReachHere();
4341 return false;
4342 }
4343 }
4344 }
4345
4346 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4347 return true;
4348 }
4349
4350 static void do_resume(OSThread* osthread) {
4351 assert(osthread->sr.is_suspended(), "thread should be suspended");
4352 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4353
4354 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4355 // failed to switch to WAKEUP_REQUEST
4356 ShouldNotReachHere();
4357 return;
4358 }
4359
4360 while (true) {
4361 if (sr_notify(osthread) == 0) {
4362 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4363 if (osthread->sr.is_running()) {
4364 return;
4365 }
4366 }
4367 } else {
4368 ShouldNotReachHere();
4369 }
4370 }
4371
4372 guarantee(osthread->sr.is_running(), "Must be running!");
4373 }
4374
4375 ///////////////////////////////////////////////////////////////////////////////////
4376 // signal handling (except suspend/resume)
4377
4378 // This routine may be used by user applications as a "hook" to catch signals.
4379 // The user-defined signal handler must pass unrecognized signals to this
4380 // routine, and if it returns true (non-zero), then the signal handler must
4381 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4382 // routine will never retun false (zero), but instead will execute a VM panic
4383 // routine kill the process.
4384 //
4385 // If this routine returns false, it is OK to call it again. This allows
4386 // the user-defined signal handler to perform checks either before or after
4387 // the VM performs its own checks. Naturally, the user code would be making
4388 // a serious error if it tried to handle an exception (such as a null check
4389 // or breakpoint) that the VM was generating for its own correct operation.
4390 //
4391 // This routine may recognize any of the following kinds of signals:
4392 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4393 // It should be consulted by handlers for any of those signals.
4394 //
4395 // The caller of this routine must pass in the three arguments supplied
4396 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4397 // field of the structure passed to sigaction(). This routine assumes that
4398 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4399 //
4400 // Note that the VM will print warnings if it detects conflicting signal
4401 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4402 //
4403 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4404 siginfo_t* siginfo,
4405 void* ucontext,
4406 int abort_if_unrecognized);
4407
4408 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4409 assert(info != NULL && uc != NULL, "it must be old kernel");
4410 int orig_errno = errno; // Preserve errno value over signal handler.
4411 JVM_handle_linux_signal(sig, info, uc, true);
4412 errno = orig_errno;
4413 }
4414
4415
4416 // This boolean allows users to forward their own non-matching signals
4417 // to JVM_handle_linux_signal, harmlessly.
4418 bool os::Linux::signal_handlers_are_installed = false;
4419
4420 // For signal-chaining
4421 struct sigaction sigact[NSIG];
4422 uint64_t sigs = 0;
4423 #if (64 < NSIG-1)
4424 #error "Not all signals can be encoded in sigs. Adapt its type!"
4425 #endif
4426 bool os::Linux::libjsig_is_loaded = false;
4427 typedef struct sigaction *(*get_signal_t)(int);
4428 get_signal_t os::Linux::get_signal_action = NULL;
4429
4430 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4431 struct sigaction *actp = NULL;
4432
4433 if (libjsig_is_loaded) {
4434 // Retrieve the old signal handler from libjsig
4435 actp = (*get_signal_action)(sig);
4436 }
4437 if (actp == NULL) {
4438 // Retrieve the preinstalled signal handler from jvm
4439 actp = get_preinstalled_handler(sig);
4440 }
4441
4442 return actp;
4443 }
4444
4445 static bool call_chained_handler(struct sigaction *actp, int sig,
4446 siginfo_t *siginfo, void *context) {
4447 // Call the old signal handler
4448 if (actp->sa_handler == SIG_DFL) {
4449 // It's more reasonable to let jvm treat it as an unexpected exception
4450 // instead of taking the default action.
4451 return false;
4452 } else if (actp->sa_handler != SIG_IGN) {
4453 if ((actp->sa_flags & SA_NODEFER) == 0) {
4454 // automaticlly block the signal
4455 sigaddset(&(actp->sa_mask), sig);
4456 }
4457
4458 sa_handler_t hand = NULL;
4459 sa_sigaction_t sa = NULL;
4460 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4461 // retrieve the chained handler
4462 if (siginfo_flag_set) {
4463 sa = actp->sa_sigaction;
4464 } else {
4465 hand = actp->sa_handler;
4466 }
4467
4468 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4469 actp->sa_handler = SIG_DFL;
4470 }
4471
4472 // try to honor the signal mask
4473 sigset_t oset;
4474 sigemptyset(&oset);
4475 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4476
4477 // call into the chained handler
4478 if (siginfo_flag_set) {
4479 (*sa)(sig, siginfo, context);
4480 } else {
4481 (*hand)(sig);
4482 }
4483
4484 // restore the signal mask
4485 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4486 }
4487 // Tell jvm's signal handler the signal is taken care of.
4488 return true;
4489 }
4490
4491 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4492 bool chained = false;
4493 // signal-chaining
4494 if (UseSignalChaining) {
4495 struct sigaction *actp = get_chained_signal_action(sig);
4496 if (actp != NULL) {
4497 chained = call_chained_handler(actp, sig, siginfo, context);
4498 }
4499 }
4500 return chained;
4501 }
4502
4503 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4504 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4505 return &sigact[sig];
4506 }
4507 return NULL;
4508 }
4509
4510 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4511 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4512 sigact[sig] = oldAct;
4513 sigs |= (uint64_t)1 << (sig-1);
4514 }
4515
4516 // for diagnostic
4517 int sigflags[NSIG];
4518
4519 int os::Linux::get_our_sigflags(int sig) {
4520 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4521 return sigflags[sig];
4522 }
4523
4524 void os::Linux::set_our_sigflags(int sig, int flags) {
4525 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4526 if (sig > 0 && sig < NSIG) {
4527 sigflags[sig] = flags;
4528 }
4529 }
4530
4531 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4532 // Check for overwrite.
4533 struct sigaction oldAct;
4534 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4535
4536 void* oldhand = oldAct.sa_sigaction
4537 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4538 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4539 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4540 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4541 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4542 if (AllowUserSignalHandlers || !set_installed) {
4543 // Do not overwrite; user takes responsibility to forward to us.
4544 return;
4545 } else if (UseSignalChaining) {
4546 // save the old handler in jvm
4547 save_preinstalled_handler(sig, oldAct);
4548 // libjsig also interposes the sigaction() call below and saves the
4549 // old sigaction on it own.
4550 } else {
4551 fatal("Encountered unexpected pre-existing sigaction handler "
4552 "%#lx for signal %d.", (long)oldhand, sig);
4553 }
4554 }
4555
4556 struct sigaction sigAct;
4557 sigfillset(&(sigAct.sa_mask));
4558 sigAct.sa_handler = SIG_DFL;
4559 if (!set_installed) {
4560 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4561 } else {
4562 sigAct.sa_sigaction = signalHandler;
4563 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4564 }
4565 // Save flags, which are set by ours
4566 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4567 sigflags[sig] = sigAct.sa_flags;
4568
4569 int ret = sigaction(sig, &sigAct, &oldAct);
4570 assert(ret == 0, "check");
4571
4572 void* oldhand2 = oldAct.sa_sigaction
4573 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4574 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4575 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4576 }
4577
4578 // install signal handlers for signals that HotSpot needs to
4579 // handle in order to support Java-level exception handling.
4580
4581 void os::Linux::install_signal_handlers() {
4582 if (!signal_handlers_are_installed) {
4583 signal_handlers_are_installed = true;
4584
4585 // signal-chaining
4586 typedef void (*signal_setting_t)();
4587 signal_setting_t begin_signal_setting = NULL;
4588 signal_setting_t end_signal_setting = NULL;
4589 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4590 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4591 if (begin_signal_setting != NULL) {
4592 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4593 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4594 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4595 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4596 libjsig_is_loaded = true;
4597 assert(UseSignalChaining, "should enable signal-chaining");
4598 }
4599 if (libjsig_is_loaded) {
4600 // Tell libjsig jvm is setting signal handlers
4601 (*begin_signal_setting)();
4602 }
4603
4604 set_signal_handler(SIGSEGV, true);
4605 set_signal_handler(SIGPIPE, true);
4606 set_signal_handler(SIGBUS, true);
4607 set_signal_handler(SIGILL, true);
4608 set_signal_handler(SIGFPE, true);
4609 #if defined(PPC64)
4610 set_signal_handler(SIGTRAP, true);
4611 #endif
4612 set_signal_handler(SIGXFSZ, true);
4613
4614 if (libjsig_is_loaded) {
4615 // Tell libjsig jvm finishes setting signal handlers
4616 (*end_signal_setting)();
4617 }
4618
4619 // We don't activate signal checker if libjsig is in place, we trust ourselves
4620 // and if UserSignalHandler is installed all bets are off.
4621 // Log that signal checking is off only if -verbose:jni is specified.
4622 if (CheckJNICalls) {
4623 if (libjsig_is_loaded) {
4624 if (PrintJNIResolving) {
4625 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4626 }
4627 check_signals = false;
4628 }
4629 if (AllowUserSignalHandlers) {
4630 if (PrintJNIResolving) {
4631 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4632 }
4633 check_signals = false;
4634 }
4635 }
4636 }
4637 }
4638
4639 // This is the fastest way to get thread cpu time on Linux.
4640 // Returns cpu time (user+sys) for any thread, not only for current.
4641 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4642 // It might work on 2.6.10+ with a special kernel/glibc patch.
4643 // For reference, please, see IEEE Std 1003.1-2004:
4644 // http://www.unix.org/single_unix_specification
4645
4646 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4647 struct timespec tp;
4648 int rc = os::Posix::clock_gettime(clockid, &tp);
4649 assert(rc == 0, "clock_gettime is expected to return 0 code");
4650
4651 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4652 }
4653
4654 void os::Linux::initialize_os_info() {
4655 assert(_os_version == 0, "OS info already initialized");
4656
4657 struct utsname _uname;
4658
4659 uint32_t major;
4660 uint32_t minor;
4661 uint32_t fix;
4662
4663 int rc;
4664
4665 // Kernel version is unknown if
4666 // verification below fails.
4667 _os_version = 0x01000000;
4668
4669 rc = uname(&_uname);
4670 if (rc != -1) {
4671
4672 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4673 if (rc == 3) {
4674
4675 if (major < 256 && minor < 256 && fix < 256) {
4676 // Kernel version format is as expected,
4677 // set it overriding unknown state.
4678 _os_version = (major << 16) |
4679 (minor << 8 ) |
4680 (fix << 0 ) ;
4681 }
4682 }
4683 }
4684 }
4685
4686 uint32_t os::Linux::os_version() {
4687 assert(_os_version != 0, "not initialized");
4688 return _os_version & 0x00FFFFFF;
4689 }
4690
4691 bool os::Linux::os_version_is_known() {
4692 assert(_os_version != 0, "not initialized");
4693 return _os_version & 0x01000000 ? false : true;
4694 }
4695
4696 /////
4697 // glibc on Linux platform uses non-documented flag
4698 // to indicate, that some special sort of signal
4699 // trampoline is used.
4700 // We will never set this flag, and we should
4701 // ignore this flag in our diagnostic
4702 #ifdef SIGNIFICANT_SIGNAL_MASK
4703 #undef SIGNIFICANT_SIGNAL_MASK
4704 #endif
4705 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4706
4707 static const char* get_signal_handler_name(address handler,
4708 char* buf, int buflen) {
4709 int offset = 0;
4710 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4711 if (found) {
4712 // skip directory names
4713 const char *p1, *p2;
4714 p1 = buf;
4715 size_t len = strlen(os::file_separator());
4716 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4717 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4718 } else {
4719 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4720 }
4721 return buf;
4722 }
4723
4724 static void print_signal_handler(outputStream* st, int sig,
4725 char* buf, size_t buflen) {
4726 struct sigaction sa;
4727
4728 sigaction(sig, NULL, &sa);
4729
4730 // See comment for SIGNIFICANT_SIGNAL_MASK define
4731 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4732
4733 st->print("%s: ", os::exception_name(sig, buf, buflen));
4734
4735 address handler = (sa.sa_flags & SA_SIGINFO)
4736 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4737 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4738
4739 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4740 st->print("SIG_DFL");
4741 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4742 st->print("SIG_IGN");
4743 } else {
4744 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4745 }
4746
4747 st->print(", sa_mask[0]=");
4748 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4749
4750 address rh = VMError::get_resetted_sighandler(sig);
4751 // May be, handler was resetted by VMError?
4752 if (rh != NULL) {
4753 handler = rh;
4754 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4755 }
4756
4757 st->print(", sa_flags=");
4758 os::Posix::print_sa_flags(st, sa.sa_flags);
4759
4760 // Check: is it our handler?
4761 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4762 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4763 // It is our signal handler
4764 // check for flags, reset system-used one!
4765 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4766 st->print(
4767 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4768 os::Linux::get_our_sigflags(sig));
4769 }
4770 }
4771 st->cr();
4772 }
4773
4774
4775 #define DO_SIGNAL_CHECK(sig) \
4776 do { \
4777 if (!sigismember(&check_signal_done, sig)) { \
4778 os::Linux::check_signal_handler(sig); \
4779 } \
4780 } while (0)
4781
4782 // This method is a periodic task to check for misbehaving JNI applications
4783 // under CheckJNI, we can add any periodic checks here
4784
4785 void os::run_periodic_checks() {
4786 if (check_signals == false) return;
4787
4788 // SEGV and BUS if overridden could potentially prevent
4789 // generation of hs*.log in the event of a crash, debugging
4790 // such a case can be very challenging, so we absolutely
4791 // check the following for a good measure:
4792 DO_SIGNAL_CHECK(SIGSEGV);
4793 DO_SIGNAL_CHECK(SIGILL);
4794 DO_SIGNAL_CHECK(SIGFPE);
4795 DO_SIGNAL_CHECK(SIGBUS);
4796 DO_SIGNAL_CHECK(SIGPIPE);
4797 DO_SIGNAL_CHECK(SIGXFSZ);
4798 #if defined(PPC64)
4799 DO_SIGNAL_CHECK(SIGTRAP);
4800 #endif
4801
4802 // ReduceSignalUsage allows the user to override these handlers
4803 // see comments at the very top and jvm_md.h
4804 if (!ReduceSignalUsage) {
4805 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4806 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4807 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4808 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4809 }
4810
4811 DO_SIGNAL_CHECK(SR_signum);
4812 }
4813
4814 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4815
4816 static os_sigaction_t os_sigaction = NULL;
4817
4818 void os::Linux::check_signal_handler(int sig) {
4819 char buf[O_BUFLEN];
4820 address jvmHandler = NULL;
4821
4822
4823 struct sigaction act;
4824 if (os_sigaction == NULL) {
4825 // only trust the default sigaction, in case it has been interposed
4826 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4827 if (os_sigaction == NULL) return;
4828 }
4829
4830 os_sigaction(sig, (struct sigaction*)NULL, &act);
4831
4832
4833 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4834
4835 address thisHandler = (act.sa_flags & SA_SIGINFO)
4836 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4837 : CAST_FROM_FN_PTR(address, act.sa_handler);
4838
4839
4840 switch (sig) {
4841 case SIGSEGV:
4842 case SIGBUS:
4843 case SIGFPE:
4844 case SIGPIPE:
4845 case SIGILL:
4846 case SIGXFSZ:
4847 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4848 break;
4849
4850 case SHUTDOWN1_SIGNAL:
4851 case SHUTDOWN2_SIGNAL:
4852 case SHUTDOWN3_SIGNAL:
4853 case BREAK_SIGNAL:
4854 jvmHandler = (address)user_handler();
4855 break;
4856
4857 default:
4858 if (sig == SR_signum) {
4859 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4860 } else {
4861 return;
4862 }
4863 break;
4864 }
4865
4866 if (thisHandler != jvmHandler) {
4867 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4868 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4869 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4870 // No need to check this sig any longer
4871 sigaddset(&check_signal_done, sig);
4872 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4873 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4874 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4875 exception_name(sig, buf, O_BUFLEN));
4876 }
4877 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4878 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4879 tty->print("expected:");
4880 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4881 tty->cr();
4882 tty->print(" found:");
4883 os::Posix::print_sa_flags(tty, act.sa_flags);
4884 tty->cr();
4885 // No need to check this sig any longer
4886 sigaddset(&check_signal_done, sig);
4887 }
4888
4889 // Dump all the signal
4890 if (sigismember(&check_signal_done, sig)) {
4891 print_signal_handlers(tty, buf, O_BUFLEN);
4892 }
4893 }
4894
4895 extern void report_error(char* file_name, int line_no, char* title,
4896 char* format, ...);
4897
4898 // this is called _before_ most of the global arguments have been parsed
4899 void os::init(void) {
4900 char dummy; // used to get a guess on initial stack address
4901
4902 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4903
4904 init_random(1234567);
4905
4906 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4907 if (Linux::page_size() == -1) {
4908 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4909 os::strerror(errno));
4910 }
4911 init_page_sizes((size_t) Linux::page_size());
4912
4913 Linux::initialize_system_info();
4914
4915 Linux::initialize_os_info();
4916
4917 // _main_thread points to the thread that created/loaded the JVM.
4918 Linux::_main_thread = pthread_self();
4919
4920 // retrieve entry point for pthread_setname_np
4921 Linux::_pthread_setname_np =
4922 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4923
4924 os::Posix::init();
4925
4926 initial_time_count = javaTimeNanos();
4927
4928 // Always warn if no monotonic clock available
4929 if (!os::Posix::supports_monotonic_clock()) {
4930 warning("No monotonic clock was available - timed services may " \
4931 "be adversely affected if the time-of-day clock changes");
4932 }
4933 }
4934
4935 // To install functions for atexit system call
4936 extern "C" {
4937 static void perfMemory_exit_helper() {
4938 perfMemory_exit();
4939 }
4940 }
4941
4942 void os::pd_init_container_support() {
4943 OSContainer::init();
4944 }
4945
4946 // this is called _after_ the global arguments have been parsed
4947 jint os::init_2(void) {
4948
4949 // This could be set after os::Posix::init() but all platforms
4950 // have to set it the same so we have to mirror Solaris.
4951 DEBUG_ONLY(os::set_mutex_init_done();)
4952
4953 os::Posix::init_2();
4954
4955 Linux::fast_thread_clock_init();
4956
4957 // initialize suspend/resume support - must do this before signal_sets_init()
4958 if (SR_initialize() != 0) {
4959 perror("SR_initialize failed");
4960 return JNI_ERR;
4961 }
4962
4963 Linux::signal_sets_init();
4964 Linux::install_signal_handlers();
4965 // Initialize data for jdk.internal.misc.Signal
4966 if (!ReduceSignalUsage) {
4967 jdk_misc_signal_init();
4968 }
4969
4970 // Check and sets minimum stack sizes against command line options
4971 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4972 return JNI_ERR;
4973 }
4974
4975 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
4976 if (!suppress_primordial_thread_resolution) {
4977 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4978 }
4979
4980 #if defined(IA32)
4981 workaround_expand_exec_shield_cs_limit();
4982 #endif
4983
4984 Linux::libpthread_init();
4985 Linux::sched_getcpu_init();
4986 log_info(os)("HotSpot is running with %s, %s",
4987 Linux::glibc_version(), Linux::libpthread_version());
4988
4989 if (UseNUMA) {
4990 if (!Linux::libnuma_init()) {
4991 UseNUMA = false;
4992 } else {
4993 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
4994 // If there's only one node (they start from 0) or if the process
4995 // is bound explicitly to a single node using membind, disable NUMA.
4996 UseNUMA = false;
4997 }
4998 }
4999
5000 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5001 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5002 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5003 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5004 // and disable adaptive resizing.
5005 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5006 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5007 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5008 UseAdaptiveSizePolicy = false;
5009 UseAdaptiveNUMAChunkSizing = false;
5010 }
5011 }
5012
5013 if (!UseNUMA && ForceNUMA) {
5014 UseNUMA = true;
5015 }
5016 }
5017
5018 if (MaxFDLimit) {
5019 // set the number of file descriptors to max. print out error
5020 // if getrlimit/setrlimit fails but continue regardless.
5021 struct rlimit nbr_files;
5022 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5023 if (status != 0) {
5024 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5025 } else {
5026 nbr_files.rlim_cur = nbr_files.rlim_max;
5027 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5028 if (status != 0) {
5029 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5030 }
5031 }
5032 }
5033
5034 // Initialize lock used to serialize thread creation (see os::create_thread)
5035 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5036
5037 // at-exit methods are called in the reverse order of their registration.
5038 // atexit functions are called on return from main or as a result of a
5039 // call to exit(3C). There can be only 32 of these functions registered
5040 // and atexit() does not set errno.
5041
5042 if (PerfAllowAtExitRegistration) {
5043 // only register atexit functions if PerfAllowAtExitRegistration is set.
5044 // atexit functions can be delayed until process exit time, which
5045 // can be problematic for embedded VM situations. Embedded VMs should
5046 // call DestroyJavaVM() to assure that VM resources are released.
5047
5048 // note: perfMemory_exit_helper atexit function may be removed in
5049 // the future if the appropriate cleanup code can be added to the
5050 // VM_Exit VMOperation's doit method.
5051 if (atexit(perfMemory_exit_helper) != 0) {
5052 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5053 }
5054 }
5055
5056 // initialize thread priority policy
5057 prio_init();
5058
5059 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5060 set_coredump_filter(DAX_SHARED_BIT);
5061 }
5062
5063 #if INCLUDE_CDS
5064 if (UseSharedSpaces && DumpPrivateMappingsInCore) {
5065 set_coredump_filter(FILE_BACKED_PVT_BIT);
5066 }
5067 #endif
5068
5069 return JNI_OK;
5070 }
5071
5072 // Mark the polling page as unreadable
5073 void os::make_polling_page_unreadable(void) {
5074 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5075 fatal("Could not disable polling page");
5076 }
5077 }
5078
5079 // Mark the polling page as readable
5080 void os::make_polling_page_readable(void) {
5081 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5082 fatal("Could not enable polling page");
5083 }
5084 }
5085
5086 // older glibc versions don't have this macro (which expands to
5087 // an optimized bit-counting function) so we have to roll our own
5088 #ifndef CPU_COUNT
5089
5090 static int _cpu_count(const cpu_set_t* cpus) {
5091 int count = 0;
5092 // only look up to the number of configured processors
5093 for (int i = 0; i < os::processor_count(); i++) {
5094 if (CPU_ISSET(i, cpus)) {
5095 count++;
5096 }
5097 }
5098 return count;
5099 }
5100
5101 #define CPU_COUNT(cpus) _cpu_count(cpus)
5102
5103 #endif // CPU_COUNT
5104
5105 // Get the current number of available processors for this process.
5106 // This value can change at any time during a process's lifetime.
5107 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5108 // If it appears there may be more than 1024 processors then we do a
5109 // dynamic check - see 6515172 for details.
5110 // If anything goes wrong we fallback to returning the number of online
5111 // processors - which can be greater than the number available to the process.
5112 int os::Linux::active_processor_count() {
5113 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5114 cpu_set_t* cpus_p = &cpus;
5115 int cpus_size = sizeof(cpu_set_t);
5116
5117 int configured_cpus = os::processor_count(); // upper bound on available cpus
5118 int cpu_count = 0;
5119
5120 // old build platforms may not support dynamic cpu sets
5121 #ifdef CPU_ALLOC
5122
5123 // To enable easy testing of the dynamic path on different platforms we
5124 // introduce a diagnostic flag: UseCpuAllocPath
5125 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5126 // kernel may use a mask bigger than cpu_set_t
5127 log_trace(os)("active_processor_count: using dynamic path %s"
5128 "- configured processors: %d",
5129 UseCpuAllocPath ? "(forced) " : "",
5130 configured_cpus);
5131 cpus_p = CPU_ALLOC(configured_cpus);
5132 if (cpus_p != NULL) {
5133 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5134 // zero it just to be safe
5135 CPU_ZERO_S(cpus_size, cpus_p);
5136 }
5137 else {
5138 // failed to allocate so fallback to online cpus
5139 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5140 log_trace(os)("active_processor_count: "
5141 "CPU_ALLOC failed (%s) - using "
5142 "online processor count: %d",
5143 os::strerror(errno), online_cpus);
5144 return online_cpus;
5145 }
5146 }
5147 else {
5148 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5149 configured_cpus);
5150 }
5151 #else // CPU_ALLOC
5152 // these stubs won't be executed
5153 #define CPU_COUNT_S(size, cpus) -1
5154 #define CPU_FREE(cpus)
5155
5156 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5157 configured_cpus);
5158 #endif // CPU_ALLOC
5159
5160 // pid 0 means the current thread - which we have to assume represents the process
5161 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5162 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5163 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5164 }
5165 else {
5166 cpu_count = CPU_COUNT(cpus_p);
5167 }
5168 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5169 }
5170 else {
5171 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5172 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5173 "which may exceed available processors", os::strerror(errno), cpu_count);
5174 }
5175
5176 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5177 CPU_FREE(cpus_p);
5178 }
5179
5180 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5181 return cpu_count;
5182 }
5183
5184 // Determine the active processor count from one of
5185 // three different sources:
5186 //
5187 // 1. User option -XX:ActiveProcessorCount
5188 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5189 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5190 //
5191 // Option 1, if specified, will always override.
5192 // If the cgroup subsystem is active and configured, we
5193 // will return the min of the cgroup and option 2 results.
5194 // This is required since tools, such as numactl, that
5195 // alter cpu affinity do not update cgroup subsystem
5196 // cpuset configuration files.
5197 int os::active_processor_count() {
5198 // User has overridden the number of active processors
5199 if (ActiveProcessorCount > 0) {
5200 log_trace(os)("active_processor_count: "
5201 "active processor count set by user : %d",
5202 ActiveProcessorCount);
5203 return ActiveProcessorCount;
5204 }
5205
5206 int active_cpus;
5207 if (OSContainer::is_containerized()) {
5208 active_cpus = OSContainer::active_processor_count();
5209 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5210 active_cpus);
5211 } else {
5212 active_cpus = os::Linux::active_processor_count();
5213 }
5214
5215 return active_cpus;
5216 }
5217
5218 uint os::processor_id() {
5219 const int id = Linux::sched_getcpu();
5220 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5221 return (uint)id;
5222 }
5223
5224 void os::set_native_thread_name(const char *name) {
5225 if (Linux::_pthread_setname_np) {
5226 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5227 snprintf(buf, sizeof(buf), "%s", name);
5228 buf[sizeof(buf) - 1] = '\0';
5229 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5230 // ERANGE should not happen; all other errors should just be ignored.
5231 assert(rc != ERANGE, "pthread_setname_np failed");
5232 }
5233 }
5234
5235 bool os::distribute_processes(uint length, uint* distribution) {
5236 // Not yet implemented.
5237 return false;
5238 }
5239
5240 bool os::bind_to_processor(uint processor_id) {
5241 // Not yet implemented.
5242 return false;
5243 }
5244
5245 ///
5246
5247 void os::SuspendedThreadTask::internal_do_task() {
5248 if (do_suspend(_thread->osthread())) {
5249 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5250 do_task(context);
5251 do_resume(_thread->osthread());
5252 }
5253 }
5254
5255 ////////////////////////////////////////////////////////////////////////////////
5256 // debug support
5257
5258 bool os::find(address addr, outputStream* st) {
5259 Dl_info dlinfo;
5260 memset(&dlinfo, 0, sizeof(dlinfo));
5261 if (dladdr(addr, &dlinfo) != 0) {
5262 st->print(PTR_FORMAT ": ", p2i(addr));
5263 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5264 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5265 p2i(addr) - p2i(dlinfo.dli_saddr));
5266 } else if (dlinfo.dli_fbase != NULL) {
5267 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5268 } else {
5269 st->print("<absolute address>");
5270 }
5271 if (dlinfo.dli_fname != NULL) {
5272 st->print(" in %s", dlinfo.dli_fname);
5273 }
5274 if (dlinfo.dli_fbase != NULL) {
5275 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5276 }
5277 st->cr();
5278
5279 if (Verbose) {
5280 // decode some bytes around the PC
5281 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5282 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5283 address lowest = (address) dlinfo.dli_sname;
5284 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5285 if (begin < lowest) begin = lowest;
5286 Dl_info dlinfo2;
5287 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5288 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5289 end = (address) dlinfo2.dli_saddr;
5290 }
5291 Disassembler::decode(begin, end, st);
5292 }
5293 return true;
5294 }
5295 return false;
5296 }
5297
5298 ////////////////////////////////////////////////////////////////////////////////
5299 // misc
5300
5301 // This does not do anything on Linux. This is basically a hook for being
5302 // able to use structured exception handling (thread-local exception filters)
5303 // on, e.g., Win32.
5304 void
5305 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5306 JavaCallArguments* args, Thread* thread) {
5307 f(value, method, args, thread);
5308 }
5309
5310 void os::print_statistics() {
5311 }
5312
5313 bool os::message_box(const char* title, const char* message) {
5314 int i;
5315 fdStream err(defaultStream::error_fd());
5316 for (i = 0; i < 78; i++) err.print_raw("=");
5317 err.cr();
5318 err.print_raw_cr(title);
5319 for (i = 0; i < 78; i++) err.print_raw("-");
5320 err.cr();
5321 err.print_raw_cr(message);
5322 for (i = 0; i < 78; i++) err.print_raw("=");
5323 err.cr();
5324
5325 char buf[16];
5326 // Prevent process from exiting upon "read error" without consuming all CPU
5327 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5328
5329 return buf[0] == 'y' || buf[0] == 'Y';
5330 }
5331
5332 // Is a (classpath) directory empty?
5333 bool os::dir_is_empty(const char* path) {
5334 DIR *dir = NULL;
5335 struct dirent *ptr;
5336
5337 dir = opendir(path);
5338 if (dir == NULL) return true;
5339
5340 // Scan the directory
5341 bool result = true;
5342 while (result && (ptr = readdir(dir)) != NULL) {
5343 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5344 result = false;
5345 }
5346 }
5347 closedir(dir);
5348 return result;
5349 }
5350
5351 // This code originates from JDK's sysOpen and open64_w
5352 // from src/solaris/hpi/src/system_md.c
5353
5354 int os::open(const char *path, int oflag, int mode) {
5355 if (strlen(path) > MAX_PATH - 1) {
5356 errno = ENAMETOOLONG;
5357 return -1;
5358 }
5359
5360 // All file descriptors that are opened in the Java process and not
5361 // specifically destined for a subprocess should have the close-on-exec
5362 // flag set. If we don't set it, then careless 3rd party native code
5363 // might fork and exec without closing all appropriate file descriptors
5364 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5365 // turn might:
5366 //
5367 // - cause end-of-file to fail to be detected on some file
5368 // descriptors, resulting in mysterious hangs, or
5369 //
5370 // - might cause an fopen in the subprocess to fail on a system
5371 // suffering from bug 1085341.
5372 //
5373 // (Yes, the default setting of the close-on-exec flag is a Unix
5374 // design flaw)
5375 //
5376 // See:
5377 // 1085341: 32-bit stdio routines should support file descriptors >255
5378 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5379 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5380 //
5381 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5382 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5383 // because it saves a system call and removes a small window where the flag
5384 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5385 // and we fall back to using FD_CLOEXEC (see below).
5386 #ifdef O_CLOEXEC
5387 oflag |= O_CLOEXEC;
5388 #endif
5389
5390 int fd = ::open64(path, oflag, mode);
5391 if (fd == -1) return -1;
5392
5393 //If the open succeeded, the file might still be a directory
5394 {
5395 struct stat64 buf64;
5396 int ret = ::fstat64(fd, &buf64);
5397 int st_mode = buf64.st_mode;
5398
5399 if (ret != -1) {
5400 if ((st_mode & S_IFMT) == S_IFDIR) {
5401 errno = EISDIR;
5402 ::close(fd);
5403 return -1;
5404 }
5405 } else {
5406 ::close(fd);
5407 return -1;
5408 }
5409 }
5410
5411 #ifdef FD_CLOEXEC
5412 // Validate that the use of the O_CLOEXEC flag on open above worked.
5413 // With recent kernels, we will perform this check exactly once.
5414 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5415 if (!O_CLOEXEC_is_known_to_work) {
5416 int flags = ::fcntl(fd, F_GETFD);
5417 if (flags != -1) {
5418 if ((flags & FD_CLOEXEC) != 0)
5419 O_CLOEXEC_is_known_to_work = 1;
5420 else
5421 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5422 }
5423 }
5424 #endif
5425
5426 return fd;
5427 }
5428
5429
5430 // create binary file, rewriting existing file if required
5431 int os::create_binary_file(const char* path, bool rewrite_existing) {
5432 int oflags = O_WRONLY | O_CREAT;
5433 if (!rewrite_existing) {
5434 oflags |= O_EXCL;
5435 }
5436 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5437 }
5438
5439 // return current position of file pointer
5440 jlong os::current_file_offset(int fd) {
5441 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5442 }
5443
5444 // move file pointer to the specified offset
5445 jlong os::seek_to_file_offset(int fd, jlong offset) {
5446 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5447 }
5448
5449 // This code originates from JDK's sysAvailable
5450 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5451
5452 int os::available(int fd, jlong *bytes) {
5453 jlong cur, end;
5454 int mode;
5455 struct stat64 buf64;
5456
5457 if (::fstat64(fd, &buf64) >= 0) {
5458 mode = buf64.st_mode;
5459 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5460 int n;
5461 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5462 *bytes = n;
5463 return 1;
5464 }
5465 }
5466 }
5467 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5468 return 0;
5469 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5470 return 0;
5471 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5472 return 0;
5473 }
5474 *bytes = end - cur;
5475 return 1;
5476 }
5477
5478 // Map a block of memory.
5479 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5480 char *addr, size_t bytes, bool read_only,
5481 bool allow_exec) {
5482 int prot;
5483 int flags = MAP_PRIVATE;
5484
5485 if (read_only) {
5486 prot = PROT_READ;
5487 } else {
5488 prot = PROT_READ | PROT_WRITE;
5489 }
5490
5491 if (allow_exec) {
5492 prot |= PROT_EXEC;
5493 }
5494
5495 if (addr != NULL) {
5496 flags |= MAP_FIXED;
5497 }
5498
5499 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5500 fd, file_offset);
5501 if (mapped_address == MAP_FAILED) {
5502 return NULL;
5503 }
5504 return mapped_address;
5505 }
5506
5507
5508 // Remap a block of memory.
5509 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5510 char *addr, size_t bytes, bool read_only,
5511 bool allow_exec) {
5512 // same as map_memory() on this OS
5513 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5514 allow_exec);
5515 }
5516
5517
5518 // Unmap a block of memory.
5519 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5520 return munmap(addr, bytes) == 0;
5521 }
5522
5523 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5524
5525 static jlong fast_cpu_time(Thread *thread) {
5526 clockid_t clockid;
5527 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5528 &clockid);
5529 if (rc == 0) {
5530 return os::Linux::fast_thread_cpu_time(clockid);
5531 } else {
5532 // It's possible to encounter a terminated native thread that failed
5533 // to detach itself from the VM - which should result in ESRCH.
5534 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5535 return -1;
5536 }
5537 }
5538
5539 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5540 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5541 // of a thread.
5542 //
5543 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5544 // the fast estimate available on the platform.
5545
5546 jlong os::current_thread_cpu_time() {
5547 if (os::Linux::supports_fast_thread_cpu_time()) {
5548 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5549 } else {
5550 // return user + sys since the cost is the same
5551 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5552 }
5553 }
5554
5555 jlong os::thread_cpu_time(Thread* thread) {
5556 // consistent with what current_thread_cpu_time() returns
5557 if (os::Linux::supports_fast_thread_cpu_time()) {
5558 return fast_cpu_time(thread);
5559 } else {
5560 return slow_thread_cpu_time(thread, true /* user + sys */);
5561 }
5562 }
5563
5564 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5565 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5566 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5567 } else {
5568 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5569 }
5570 }
5571
5572 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5573 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5574 return fast_cpu_time(thread);
5575 } else {
5576 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5577 }
5578 }
5579
5580 // -1 on error.
5581 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5582 pid_t tid = thread->osthread()->thread_id();
5583 char *s;
5584 char stat[2048];
5585 int statlen;
5586 char proc_name[64];
5587 int count;
5588 long sys_time, user_time;
5589 char cdummy;
5590 int idummy;
5591 long ldummy;
5592 FILE *fp;
5593
5594 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5595 fp = fopen(proc_name, "r");
5596 if (fp == NULL) return -1;
5597 statlen = fread(stat, 1, 2047, fp);
5598 stat[statlen] = '\0';
5599 fclose(fp);
5600
5601 // Skip pid and the command string. Note that we could be dealing with
5602 // weird command names, e.g. user could decide to rename java launcher
5603 // to "java 1.4.2 :)", then the stat file would look like
5604 // 1234 (java 1.4.2 :)) R ... ...
5605 // We don't really need to know the command string, just find the last
5606 // occurrence of ")" and then start parsing from there. See bug 4726580.
5607 s = strrchr(stat, ')');
5608 if (s == NULL) return -1;
5609
5610 // Skip blank chars
5611 do { s++; } while (s && isspace(*s));
5612
5613 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5614 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5615 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5616 &user_time, &sys_time);
5617 if (count != 13) return -1;
5618 if (user_sys_cpu_time) {
5619 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5620 } else {
5621 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5622 }
5623 }
5624
5625 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5626 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5627 info_ptr->may_skip_backward = false; // elapsed time not wall time
5628 info_ptr->may_skip_forward = false; // elapsed time not wall time
5629 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5630 }
5631
5632 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5633 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5634 info_ptr->may_skip_backward = false; // elapsed time not wall time
5635 info_ptr->may_skip_forward = false; // elapsed time not wall time
5636 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5637 }
5638
5639 bool os::is_thread_cpu_time_supported() {
5640 return true;
5641 }
5642
5643 // System loadavg support. Returns -1 if load average cannot be obtained.
5644 // Linux doesn't yet have a (official) notion of processor sets,
5645 // so just return the system wide load average.
5646 int os::loadavg(double loadavg[], int nelem) {
5647 return ::getloadavg(loadavg, nelem);
5648 }
5649
5650 void os::pause() {
5651 char filename[MAX_PATH];
5652 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5653 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5654 } else {
5655 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5656 }
5657
5658 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5659 if (fd != -1) {
5660 struct stat buf;
5661 ::close(fd);
5662 while (::stat(filename, &buf) == 0) {
5663 (void)::poll(NULL, 0, 100);
5664 }
5665 } else {
5666 jio_fprintf(stderr,
5667 "Could not open pause file '%s', continuing immediately.\n", filename);
5668 }
5669 }
5670
5671 extern char** environ;
5672
5673 // Run the specified command in a separate process. Return its exit value,
5674 // or -1 on failure (e.g. can't fork a new process).
5675 // Unlike system(), this function can be called from signal handler. It
5676 // doesn't block SIGINT et al.
5677 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5678 const char * argv[4] = {"sh", "-c", cmd, NULL};
5679
5680 pid_t pid ;
5681
5682 if (use_vfork_if_available) {
5683 pid = vfork();
5684 } else {
5685 pid = fork();
5686 }
5687
5688 if (pid < 0) {
5689 // fork failed
5690 return -1;
5691
5692 } else if (pid == 0) {
5693 // child process
5694
5695 execve("/bin/sh", (char* const*)argv, environ);
5696
5697 // execve failed
5698 _exit(-1);
5699
5700 } else {
5701 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5702 // care about the actual exit code, for now.
5703
5704 int status;
5705
5706 // Wait for the child process to exit. This returns immediately if
5707 // the child has already exited. */
5708 while (waitpid(pid, &status, 0) < 0) {
5709 switch (errno) {
5710 case ECHILD: return 0;
5711 case EINTR: break;
5712 default: return -1;
5713 }
5714 }
5715
5716 if (WIFEXITED(status)) {
5717 // The child exited normally; get its exit code.
5718 return WEXITSTATUS(status);
5719 } else if (WIFSIGNALED(status)) {
5720 // The child exited because of a signal
5721 // The best value to return is 0x80 + signal number,
5722 // because that is what all Unix shells do, and because
5723 // it allows callers to distinguish between process exit and
5724 // process death by signal.
5725 return 0x80 + WTERMSIG(status);
5726 } else {
5727 // Unknown exit code; pass it through
5728 return status;
5729 }
5730 }
5731 }
5732
5733 // Get the default path to the core file
5734 // Returns the length of the string
5735 int os::get_core_path(char* buffer, size_t bufferSize) {
5736 /*
5737 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5738 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5739 */
5740 const int core_pattern_len = 129;
5741 char core_pattern[core_pattern_len] = {0};
5742
5743 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5744 if (core_pattern_file == -1) {
5745 return -1;
5746 }
5747
5748 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5749 ::close(core_pattern_file);
5750 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5751 return -1;
5752 }
5753 if (core_pattern[ret-1] == '\n') {
5754 core_pattern[ret-1] = '\0';
5755 } else {
5756 core_pattern[ret] = '\0';
5757 }
5758
5759 // Replace the %p in the core pattern with the process id. NOTE: we do this
5760 // only if the pattern doesn't start with "|", and we support only one %p in
5761 // the pattern.
5762 char *pid_pos = strstr(core_pattern, "%p");
5763 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5764 int written;
5765
5766 if (core_pattern[0] == '/') {
5767 if (pid_pos != NULL) {
5768 *pid_pos = '\0';
5769 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5770 current_process_id(), tail);
5771 } else {
5772 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5773 }
5774 } else {
5775 char cwd[PATH_MAX];
5776
5777 const char* p = get_current_directory(cwd, PATH_MAX);
5778 if (p == NULL) {
5779 return -1;
5780 }
5781
5782 if (core_pattern[0] == '|') {
5783 written = jio_snprintf(buffer, bufferSize,
5784 "\"%s\" (or dumping to %s/core.%d)",
5785 &core_pattern[1], p, current_process_id());
5786 } else if (pid_pos != NULL) {
5787 *pid_pos = '\0';
5788 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5789 current_process_id(), tail);
5790 } else {
5791 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5792 }
5793 }
5794
5795 if (written < 0) {
5796 return -1;
5797 }
5798
5799 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5800 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5801
5802 if (core_uses_pid_file != -1) {
5803 char core_uses_pid = 0;
5804 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5805 ::close(core_uses_pid_file);
5806
5807 if (core_uses_pid == '1') {
5808 jio_snprintf(buffer + written, bufferSize - written,
5809 ".%d", current_process_id());
5810 }
5811 }
5812 }
5813
5814 return strlen(buffer);
5815 }
5816
5817 bool os::start_debugging(char *buf, int buflen) {
5818 int len = (int)strlen(buf);
5819 char *p = &buf[len];
5820
5821 jio_snprintf(p, buflen-len,
5822 "\n\n"
5823 "Do you want to debug the problem?\n\n"
5824 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5825 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5826 "Otherwise, press RETURN to abort...",
5827 os::current_process_id(), os::current_process_id(),
5828 os::current_thread_id(), os::current_thread_id());
5829
5830 bool yes = os::message_box("Unexpected Error", buf);
5831
5832 if (yes) {
5833 // yes, user asked VM to launch debugger
5834 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5835 os::current_process_id(), os::current_process_id());
5836
5837 os::fork_and_exec(buf);
5838 yes = false;
5839 }
5840 return yes;
5841 }
5842
5843
5844 // Java/Compiler thread:
5845 //
5846 // Low memory addresses
5847 // P0 +------------------------+
5848 // | |\ Java thread created by VM does not have glibc
5849 // | glibc guard page | - guard page, attached Java thread usually has
5850 // | |/ 1 glibc guard page.
5851 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5852 // | |\
5853 // | HotSpot Guard Pages | - red, yellow and reserved pages
5854 // | |/
5855 // +------------------------+ JavaThread::stack_reserved_zone_base()
5856 // | |\
5857 // | Normal Stack | -
5858 // | |/
5859 // P2 +------------------------+ Thread::stack_base()
5860 //
5861 // Non-Java thread:
5862 //
5863 // Low memory addresses
5864 // P0 +------------------------+
5865 // | |\
5866 // | glibc guard page | - usually 1 page
5867 // | |/
5868 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5869 // | |\
5870 // | Normal Stack | -
5871 // | |/
5872 // P2 +------------------------+ Thread::stack_base()
5873 //
5874 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5875 // returned from pthread_attr_getstack().
5876 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5877 // of the stack size given in pthread_attr. We work around this for
5878 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5879 //
5880 #ifndef ZERO
5881 static void current_stack_region(address * bottom, size_t * size) {
5882 if (os::is_primordial_thread()) {
5883 // primordial thread needs special handling because pthread_getattr_np()
5884 // may return bogus value.
5885 *bottom = os::Linux::initial_thread_stack_bottom();
5886 *size = os::Linux::initial_thread_stack_size();
5887 } else {
5888 pthread_attr_t attr;
5889
5890 int rslt = pthread_getattr_np(pthread_self(), &attr);
5891
5892 // JVM needs to know exact stack location, abort if it fails
5893 if (rslt != 0) {
5894 if (rslt == ENOMEM) {
5895 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5896 } else {
5897 fatal("pthread_getattr_np failed with error = %d", rslt);
5898 }
5899 }
5900
5901 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5902 fatal("Cannot locate current stack attributes!");
5903 }
5904
5905 // Work around NPTL stack guard error.
5906 size_t guard_size = 0;
5907 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5908 if (rslt != 0) {
5909 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5910 }
5911 *bottom += guard_size;
5912 *size -= guard_size;
5913
5914 pthread_attr_destroy(&attr);
5915
5916 }
5917 assert(os::current_stack_pointer() >= *bottom &&
5918 os::current_stack_pointer() < *bottom + *size, "just checking");
5919 }
5920
5921 address os::current_stack_base() {
5922 address bottom;
5923 size_t size;
5924 current_stack_region(&bottom, &size);
5925 return (bottom + size);
5926 }
5927
5928 size_t os::current_stack_size() {
5929 // This stack size includes the usable stack and HotSpot guard pages
5930 // (for the threads that have Hotspot guard pages).
5931 address bottom;
5932 size_t size;
5933 current_stack_region(&bottom, &size);
5934 return size;
5935 }
5936 #endif
5937
5938 static inline struct timespec get_mtime(const char* filename) {
5939 struct stat st;
5940 int ret = os::stat(filename, &st);
5941 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
5942 return st.st_mtim;
5943 }
5944
5945 int os::compare_file_modified_times(const char* file1, const char* file2) {
5946 struct timespec filetime1 = get_mtime(file1);
5947 struct timespec filetime2 = get_mtime(file2);
5948 int diff = filetime1.tv_sec - filetime2.tv_sec;
5949 if (diff == 0) {
5950 return filetime1.tv_nsec - filetime2.tv_nsec;
5951 }
5952 return diff;
5953 }
5954
5955 /////////////// Unit tests ///////////////
5956
5957 #ifndef PRODUCT
5958
5959 class TestReserveMemorySpecial : AllStatic {
5960 public:
5961 static void small_page_write(void* addr, size_t size) {
5962 size_t page_size = os::vm_page_size();
5963
5964 char* end = (char*)addr + size;
5965 for (char* p = (char*)addr; p < end; p += page_size) {
5966 *p = 1;
5967 }
5968 }
5969
5970 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5971 if (!UseHugeTLBFS) {
5972 return;
5973 }
5974
5975 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5976
5977 if (addr != NULL) {
5978 small_page_write(addr, size);
5979
5980 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5981 }
5982 }
5983
5984 static void test_reserve_memory_special_huge_tlbfs_only() {
5985 if (!UseHugeTLBFS) {
5986 return;
5987 }
5988
5989 size_t lp = os::large_page_size();
5990
5991 for (size_t size = lp; size <= lp * 10; size += lp) {
5992 test_reserve_memory_special_huge_tlbfs_only(size);
5993 }
5994 }
5995
5996 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5997 size_t lp = os::large_page_size();
5998 size_t ag = os::vm_allocation_granularity();
5999
6000 // sizes to test
6001 const size_t sizes[] = {
6002 lp, lp + ag, lp + lp / 2, lp * 2,
6003 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6004 lp * 10, lp * 10 + lp / 2
6005 };
6006 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6007
6008 // For each size/alignment combination, we test three scenarios:
6009 // 1) with req_addr == NULL
6010 // 2) with a non-null req_addr at which we expect to successfully allocate
6011 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6012 // expect the allocation to either fail or to ignore req_addr
6013
6014 // Pre-allocate two areas; they shall be as large as the largest allocation
6015 // and aligned to the largest alignment we will be testing.
6016 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6017 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6018 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6019 -1, 0);
6020 assert(mapping1 != MAP_FAILED, "should work");
6021
6022 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6023 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6024 -1, 0);
6025 assert(mapping2 != MAP_FAILED, "should work");
6026
6027 // Unmap the first mapping, but leave the second mapping intact: the first
6028 // mapping will serve as a value for a "good" req_addr (case 2). The second
6029 // mapping, still intact, as "bad" req_addr (case 3).
6030 ::munmap(mapping1, mapping_size);
6031
6032 // Case 1
6033 for (int i = 0; i < num_sizes; i++) {
6034 const size_t size = sizes[i];
6035 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6036 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6037 if (p != NULL) {
6038 assert(is_aligned(p, alignment), "must be");
6039 small_page_write(p, size);
6040 os::Linux::release_memory_special_huge_tlbfs(p, size);
6041 }
6042 }
6043 }
6044
6045 // Case 2
6046 for (int i = 0; i < num_sizes; i++) {
6047 const size_t size = sizes[i];
6048 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6049 char* const req_addr = align_up(mapping1, alignment);
6050 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6051 if (p != NULL) {
6052 assert(p == req_addr, "must be");
6053 small_page_write(p, size);
6054 os::Linux::release_memory_special_huge_tlbfs(p, size);
6055 }
6056 }
6057 }
6058
6059 // Case 3
6060 for (int i = 0; i < num_sizes; i++) {
6061 const size_t size = sizes[i];
6062 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6063 char* const req_addr = align_up(mapping2, alignment);
6064 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6065 // as the area around req_addr contains already existing mappings, the API should always
6066 // return NULL (as per contract, it cannot return another address)
6067 assert(p == NULL, "must be");
6068 }
6069 }
6070
6071 ::munmap(mapping2, mapping_size);
6072
6073 }
6074
6075 static void test_reserve_memory_special_huge_tlbfs() {
6076 if (!UseHugeTLBFS) {
6077 return;
6078 }
6079
6080 test_reserve_memory_special_huge_tlbfs_only();
6081 test_reserve_memory_special_huge_tlbfs_mixed();
6082 }
6083
6084 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6085 if (!UseSHM) {
6086 return;
6087 }
6088
6089 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6090
6091 if (addr != NULL) {
6092 assert(is_aligned(addr, alignment), "Check");
6093 assert(is_aligned(addr, os::large_page_size()), "Check");
6094
6095 small_page_write(addr, size);
6096
6097 os::Linux::release_memory_special_shm(addr, size);
6098 }
6099 }
6100
6101 static void test_reserve_memory_special_shm() {
6102 size_t lp = os::large_page_size();
6103 size_t ag = os::vm_allocation_granularity();
6104
6105 for (size_t size = ag; size < lp * 3; size += ag) {
6106 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6107 test_reserve_memory_special_shm(size, alignment);
6108 }
6109 }
6110 }
6111
6112 static void test() {
6113 test_reserve_memory_special_huge_tlbfs();
6114 test_reserve_memory_special_shm();
6115 }
6116 };
6117
6118 void TestReserveMemorySpecial_test() {
6119 TestReserveMemorySpecial::test();
6120 }
6121
6122 #endif