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 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
331 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
332 #else
333 #define DEFAULT_LIBPATH "/lib:/usr/lib"
334 #endif
335
336 // Base path of extensions installed on the system.
337 #define SYS_EXT_DIR "/usr/java/packages"
338 #define EXTENSIONS_DIR "/lib/ext"
339
340 // Buffer that fits several sprintfs.
341 // Note that the space for the colon and the trailing null are provided
342 // by the nulls included by the sizeof operator.
343 const size_t bufsize =
344 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
345 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
346 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
347
348 // sysclasspath, java_home, dll_dir
349 {
350 char *pslash;
351 os::jvm_path(buf, bufsize);
352
353 // Found the full path to libjvm.so.
354 // Now cut the path to <java_home>/jre if we can.
355 pslash = strrchr(buf, '/');
356 if (pslash != NULL) {
357 *pslash = '\0'; // Get rid of /libjvm.so.
358 }
359 pslash = strrchr(buf, '/');
360 if (pslash != NULL) {
361 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
362 }
363 Arguments::set_dll_dir(buf);
364
365 if (pslash != NULL) {
366 pslash = strrchr(buf, '/');
367 if (pslash != NULL) {
368 *pslash = '\0'; // Get rid of /lib.
369 }
370 }
371 Arguments::set_java_home(buf);
372 if (!set_boot_path('/', ':')) {
373 vm_exit_during_initialization("Failed setting boot class path.", NULL);
374 }
375 }
376
377 // Where to look for native libraries.
378 //
379 // Note: Due to a legacy implementation, most of the library path
380 // is set in the launcher. This was to accomodate linking restrictions
381 // on legacy Linux implementations (which are no longer supported).
382 // Eventually, all the library path setting will be done here.
383 //
384 // However, to prevent the proliferation of improperly built native
385 // libraries, the new path component /usr/java/packages is added here.
386 // Eventually, all the library path setting will be done here.
387 {
388 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
389 // should always exist (until the legacy problem cited above is
390 // addressed).
391 const char *v = ::getenv("LD_LIBRARY_PATH");
392 const char *v_colon = ":";
393 if (v == NULL) { v = ""; v_colon = ""; }
394 // That's +1 for the colon and +1 for the trailing '\0'.
395 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
396 strlen(v) + 1 +
397 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
398 mtInternal);
399 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
400 Arguments::set_library_path(ld_library_path);
401 FREE_C_HEAP_ARRAY(char, ld_library_path);
402 }
403
404 // Extensions directories.
405 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
406 Arguments::set_ext_dirs(buf);
407
408 FREE_C_HEAP_ARRAY(char, buf);
409
410 #undef DEFAULT_LIBPATH
411 #undef SYS_EXT_DIR
412 #undef EXTENSIONS_DIR
413 }
414
415 ////////////////////////////////////////////////////////////////////////////////
416 // breakpoint support
417
418 void os::breakpoint() {
419 BREAKPOINT;
420 }
421
422 extern "C" void breakpoint() {
423 // use debugger to set breakpoint here
424 }
425
426 ////////////////////////////////////////////////////////////////////////////////
427 // signal support
428
429 debug_only(static bool signal_sets_initialized = false);
430 static sigset_t unblocked_sigs, vm_sigs;
431
432 void os::Linux::signal_sets_init() {
433 // Should also have an assertion stating we are still single-threaded.
434 assert(!signal_sets_initialized, "Already initialized");
435 // Fill in signals that are necessarily unblocked for all threads in
436 // the VM. Currently, we unblock the following signals:
437 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
438 // by -Xrs (=ReduceSignalUsage));
439 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
440 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
441 // the dispositions or masks wrt these signals.
442 // Programs embedding the VM that want to use the above signals for their
443 // own purposes must, at this time, use the "-Xrs" option to prevent
444 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
445 // (See bug 4345157, and other related bugs).
446 // In reality, though, unblocking these signals is really a nop, since
447 // these signals are not blocked by default.
448 sigemptyset(&unblocked_sigs);
449 sigaddset(&unblocked_sigs, SIGILL);
450 sigaddset(&unblocked_sigs, SIGSEGV);
451 sigaddset(&unblocked_sigs, SIGBUS);
452 sigaddset(&unblocked_sigs, SIGFPE);
453 #if defined(PPC64)
454 sigaddset(&unblocked_sigs, SIGTRAP);
455 #endif
456 sigaddset(&unblocked_sigs, SR_signum);
457
458 if (!ReduceSignalUsage) {
459 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
460 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
461 }
462 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
464 }
465 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
466 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
467 }
468 }
469 // Fill in signals that are blocked by all but the VM thread.
470 sigemptyset(&vm_sigs);
471 if (!ReduceSignalUsage) {
472 sigaddset(&vm_sigs, BREAK_SIGNAL);
473 }
474 debug_only(signal_sets_initialized = true);
475
476 }
477
478 // These are signals that are unblocked while a thread is running Java.
479 // (For some reason, they get blocked by default.)
480 sigset_t* os::Linux::unblocked_signals() {
481 assert(signal_sets_initialized, "Not initialized");
482 return &unblocked_sigs;
483 }
484
485 // These are the signals that are blocked while a (non-VM) thread is
486 // running Java. Only the VM thread handles these signals.
487 sigset_t* os::Linux::vm_signals() {
488 assert(signal_sets_initialized, "Not initialized");
489 return &vm_sigs;
490 }
491
492 void os::Linux::hotspot_sigmask(Thread* thread) {
493
494 //Save caller's signal mask before setting VM signal mask
495 sigset_t caller_sigmask;
496 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
497
498 OSThread* osthread = thread->osthread();
499 osthread->set_caller_sigmask(caller_sigmask);
500
501 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
502
503 if (!ReduceSignalUsage) {
504 if (thread->is_VM_thread()) {
505 // Only the VM thread handles BREAK_SIGNAL ...
506 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
507 } else {
508 // ... all other threads block BREAK_SIGNAL
509 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
510 }
511 }
512 }
513
514 //////////////////////////////////////////////////////////////////////////////
515 // detecting pthread library
516
517 void os::Linux::libpthread_init() {
518 // Save glibc and pthread version strings.
519 #if !defined(_CS_GNU_LIBC_VERSION) || \
520 !defined(_CS_GNU_LIBPTHREAD_VERSION)
521 #error "glibc too old (< 2.3.2)"
522 #endif
523
524 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
525 assert(n > 0, "cannot retrieve glibc version");
526 char *str = (char *)malloc(n, mtInternal);
527 confstr(_CS_GNU_LIBC_VERSION, str, n);
528 os::Linux::set_glibc_version(str);
529
530 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
531 assert(n > 0, "cannot retrieve pthread version");
532 str = (char *)malloc(n, mtInternal);
533 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
534 os::Linux::set_libpthread_version(str);
535 }
536
537 /////////////////////////////////////////////////////////////////////////////
538 // thread stack expansion
539
540 // os::Linux::manually_expand_stack() takes care of expanding the thread
541 // stack. Note that this is normally not needed: pthread stacks allocate
542 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
543 // committed. Therefore it is not necessary to expand the stack manually.
544 //
545 // Manually expanding the stack was historically needed on LinuxThreads
546 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
547 // it is kept to deal with very rare corner cases:
548 //
549 // For one, user may run the VM on an own implementation of threads
550 // whose stacks are - like the old LinuxThreads - implemented using
551 // mmap(MAP_GROWSDOWN).
552 //
553 // Also, this coding may be needed if the VM is running on the primordial
554 // thread. Normally we avoid running on the primordial thread; however,
555 // user may still invoke the VM on the primordial thread.
556 //
557 // The following historical comment describes the details about running
558 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
559
560
561 // Force Linux kernel to expand current thread stack. If "bottom" is close
562 // to the stack guard, caller should block all signals.
563 //
564 // MAP_GROWSDOWN:
565 // A special mmap() flag that is used to implement thread stacks. It tells
566 // kernel that the memory region should extend downwards when needed. This
567 // allows early versions of LinuxThreads to only mmap the first few pages
568 // when creating a new thread. Linux kernel will automatically expand thread
569 // stack as needed (on page faults).
570 //
571 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
572 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
573 // region, it's hard to tell if the fault is due to a legitimate stack
574 // access or because of reading/writing non-exist memory (e.g. buffer
575 // overrun). As a rule, if the fault happens below current stack pointer,
576 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
577 // application (see Linux kernel fault.c).
578 //
579 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
580 // stack overflow detection.
581 //
582 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
583 // not use MAP_GROWSDOWN.
584 //
585 // To get around the problem and allow stack banging on Linux, we need to
586 // manually expand thread stack after receiving the SIGSEGV.
587 //
588 // There are two ways to expand thread stack to address "bottom", we used
589 // both of them in JVM before 1.5:
590 // 1. adjust stack pointer first so that it is below "bottom", and then
591 // touch "bottom"
592 // 2. mmap() the page in question
593 //
594 // Now alternate signal stack is gone, it's harder to use 2. For instance,
595 // if current sp is already near the lower end of page 101, and we need to
596 // call mmap() to map page 100, it is possible that part of the mmap() frame
597 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
598 // That will destroy the mmap() frame and cause VM to crash.
599 //
600 // The following code works by adjusting sp first, then accessing the "bottom"
601 // page to force a page fault. Linux kernel will then automatically expand the
602 // stack mapping.
603 //
604 // _expand_stack_to() assumes its frame size is less than page size, which
605 // should always be true if the function is not inlined.
606
607 static void NOINLINE _expand_stack_to(address bottom) {
608 address sp;
609 size_t size;
610 volatile char *p;
611
612 // Adjust bottom to point to the largest address within the same page, it
613 // gives us a one-page buffer if alloca() allocates slightly more memory.
614 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
615 bottom += os::Linux::page_size() - 1;
616
617 // sp might be slightly above current stack pointer; if that's the case, we
618 // will alloca() a little more space than necessary, which is OK. Don't use
619 // os::current_stack_pointer(), as its result can be slightly below current
620 // stack pointer, causing us to not alloca enough to reach "bottom".
621 sp = (address)&sp;
622
623 if (sp > bottom) {
624 size = sp - bottom;
625 p = (volatile char *)alloca(size);
626 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
627 p[0] = '\0';
628 }
629 }
630
631 void os::Linux::expand_stack_to(address bottom) {
632 _expand_stack_to(bottom);
633 }
634
635 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
636 assert(t!=NULL, "just checking");
637 assert(t->osthread()->expanding_stack(), "expand should be set");
638 assert(t->stack_base() != NULL, "stack_base was not initialized");
639
640 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
641 sigset_t mask_all, old_sigset;
642 sigfillset(&mask_all);
643 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
644 _expand_stack_to(addr);
645 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
646 return true;
647 }
648 return false;
649 }
650
651 //////////////////////////////////////////////////////////////////////////////
652 // create new thread
653
654 // Thread start routine for all newly created threads
655 static void *thread_native_entry(Thread *thread) {
656 // Try to randomize the cache line index of hot stack frames.
657 // This helps when threads of the same stack traces evict each other's
658 // cache lines. The threads can be either from the same JVM instance, or
659 // from different JVM instances. The benefit is especially true for
660 // processors with hyperthreading technology.
661 static int counter = 0;
662 int pid = os::current_process_id();
663 alloca(((pid ^ counter++) & 7) * 128);
664
665 thread->initialize_thread_current();
666
667 OSThread* osthread = thread->osthread();
668 Monitor* sync = osthread->startThread_lock();
669
670 osthread->set_thread_id(os::current_thread_id());
671
672 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
673 os::current_thread_id(), (uintx) pthread_self());
674
675 if (UseNUMA) {
676 int lgrp_id = os::numa_get_group_id();
677 if (lgrp_id != -1) {
678 thread->set_lgrp_id(lgrp_id);
679 }
680 }
681 // initialize signal mask for this thread
682 os::Linux::hotspot_sigmask(thread);
683
684 // initialize floating point control register
685 os::Linux::init_thread_fpu_state();
686
687 // handshaking with parent thread
688 {
689 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
690
691 // notify parent thread
692 osthread->set_state(INITIALIZED);
693 sync->notify_all();
694
695 // wait until os::start_thread()
696 while (osthread->get_state() == INITIALIZED) {
697 sync->wait(Mutex::_no_safepoint_check_flag);
698 }
699 }
700
701 // call one more level start routine
702 thread->run();
703
704 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
705 os::current_thread_id(), (uintx) pthread_self());
706
707 // If a thread has not deleted itself ("delete this") as part of its
708 // termination sequence, we have to ensure thread-local-storage is
709 // cleared before we actually terminate. No threads should ever be
710 // deleted asynchronously with respect to their termination.
711 if (Thread::current_or_null_safe() != NULL) {
712 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
713 thread->clear_thread_current();
714 }
715
716 return 0;
717 }
718
719 bool os::create_thread(Thread* thread, ThreadType thr_type,
720 size_t req_stack_size) {
721 assert(thread->osthread() == NULL, "caller responsible");
722
723 // Allocate the OSThread object
724 OSThread* osthread = new OSThread(NULL, NULL);
725 if (osthread == NULL) {
726 return false;
727 }
728
729 // set the correct thread state
730 osthread->set_thread_type(thr_type);
731
732 // Initial state is ALLOCATED but not INITIALIZED
733 osthread->set_state(ALLOCATED);
734
735 thread->set_osthread(osthread);
736
737 // init thread attributes
738 pthread_attr_t attr;
739 pthread_attr_init(&attr);
740 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
741
742 // Calculate stack size if it's not specified by caller.
743 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
744 // In the Linux NPTL pthread implementation the guard size mechanism
745 // is not implemented properly. The posix standard requires adding
746 // the size of the guard pages to the stack size, instead Linux
747 // takes the space out of 'stacksize'. Thus we adapt the requested
748 // stack_size by the size of the guard pages to mimick proper
749 // behaviour. However, be careful not to end up with a size
750 // of zero due to overflow. Don't add the guard page in that case.
751 size_t guard_size = os::Linux::default_guard_size(thr_type);
752 if (stack_size <= SIZE_MAX - guard_size) {
753 stack_size += guard_size;
754 }
755 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
756
757 int status = pthread_attr_setstacksize(&attr, stack_size);
758 assert_status(status == 0, status, "pthread_attr_setstacksize");
759
760 // Configure glibc guard page.
761 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
762
763 ThreadState state;
764
765 {
766 pthread_t tid;
767 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
768
769 char buf[64];
770 if (ret == 0) {
771 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
772 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
773 } else {
774 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
775 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
776 }
777
778 pthread_attr_destroy(&attr);
779
780 if (ret != 0) {
781 // Need to clean up stuff we've allocated so far
782 thread->set_osthread(NULL);
783 delete osthread;
784 return false;
785 }
786
787 // Store pthread info into the OSThread
788 osthread->set_pthread_id(tid);
789
790 // Wait until child thread is either initialized or aborted
791 {
792 Monitor* sync_with_child = osthread->startThread_lock();
793 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
794 while ((state = osthread->get_state()) == ALLOCATED) {
795 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
796 }
797 }
798 }
799
800 // Aborted due to thread limit being reached
801 if (state == ZOMBIE) {
802 thread->set_osthread(NULL);
803 delete osthread;
804 return false;
805 }
806
807 // The thread is returned suspended (in state INITIALIZED),
808 // and is started higher up in the call chain
809 assert(state == INITIALIZED, "race condition");
810 return true;
811 }
812
813 /////////////////////////////////////////////////////////////////////////////
814 // attach existing thread
815
816 // bootstrap the main thread
817 bool os::create_main_thread(JavaThread* thread) {
818 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
819 return create_attached_thread(thread);
820 }
821
822 bool os::create_attached_thread(JavaThread* thread) {
823 #ifdef ASSERT
824 thread->verify_not_published();
825 #endif
826
827 // Allocate the OSThread object
828 OSThread* osthread = new OSThread(NULL, NULL);
829
830 if (osthread == NULL) {
831 return false;
832 }
833
834 // Store pthread info into the OSThread
835 osthread->set_thread_id(os::Linux::gettid());
836 osthread->set_pthread_id(::pthread_self());
837
838 // initialize floating point control register
839 os::Linux::init_thread_fpu_state();
840
841 // Initial thread state is RUNNABLE
842 osthread->set_state(RUNNABLE);
843
844 thread->set_osthread(osthread);
845
846 if (UseNUMA) {
847 int lgrp_id = os::numa_get_group_id();
848 if (lgrp_id != -1) {
849 thread->set_lgrp_id(lgrp_id);
850 }
851 }
852
853 if (os::is_primordial_thread()) {
854 // If current thread is primordial thread, its stack is mapped on demand,
855 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
856 // the entire stack region to avoid SEGV in stack banging.
857 // It is also useful to get around the heap-stack-gap problem on SuSE
858 // kernel (see 4821821 for details). We first expand stack to the top
859 // of yellow zone, then enable stack yellow zone (order is significant,
860 // enabling yellow zone first will crash JVM on SuSE Linux), so there
861 // is no gap between the last two virtual memory regions.
862
863 JavaThread *jt = (JavaThread *)thread;
864 address addr = jt->stack_reserved_zone_base();
865 assert(addr != NULL, "initialization problem?");
866 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
867
868 osthread->set_expanding_stack();
869 os::Linux::manually_expand_stack(jt, addr);
870 osthread->clear_expanding_stack();
871 }
872
873 // initialize signal mask for this thread
874 // and save the caller's signal mask
875 os::Linux::hotspot_sigmask(thread);
876
877 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
878 os::current_thread_id(), (uintx) pthread_self());
879
880 return true;
881 }
882
883 void os::pd_start_thread(Thread* thread) {
884 OSThread * osthread = thread->osthread();
885 assert(osthread->get_state() != INITIALIZED, "just checking");
886 Monitor* sync_with_child = osthread->startThread_lock();
887 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
888 sync_with_child->notify();
889 }
890
891 // Free Linux resources related to the OSThread
892 void os::free_thread(OSThread* osthread) {
893 assert(osthread != NULL, "osthread not set");
894
895 // We are told to free resources of the argument thread,
896 // but we can only really operate on the current thread.
897 assert(Thread::current()->osthread() == osthread,
898 "os::free_thread but not current thread");
899
900 #ifdef ASSERT
901 sigset_t current;
902 sigemptyset(¤t);
903 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
904 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
905 #endif
906
907 // Restore caller's signal mask
908 sigset_t sigmask = osthread->caller_sigmask();
909 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
910
911 delete osthread;
912 }
913
914 //////////////////////////////////////////////////////////////////////////////
915 // primordial thread
916
917 // Check if current thread is the primordial thread, similar to Solaris thr_main.
918 bool os::is_primordial_thread(void) {
919 if (suppress_primordial_thread_resolution) {
920 return false;
921 }
922 char dummy;
923 // If called before init complete, thread stack bottom will be null.
924 // Can be called if fatal error occurs before initialization.
925 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
926 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
927 os::Linux::initial_thread_stack_size() != 0,
928 "os::init did not locate primordial thread's stack region");
929 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
930 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
931 os::Linux::initial_thread_stack_size()) {
932 return true;
933 } else {
934 return false;
935 }
936 }
937
938 // Find the virtual memory area that contains addr
939 static bool find_vma(address addr, address* vma_low, address* vma_high) {
940 FILE *fp = fopen("/proc/self/maps", "r");
941 if (fp) {
942 address low, high;
943 while (!feof(fp)) {
944 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
945 if (low <= addr && addr < high) {
946 if (vma_low) *vma_low = low;
947 if (vma_high) *vma_high = high;
948 fclose(fp);
949 return true;
950 }
951 }
952 for (;;) {
953 int ch = fgetc(fp);
954 if (ch == EOF || ch == (int)'\n') break;
955 }
956 }
957 fclose(fp);
958 }
959 return false;
960 }
961
962 // Locate primordial thread stack. This special handling of primordial thread stack
963 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
964 // bogus value for the primordial process thread. While the launcher has created
965 // the VM in a new thread since JDK 6, we still have to allow for the use of the
966 // JNI invocation API from a primordial thread.
967 void os::Linux::capture_initial_stack(size_t max_size) {
968
969 // max_size is either 0 (which means accept OS default for thread stacks) or
970 // a user-specified value known to be at least the minimum needed. If we
971 // are actually on the primordial thread we can make it appear that we have a
972 // smaller max_size stack by inserting the guard pages at that location. But we
973 // cannot do anything to emulate a larger stack than what has been provided by
974 // the OS or threading library. In fact if we try to use a stack greater than
975 // what is set by rlimit then we will crash the hosting process.
976
977 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
978 // If this is "unlimited" then it will be a huge value.
979 struct rlimit rlim;
980 getrlimit(RLIMIT_STACK, &rlim);
981 size_t stack_size = rlim.rlim_cur;
982
983 // 6308388: a bug in ld.so will relocate its own .data section to the
984 // lower end of primordial stack; reduce ulimit -s value a little bit
985 // so we won't install guard page on ld.so's data section.
986 // But ensure we don't underflow the stack size - allow 1 page spare
987 if (stack_size >= (size_t)(3 * page_size())) {
988 stack_size -= 2 * page_size();
989 }
990
991 // Try to figure out where the stack base (top) is. This is harder.
992 //
993 // When an application is started, glibc saves the initial stack pointer in
994 // a global variable "__libc_stack_end", which is then used by system
995 // libraries. __libc_stack_end should be pretty close to stack top. The
996 // variable is available since the very early days. However, because it is
997 // a private interface, it could disappear in the future.
998 //
999 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1000 // to __libc_stack_end, it is very close to stack top, but isn't the real
1001 // stack top. Note that /proc may not exist if VM is running as a chroot
1002 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1003 // /proc/<pid>/stat could change in the future (though unlikely).
1004 //
1005 // We try __libc_stack_end first. If that doesn't work, look for
1006 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1007 // as a hint, which should work well in most cases.
1008
1009 uintptr_t stack_start;
1010
1011 // try __libc_stack_end first
1012 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1013 if (p && *p) {
1014 stack_start = *p;
1015 } else {
1016 // see if we can get the start_stack field from /proc/self/stat
1017 FILE *fp;
1018 int pid;
1019 char state;
1020 int ppid;
1021 int pgrp;
1022 int session;
1023 int nr;
1024 int tpgrp;
1025 unsigned long flags;
1026 unsigned long minflt;
1027 unsigned long cminflt;
1028 unsigned long majflt;
1029 unsigned long cmajflt;
1030 unsigned long utime;
1031 unsigned long stime;
1032 long cutime;
1033 long cstime;
1034 long prio;
1035 long nice;
1036 long junk;
1037 long it_real;
1038 uintptr_t start;
1039 uintptr_t vsize;
1040 intptr_t rss;
1041 uintptr_t rsslim;
1042 uintptr_t scodes;
1043 uintptr_t ecode;
1044 int i;
1045
1046 // Figure what the primordial thread stack base is. Code is inspired
1047 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1048 // followed by command name surrounded by parentheses, state, etc.
1049 char stat[2048];
1050 int statlen;
1051
1052 fp = fopen("/proc/self/stat", "r");
1053 if (fp) {
1054 statlen = fread(stat, 1, 2047, fp);
1055 stat[statlen] = '\0';
1056 fclose(fp);
1057
1058 // Skip pid and the command string. Note that we could be dealing with
1059 // weird command names, e.g. user could decide to rename java launcher
1060 // to "java 1.4.2 :)", then the stat file would look like
1061 // 1234 (java 1.4.2 :)) R ... ...
1062 // We don't really need to know the command string, just find the last
1063 // occurrence of ")" and then start parsing from there. See bug 4726580.
1064 char * s = strrchr(stat, ')');
1065
1066 i = 0;
1067 if (s) {
1068 // Skip blank chars
1069 do { s++; } while (s && isspace(*s));
1070
1071 #define _UFM UINTX_FORMAT
1072 #define _DFM INTX_FORMAT
1073
1074 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1075 // 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
1076 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,
1077 &state, // 3 %c
1078 &ppid, // 4 %d
1079 &pgrp, // 5 %d
1080 &session, // 6 %d
1081 &nr, // 7 %d
1082 &tpgrp, // 8 %d
1083 &flags, // 9 %lu
1084 &minflt, // 10 %lu
1085 &cminflt, // 11 %lu
1086 &majflt, // 12 %lu
1087 &cmajflt, // 13 %lu
1088 &utime, // 14 %lu
1089 &stime, // 15 %lu
1090 &cutime, // 16 %ld
1091 &cstime, // 17 %ld
1092 &prio, // 18 %ld
1093 &nice, // 19 %ld
1094 &junk, // 20 %ld
1095 &it_real, // 21 %ld
1096 &start, // 22 UINTX_FORMAT
1097 &vsize, // 23 UINTX_FORMAT
1098 &rss, // 24 INTX_FORMAT
1099 &rsslim, // 25 UINTX_FORMAT
1100 &scodes, // 26 UINTX_FORMAT
1101 &ecode, // 27 UINTX_FORMAT
1102 &stack_start); // 28 UINTX_FORMAT
1103 }
1104
1105 #undef _UFM
1106 #undef _DFM
1107
1108 if (i != 28 - 2) {
1109 assert(false, "Bad conversion from /proc/self/stat");
1110 // product mode - assume we are the primordial thread, good luck in the
1111 // embedded case.
1112 warning("Can't detect primordial thread stack location - bad conversion");
1113 stack_start = (uintptr_t) &rlim;
1114 }
1115 } else {
1116 // For some reason we can't open /proc/self/stat (for example, running on
1117 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1118 // most cases, so don't abort:
1119 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1120 stack_start = (uintptr_t) &rlim;
1121 }
1122 }
1123
1124 // Now we have a pointer (stack_start) very close to the stack top, the
1125 // next thing to do is to figure out the exact location of stack top. We
1126 // can find out the virtual memory area that contains stack_start by
1127 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1128 // and its upper limit is the real stack top. (again, this would fail if
1129 // running inside chroot, because /proc may not exist.)
1130
1131 uintptr_t stack_top;
1132 address low, high;
1133 if (find_vma((address)stack_start, &low, &high)) {
1134 // success, "high" is the true stack top. (ignore "low", because initial
1135 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1136 stack_top = (uintptr_t)high;
1137 } else {
1138 // failed, likely because /proc/self/maps does not exist
1139 warning("Can't detect primordial thread stack location - find_vma failed");
1140 // best effort: stack_start is normally within a few pages below the real
1141 // stack top, use it as stack top, and reduce stack size so we won't put
1142 // guard page outside stack.
1143 stack_top = stack_start;
1144 stack_size -= 16 * page_size();
1145 }
1146
1147 // stack_top could be partially down the page so align it
1148 stack_top = align_up(stack_top, page_size());
1149
1150 // Allowed stack value is minimum of max_size and what we derived from rlimit
1151 if (max_size > 0) {
1152 _initial_thread_stack_size = MIN2(max_size, stack_size);
1153 } else {
1154 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1155 // clamp it at 8MB as we do on Solaris
1156 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1157 }
1158 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1159 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1160
1161 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1162
1163 if (log_is_enabled(Info, os, thread)) {
1164 // See if we seem to be on primordial process thread
1165 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1166 uintptr_t(&rlim) < stack_top;
1167
1168 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1169 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1170 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1171 stack_top, intptr_t(_initial_thread_stack_bottom));
1172 }
1173 }
1174
1175 ////////////////////////////////////////////////////////////////////////////////
1176 // time support
1177
1178 #ifndef SUPPORTS_CLOCK_MONOTONIC
1179 #error "Build platform doesn't support clock_gettime and related functionality"
1180 #endif
1181
1182 // Time since start-up in seconds to a fine granularity.
1183 // Used by VMSelfDestructTimer and the MemProfiler.
1184 double os::elapsedTime() {
1185
1186 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1187 }
1188
1189 jlong os::elapsed_counter() {
1190 return javaTimeNanos() - initial_time_count;
1191 }
1192
1193 jlong os::elapsed_frequency() {
1194 return NANOSECS_PER_SEC; // nanosecond resolution
1195 }
1196
1197 bool os::supports_vtime() { return true; }
1198 bool os::enable_vtime() { return false; }
1199 bool os::vtime_enabled() { return false; }
1200
1201 double os::elapsedVTime() {
1202 struct rusage usage;
1203 int retval = getrusage(RUSAGE_THREAD, &usage);
1204 if (retval == 0) {
1205 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);
1206 } else {
1207 // better than nothing, but not much
1208 return elapsedTime();
1209 }
1210 }
1211
1212 jlong os::javaTimeMillis() {
1213 timeval time;
1214 int status = gettimeofday(&time, NULL);
1215 assert(status != -1, "linux error");
1216 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1217 }
1218
1219 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1220 timeval time;
1221 int status = gettimeofday(&time, NULL);
1222 assert(status != -1, "linux error");
1223 seconds = jlong(time.tv_sec);
1224 nanos = jlong(time.tv_usec) * 1000;
1225 }
1226
1227 void os::Linux::fast_thread_clock_init() {
1228 if (!UseLinuxPosixThreadCPUClocks) {
1229 return;
1230 }
1231 clockid_t clockid;
1232 struct timespec tp;
1233 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1234 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1235
1236 // Switch to using fast clocks for thread cpu time if
1237 // the clock_getres() returns 0 error code.
1238 // Note, that some kernels may support the current thread
1239 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1240 // returned by the pthread_getcpuclockid().
1241 // If the fast Posix clocks are supported then the clock_getres()
1242 // must return at least tp.tv_sec == 0 which means a resolution
1243 // better than 1 sec. This is extra check for reliability.
1244
1245 if (pthread_getcpuclockid_func &&
1246 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1247 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1248 _supports_fast_thread_cpu_time = true;
1249 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1250 }
1251 }
1252
1253 jlong os::javaTimeNanos() {
1254 if (os::supports_monotonic_clock()) {
1255 struct timespec tp;
1256 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1257 assert(status == 0, "gettime error");
1258 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1259 return result;
1260 } else {
1261 timeval time;
1262 int status = gettimeofday(&time, NULL);
1263 assert(status != -1, "linux error");
1264 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1265 return 1000 * usecs;
1266 }
1267 }
1268
1269 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1270 if (os::supports_monotonic_clock()) {
1271 info_ptr->max_value = ALL_64_BITS;
1272
1273 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1274 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1275 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1276 } else {
1277 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1278 info_ptr->max_value = ALL_64_BITS;
1279
1280 // gettimeofday is a real time clock so it skips
1281 info_ptr->may_skip_backward = true;
1282 info_ptr->may_skip_forward = true;
1283 }
1284
1285 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1286 }
1287
1288 // Return the real, user, and system times in seconds from an
1289 // arbitrary fixed point in the past.
1290 bool os::getTimesSecs(double* process_real_time,
1291 double* process_user_time,
1292 double* process_system_time) {
1293 struct tms ticks;
1294 clock_t real_ticks = times(&ticks);
1295
1296 if (real_ticks == (clock_t) (-1)) {
1297 return false;
1298 } else {
1299 double ticks_per_second = (double) clock_tics_per_sec;
1300 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1301 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1302 *process_real_time = ((double) real_ticks) / ticks_per_second;
1303
1304 return true;
1305 }
1306 }
1307
1308
1309 char * os::local_time_string(char *buf, size_t buflen) {
1310 struct tm t;
1311 time_t long_time;
1312 time(&long_time);
1313 localtime_r(&long_time, &t);
1314 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1315 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1316 t.tm_hour, t.tm_min, t.tm_sec);
1317 return buf;
1318 }
1319
1320 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1321 return localtime_r(clock, res);
1322 }
1323
1324 ////////////////////////////////////////////////////////////////////////////////
1325 // runtime exit support
1326
1327 // Note: os::shutdown() might be called very early during initialization, or
1328 // called from signal handler. Before adding something to os::shutdown(), make
1329 // sure it is async-safe and can handle partially initialized VM.
1330 void os::shutdown() {
1331
1332 // allow PerfMemory to attempt cleanup of any persistent resources
1333 perfMemory_exit();
1334
1335 // needs to remove object in file system
1336 AttachListener::abort();
1337
1338 // flush buffered output, finish log files
1339 ostream_abort();
1340
1341 // Check for abort hook
1342 abort_hook_t abort_hook = Arguments::abort_hook();
1343 if (abort_hook != NULL) {
1344 abort_hook();
1345 }
1346
1347 }
1348
1349 // Note: os::abort() might be called very early during initialization, or
1350 // called from signal handler. Before adding something to os::abort(), make
1351 // sure it is async-safe and can handle partially initialized VM.
1352 void os::abort(bool dump_core, void* siginfo, const void* context) {
1353 os::shutdown();
1354 if (dump_core) {
1355 #ifndef PRODUCT
1356 fdStream out(defaultStream::output_fd());
1357 out.print_raw("Current thread is ");
1358 char buf[16];
1359 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1360 out.print_raw_cr(buf);
1361 out.print_raw_cr("Dumping core ...");
1362 #endif
1363 ::abort(); // dump core
1364 }
1365
1366 ::exit(1);
1367 }
1368
1369 // Die immediately, no exit hook, no abort hook, no cleanup.
1370 void os::die() {
1371 ::abort();
1372 }
1373
1374 // thread_id is kernel thread id (similar to Solaris LWP id)
1375 intx os::current_thread_id() { return os::Linux::gettid(); }
1376 int os::current_process_id() {
1377 return ::getpid();
1378 }
1379
1380 // DLL functions
1381
1382 const char* os::dll_file_extension() { return ".so"; }
1383
1384 // This must be hard coded because it's the system's temporary
1385 // directory not the java application's temp directory, ala java.io.tmpdir.
1386 const char* os::get_temp_directory() { return "/tmp"; }
1387
1388 static bool file_exists(const char* filename) {
1389 struct stat statbuf;
1390 if (filename == NULL || strlen(filename) == 0) {
1391 return false;
1392 }
1393 return os::stat(filename, &statbuf) == 0;
1394 }
1395
1396 // check if addr is inside libjvm.so
1397 bool os::address_is_in_vm(address addr) {
1398 static address libjvm_base_addr;
1399 Dl_info dlinfo;
1400
1401 if (libjvm_base_addr == NULL) {
1402 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1403 libjvm_base_addr = (address)dlinfo.dli_fbase;
1404 }
1405 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1406 }
1407
1408 if (dladdr((void *)addr, &dlinfo) != 0) {
1409 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1410 }
1411
1412 return false;
1413 }
1414
1415 bool os::dll_address_to_function_name(address addr, char *buf,
1416 int buflen, int *offset,
1417 bool demangle) {
1418 // buf is not optional, but offset is optional
1419 assert(buf != NULL, "sanity check");
1420
1421 Dl_info dlinfo;
1422
1423 if (dladdr((void*)addr, &dlinfo) != 0) {
1424 // see if we have a matching symbol
1425 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1426 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1427 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1428 }
1429 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1430 return true;
1431 }
1432 // no matching symbol so try for just file info
1433 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1434 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1435 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1436 return true;
1437 }
1438 }
1439 }
1440
1441 buf[0] = '\0';
1442 if (offset != NULL) *offset = -1;
1443 return false;
1444 }
1445
1446 struct _address_to_library_name {
1447 address addr; // input : memory address
1448 size_t buflen; // size of fname
1449 char* fname; // output: library name
1450 address base; // library base addr
1451 };
1452
1453 static int address_to_library_name_callback(struct dl_phdr_info *info,
1454 size_t size, void *data) {
1455 int i;
1456 bool found = false;
1457 address libbase = NULL;
1458 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1459
1460 // iterate through all loadable segments
1461 for (i = 0; i < info->dlpi_phnum; i++) {
1462 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1463 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1464 // base address of a library is the lowest address of its loaded
1465 // segments.
1466 if (libbase == NULL || libbase > segbase) {
1467 libbase = segbase;
1468 }
1469 // see if 'addr' is within current segment
1470 if (segbase <= d->addr &&
1471 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1472 found = true;
1473 }
1474 }
1475 }
1476
1477 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1478 // so dll_address_to_library_name() can fall through to use dladdr() which
1479 // can figure out executable name from argv[0].
1480 if (found && info->dlpi_name && info->dlpi_name[0]) {
1481 d->base = libbase;
1482 if (d->fname) {
1483 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1484 }
1485 return 1;
1486 }
1487 return 0;
1488 }
1489
1490 bool os::dll_address_to_library_name(address addr, char* buf,
1491 int buflen, int* offset) {
1492 // buf is not optional, but offset is optional
1493 assert(buf != NULL, "sanity check");
1494
1495 Dl_info dlinfo;
1496 struct _address_to_library_name data;
1497
1498 // There is a bug in old glibc dladdr() implementation that it could resolve
1499 // to wrong library name if the .so file has a base address != NULL. Here
1500 // we iterate through the program headers of all loaded libraries to find
1501 // out which library 'addr' really belongs to. This workaround can be
1502 // removed once the minimum requirement for glibc is moved to 2.3.x.
1503 data.addr = addr;
1504 data.fname = buf;
1505 data.buflen = buflen;
1506 data.base = NULL;
1507 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1508
1509 if (rslt) {
1510 // buf already contains library name
1511 if (offset) *offset = addr - data.base;
1512 return true;
1513 }
1514 if (dladdr((void*)addr, &dlinfo) != 0) {
1515 if (dlinfo.dli_fname != NULL) {
1516 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1517 }
1518 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1519 *offset = addr - (address)dlinfo.dli_fbase;
1520 }
1521 return true;
1522 }
1523
1524 buf[0] = '\0';
1525 if (offset) *offset = -1;
1526 return false;
1527 }
1528
1529 // Loads .dll/.so and
1530 // in case of error it checks if .dll/.so was built for the
1531 // same architecture as Hotspot is running on
1532
1533
1534 // Remember the stack's state. The Linux dynamic linker will change
1535 // the stack to 'executable' at most once, so we must safepoint only once.
1536 bool os::Linux::_stack_is_executable = false;
1537
1538 // VM operation that loads a library. This is necessary if stack protection
1539 // of the Java stacks can be lost during loading the library. If we
1540 // do not stop the Java threads, they can stack overflow before the stacks
1541 // are protected again.
1542 class VM_LinuxDllLoad: public VM_Operation {
1543 private:
1544 const char *_filename;
1545 char *_ebuf;
1546 int _ebuflen;
1547 void *_lib;
1548 public:
1549 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1550 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1551 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1552 void doit() {
1553 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1554 os::Linux::_stack_is_executable = true;
1555 }
1556 void* loaded_library() { return _lib; }
1557 };
1558
1559 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1560 void * result = NULL;
1561 bool load_attempted = false;
1562
1563 // Check whether the library to load might change execution rights
1564 // of the stack. If they are changed, the protection of the stack
1565 // guard pages will be lost. We need a safepoint to fix this.
1566 //
1567 // See Linux man page execstack(8) for more info.
1568 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1569 if (!ElfFile::specifies_noexecstack(filename)) {
1570 if (!is_init_completed()) {
1571 os::Linux::_stack_is_executable = true;
1572 // This is OK - No Java threads have been created yet, and hence no
1573 // stack guard pages to fix.
1574 //
1575 // Dynamic loader will make all stacks executable after
1576 // this function returns, and will not do that again.
1577 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1578 } else {
1579 warning("You have loaded library %s which might have disabled stack guard. "
1580 "The VM will try to fix the stack guard now.\n"
1581 "It's highly recommended that you fix the library with "
1582 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1583 filename);
1584
1585 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1586 JavaThread *jt = JavaThread::current();
1587 if (jt->thread_state() != _thread_in_native) {
1588 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1589 // that requires ExecStack. Cannot enter safe point. Let's give up.
1590 warning("Unable to fix stack guard. Giving up.");
1591 } else {
1592 if (!LoadExecStackDllInVMThread) {
1593 // This is for the case where the DLL has an static
1594 // constructor function that executes JNI code. We cannot
1595 // load such DLLs in the VMThread.
1596 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1597 }
1598
1599 ThreadInVMfromNative tiv(jt);
1600 debug_only(VMNativeEntryWrapper vew;)
1601
1602 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1603 VMThread::execute(&op);
1604 if (LoadExecStackDllInVMThread) {
1605 result = op.loaded_library();
1606 }
1607 load_attempted = true;
1608 }
1609 }
1610 }
1611 }
1612
1613 if (!load_attempted) {
1614 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1615 }
1616
1617 if (result != NULL) {
1618 // Successful loading
1619 return result;
1620 }
1621
1622 Elf32_Ehdr elf_head;
1623 int diag_msg_max_length=ebuflen-strlen(ebuf);
1624 char* diag_msg_buf=ebuf+strlen(ebuf);
1625
1626 if (diag_msg_max_length==0) {
1627 // No more space in ebuf for additional diagnostics message
1628 return NULL;
1629 }
1630
1631
1632 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1633
1634 if (file_descriptor < 0) {
1635 // Can't open library, report dlerror() message
1636 return NULL;
1637 }
1638
1639 bool failed_to_read_elf_head=
1640 (sizeof(elf_head)!=
1641 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1642
1643 ::close(file_descriptor);
1644 if (failed_to_read_elf_head) {
1645 // file i/o error - report dlerror() msg
1646 return NULL;
1647 }
1648
1649 typedef struct {
1650 Elf32_Half code; // Actual value as defined in elf.h
1651 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1652 unsigned char elf_class; // 32 or 64 bit
1653 unsigned char endianess; // MSB or LSB
1654 char* name; // String representation
1655 } arch_t;
1656
1657 #ifndef EM_486
1658 #define EM_486 6 /* Intel 80486 */
1659 #endif
1660 #ifndef EM_AARCH64
1661 #define EM_AARCH64 183 /* ARM AARCH64 */
1662 #endif
1663
1664 static const arch_t arch_array[]={
1665 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1666 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1667 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1668 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1669 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1670 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1671 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1672 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1673 #if defined(VM_LITTLE_ENDIAN)
1674 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1675 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1676 #else
1677 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1678 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1679 #endif
1680 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1681 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1682 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1683 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1684 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1685 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1686 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1687 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1688 };
1689
1690 #if (defined IA32)
1691 static Elf32_Half running_arch_code=EM_386;
1692 #elif (defined AMD64) || (defined X32)
1693 static Elf32_Half running_arch_code=EM_X86_64;
1694 #elif (defined IA64)
1695 static Elf32_Half running_arch_code=EM_IA_64;
1696 #elif (defined __sparc) && (defined _LP64)
1697 static Elf32_Half running_arch_code=EM_SPARCV9;
1698 #elif (defined __sparc) && (!defined _LP64)
1699 static Elf32_Half running_arch_code=EM_SPARC;
1700 #elif (defined __powerpc64__)
1701 static Elf32_Half running_arch_code=EM_PPC64;
1702 #elif (defined __powerpc__)
1703 static Elf32_Half running_arch_code=EM_PPC;
1704 #elif (defined AARCH64)
1705 static Elf32_Half running_arch_code=EM_AARCH64;
1706 #elif (defined ARM)
1707 static Elf32_Half running_arch_code=EM_ARM;
1708 #elif (defined S390)
1709 static Elf32_Half running_arch_code=EM_S390;
1710 #elif (defined ALPHA)
1711 static Elf32_Half running_arch_code=EM_ALPHA;
1712 #elif (defined MIPSEL)
1713 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1714 #elif (defined PARISC)
1715 static Elf32_Half running_arch_code=EM_PARISC;
1716 #elif (defined MIPS)
1717 static Elf32_Half running_arch_code=EM_MIPS;
1718 #elif (defined M68K)
1719 static Elf32_Half running_arch_code=EM_68K;
1720 #elif (defined SH)
1721 static Elf32_Half running_arch_code=EM_SH;
1722 #else
1723 #error Method os::dll_load requires that one of following is defined:\
1724 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1725 #endif
1726
1727 // Identify compatability class for VM's architecture and library's architecture
1728 // Obtain string descriptions for architectures
1729
1730 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1731 int running_arch_index=-1;
1732
1733 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1734 if (running_arch_code == arch_array[i].code) {
1735 running_arch_index = i;
1736 }
1737 if (lib_arch.code == arch_array[i].code) {
1738 lib_arch.compat_class = arch_array[i].compat_class;
1739 lib_arch.name = arch_array[i].name;
1740 }
1741 }
1742
1743 assert(running_arch_index != -1,
1744 "Didn't find running architecture code (running_arch_code) in arch_array");
1745 if (running_arch_index == -1) {
1746 // Even though running architecture detection failed
1747 // we may still continue with reporting dlerror() message
1748 return NULL;
1749 }
1750
1751 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1752 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1753 return NULL;
1754 }
1755
1756 #ifndef S390
1757 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1758 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1759 return NULL;
1760 }
1761 #endif // !S390
1762
1763 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1764 if (lib_arch.name!=NULL) {
1765 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1766 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1767 lib_arch.name, arch_array[running_arch_index].name);
1768 } else {
1769 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1770 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1771 lib_arch.code,
1772 arch_array[running_arch_index].name);
1773 }
1774 }
1775
1776 return NULL;
1777 }
1778
1779 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1780 int ebuflen) {
1781 void * result = ::dlopen(filename, RTLD_LAZY);
1782 if (result == NULL) {
1783 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1784 ebuf[ebuflen-1] = '\0';
1785 }
1786 return result;
1787 }
1788
1789 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1790 int ebuflen) {
1791 void * result = NULL;
1792 if (LoadExecStackDllInVMThread) {
1793 result = dlopen_helper(filename, ebuf, ebuflen);
1794 }
1795
1796 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1797 // library that requires an executable stack, or which does not have this
1798 // stack attribute set, dlopen changes the stack attribute to executable. The
1799 // read protection of the guard pages gets lost.
1800 //
1801 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1802 // may have been queued at the same time.
1803
1804 if (!_stack_is_executable) {
1805 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1806 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1807 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1808 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1809 warning("Attempt to reguard stack yellow zone failed.");
1810 }
1811 }
1812 }
1813 }
1814
1815 return result;
1816 }
1817
1818 void* os::dll_lookup(void* handle, const char* name) {
1819 void* res = dlsym(handle, name);
1820 return res;
1821 }
1822
1823 void* os::get_default_process_handle() {
1824 return (void*)::dlopen(NULL, RTLD_LAZY);
1825 }
1826
1827 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1828 int fd = ::open(filename, O_RDONLY);
1829 if (fd == -1) {
1830 return false;
1831 }
1832
1833 if (hdr != NULL) {
1834 st->print_cr("%s", hdr);
1835 }
1836
1837 char buf[33];
1838 int bytes;
1839 buf[32] = '\0';
1840 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1841 st->print_raw(buf, bytes);
1842 }
1843
1844 ::close(fd);
1845
1846 return true;
1847 }
1848
1849 void os::print_dll_info(outputStream *st) {
1850 st->print_cr("Dynamic libraries:");
1851
1852 char fname[32];
1853 pid_t pid = os::Linux::gettid();
1854
1855 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1856
1857 if (!_print_ascii_file(fname, st)) {
1858 st->print("Can not get library information for pid = %d\n", pid);
1859 }
1860 }
1861
1862 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1863 FILE *procmapsFile = NULL;
1864
1865 // Open the procfs maps file for the current process
1866 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1867 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1868 char line[PATH_MAX + 100];
1869
1870 // Read line by line from 'file'
1871 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1872 u8 base, top, offset, inode;
1873 char permissions[5];
1874 char device[6];
1875 char name[PATH_MAX + 1];
1876
1877 // Parse fields from line
1878 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1879 &base, &top, permissions, &offset, device, &inode, name);
1880
1881 // Filter by device id '00:00' so that we only get file system mapped files.
1882 if (strcmp(device, "00:00") != 0) {
1883
1884 // Call callback with the fields of interest
1885 if(callback(name, (address)base, (address)top, param)) {
1886 // Oops abort, callback aborted
1887 fclose(procmapsFile);
1888 return 1;
1889 }
1890 }
1891 }
1892 fclose(procmapsFile);
1893 }
1894 return 0;
1895 }
1896
1897 void os::print_os_info_brief(outputStream* st) {
1898 os::Linux::print_distro_info(st);
1899
1900 os::Posix::print_uname_info(st);
1901
1902 os::Linux::print_libversion_info(st);
1903
1904 }
1905
1906 void os::print_os_info(outputStream* st) {
1907 st->print("OS:");
1908
1909 os::Linux::print_distro_info(st);
1910
1911 os::Posix::print_uname_info(st);
1912
1913 // Print warning if unsafe chroot environment detected
1914 if (unsafe_chroot_detected) {
1915 st->print("WARNING!! ");
1916 st->print_cr("%s", unstable_chroot_error);
1917 }
1918
1919 os::Linux::print_libversion_info(st);
1920
1921 os::Posix::print_rlimit_info(st);
1922
1923 os::Posix::print_load_average(st);
1924
1925 os::Linux::print_full_memory_info(st);
1926
1927 os::Linux::print_proc_sys_info(st);
1928
1929 os::Linux::print_ld_preload_file(st);
1930
1931 os::Linux::print_container_info(st);
1932 }
1933
1934 // Try to identify popular distros.
1935 // Most Linux distributions have a /etc/XXX-release file, which contains
1936 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1937 // file that also contains the OS version string. Some have more than one
1938 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1939 // /etc/redhat-release.), so the order is important.
1940 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1941 // their own specific XXX-release file as well as a redhat-release file.
1942 // Because of this the XXX-release file needs to be searched for before the
1943 // redhat-release file.
1944 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1945 // search for redhat-release / SuSE-release needs to be before lsb-release.
1946 // Since the lsb-release file is the new standard it needs to be searched
1947 // before the older style release files.
1948 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1949 // next to last resort. The os-release file is a new standard that contains
1950 // distribution information and the system-release file seems to be an old
1951 // standard that has been replaced by the lsb-release and os-release files.
1952 // Searching for the debian_version file is the last resort. It contains
1953 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1954 // "Debian " is printed before the contents of the debian_version file.
1955
1956 const char* distro_files[] = {
1957 "/etc/oracle-release",
1958 "/etc/mandriva-release",
1959 "/etc/mandrake-release",
1960 "/etc/sun-release",
1961 "/etc/redhat-release",
1962 "/etc/SuSE-release",
1963 "/etc/lsb-release",
1964 "/etc/turbolinux-release",
1965 "/etc/gentoo-release",
1966 "/etc/ltib-release",
1967 "/etc/angstrom-version",
1968 "/etc/system-release",
1969 "/etc/os-release",
1970 NULL };
1971
1972 void os::Linux::print_distro_info(outputStream* st) {
1973 for (int i = 0;; i++) {
1974 const char* file = distro_files[i];
1975 if (file == NULL) {
1976 break; // done
1977 }
1978 // If file prints, we found it.
1979 if (_print_ascii_file(file, st)) {
1980 return;
1981 }
1982 }
1983
1984 if (file_exists("/etc/debian_version")) {
1985 st->print("Debian ");
1986 _print_ascii_file("/etc/debian_version", st);
1987 } else {
1988 st->print("Linux");
1989 }
1990 st->cr();
1991 }
1992
1993 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
1994 char buf[256];
1995 while (fgets(buf, sizeof(buf), fp)) {
1996 // Edit out extra stuff in expected format
1997 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
1998 char* ptr = strstr(buf, "\""); // the name is in quotes
1999 if (ptr != NULL) {
2000 ptr++; // go beyond first quote
2001 char* nl = strchr(ptr, '\"');
2002 if (nl != NULL) *nl = '\0';
2003 strncpy(distro, ptr, length);
2004 } else {
2005 ptr = strstr(buf, "=");
2006 ptr++; // go beyond equals then
2007 char* nl = strchr(ptr, '\n');
2008 if (nl != NULL) *nl = '\0';
2009 strncpy(distro, ptr, length);
2010 }
2011 return;
2012 } else if (get_first_line) {
2013 char* nl = strchr(buf, '\n');
2014 if (nl != NULL) *nl = '\0';
2015 strncpy(distro, buf, length);
2016 return;
2017 }
2018 }
2019 // print last line and close
2020 char* nl = strchr(buf, '\n');
2021 if (nl != NULL) *nl = '\0';
2022 strncpy(distro, buf, length);
2023 }
2024
2025 static void parse_os_info(char* distro, size_t length, const char* file) {
2026 FILE* fp = fopen(file, "r");
2027 if (fp != NULL) {
2028 // if suse format, print out first line
2029 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2030 parse_os_info_helper(fp, distro, length, get_first_line);
2031 fclose(fp);
2032 }
2033 }
2034
2035 void os::get_summary_os_info(char* buf, size_t buflen) {
2036 for (int i = 0;; i++) {
2037 const char* file = distro_files[i];
2038 if (file == NULL) {
2039 break; // ran out of distro_files
2040 }
2041 if (file_exists(file)) {
2042 parse_os_info(buf, buflen, file);
2043 return;
2044 }
2045 }
2046 // special case for debian
2047 if (file_exists("/etc/debian_version")) {
2048 strncpy(buf, "Debian ", buflen);
2049 if (buflen > 7) {
2050 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2051 }
2052 } else {
2053 strncpy(buf, "Linux", buflen);
2054 }
2055 }
2056
2057 void os::Linux::print_libversion_info(outputStream* st) {
2058 // libc, pthread
2059 st->print("libc:");
2060 st->print("%s ", os::Linux::glibc_version());
2061 st->print("%s ", os::Linux::libpthread_version());
2062 st->cr();
2063 }
2064
2065 void os::Linux::print_proc_sys_info(outputStream* st) {
2066 st->cr();
2067 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2068 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2069 st->cr();
2070 st->cr();
2071
2072 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2073 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2074 st->cr();
2075 st->cr();
2076
2077 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2078 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2079 st->cr();
2080 st->cr();
2081 }
2082
2083 void os::Linux::print_full_memory_info(outputStream* st) {
2084 st->print("\n/proc/meminfo:\n");
2085 _print_ascii_file("/proc/meminfo", st);
2086 st->cr();
2087 }
2088
2089 void os::Linux::print_ld_preload_file(outputStream* st) {
2090 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2091 st->cr();
2092 }
2093
2094 void os::Linux::print_container_info(outputStream* st) {
2095 if (!OSContainer::is_containerized()) {
2096 return;
2097 }
2098
2099 st->print("container (cgroup) information:\n");
2100
2101 const char *p_ct = OSContainer::container_type();
2102 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2103
2104 char *p = OSContainer::cpu_cpuset_cpus();
2105 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2106 free(p);
2107
2108 p = OSContainer::cpu_cpuset_memory_nodes();
2109 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2110 free(p);
2111
2112 int i = OSContainer::active_processor_count();
2113 if (i > 0) {
2114 st->print("active_processor_count: %d\n", i);
2115 } else {
2116 st->print("active_processor_count: failed\n");
2117 }
2118
2119 i = OSContainer::cpu_quota();
2120 st->print("cpu_quota: %d\n", i);
2121
2122 i = OSContainer::cpu_period();
2123 st->print("cpu_period: %d\n", i);
2124
2125 i = OSContainer::cpu_shares();
2126 st->print("cpu_shares: %d\n", i);
2127
2128 jlong j = OSContainer::memory_limit_in_bytes();
2129 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2130
2131 j = OSContainer::memory_and_swap_limit_in_bytes();
2132 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2133
2134 j = OSContainer::memory_soft_limit_in_bytes();
2135 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2136
2137 j = OSContainer::OSContainer::memory_usage_in_bytes();
2138 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2139
2140 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2141 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2142 st->cr();
2143 }
2144
2145 void os::print_memory_info(outputStream* st) {
2146
2147 st->print("Memory:");
2148 st->print(" %dk page", os::vm_page_size()>>10);
2149
2150 // values in struct sysinfo are "unsigned long"
2151 struct sysinfo si;
2152 sysinfo(&si);
2153
2154 st->print(", physical " UINT64_FORMAT "k",
2155 os::physical_memory() >> 10);
2156 st->print("(" UINT64_FORMAT "k free)",
2157 os::available_memory() >> 10);
2158 st->print(", swap " UINT64_FORMAT "k",
2159 ((jlong)si.totalswap * si.mem_unit) >> 10);
2160 st->print("(" UINT64_FORMAT "k free)",
2161 ((jlong)si.freeswap * si.mem_unit) >> 10);
2162 st->cr();
2163 }
2164
2165 // Print the first "model name" line and the first "flags" line
2166 // that we find and nothing more. We assume "model name" comes
2167 // before "flags" so if we find a second "model name", then the
2168 // "flags" field is considered missing.
2169 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2170 #if defined(IA32) || defined(AMD64)
2171 // Other platforms have less repetitive cpuinfo files
2172 FILE *fp = fopen("/proc/cpuinfo", "r");
2173 if (fp) {
2174 while (!feof(fp)) {
2175 if (fgets(buf, buflen, fp)) {
2176 // Assume model name comes before flags
2177 bool model_name_printed = false;
2178 if (strstr(buf, "model name") != NULL) {
2179 if (!model_name_printed) {
2180 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2181 st->print_raw(buf);
2182 model_name_printed = true;
2183 } else {
2184 // model name printed but not flags? Odd, just return
2185 fclose(fp);
2186 return true;
2187 }
2188 }
2189 // print the flags line too
2190 if (strstr(buf, "flags") != NULL) {
2191 st->print_raw(buf);
2192 fclose(fp);
2193 return true;
2194 }
2195 }
2196 }
2197 fclose(fp);
2198 }
2199 #endif // x86 platforms
2200 return false;
2201 }
2202
2203 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2204 // Only print the model name if the platform provides this as a summary
2205 if (!print_model_name_and_flags(st, buf, buflen)) {
2206 st->print("\n/proc/cpuinfo:\n");
2207 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2208 st->print_cr(" <Not Available>");
2209 }
2210 }
2211 }
2212
2213 #if defined(AMD64) || defined(IA32) || defined(X32)
2214 const char* search_string = "model name";
2215 #elif defined(M68K)
2216 const char* search_string = "CPU";
2217 #elif defined(PPC64)
2218 const char* search_string = "cpu";
2219 #elif defined(S390)
2220 const char* search_string = "processor";
2221 #elif defined(SPARC)
2222 const char* search_string = "cpu";
2223 #else
2224 const char* search_string = "Processor";
2225 #endif
2226
2227 // Parses the cpuinfo file for string representing the model name.
2228 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2229 FILE* fp = fopen("/proc/cpuinfo", "r");
2230 if (fp != NULL) {
2231 while (!feof(fp)) {
2232 char buf[256];
2233 if (fgets(buf, sizeof(buf), fp)) {
2234 char* start = strstr(buf, search_string);
2235 if (start != NULL) {
2236 char *ptr = start + strlen(search_string);
2237 char *end = buf + strlen(buf);
2238 while (ptr != end) {
2239 // skip whitespace and colon for the rest of the name.
2240 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2241 break;
2242 }
2243 ptr++;
2244 }
2245 if (ptr != end) {
2246 // reasonable string, get rid of newline and keep the rest
2247 char* nl = strchr(buf, '\n');
2248 if (nl != NULL) *nl = '\0';
2249 strncpy(cpuinfo, ptr, length);
2250 fclose(fp);
2251 return;
2252 }
2253 }
2254 }
2255 }
2256 fclose(fp);
2257 }
2258 // cpuinfo not found or parsing failed, just print generic string. The entire
2259 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2260 #if defined(AARCH64)
2261 strncpy(cpuinfo, "AArch64", length);
2262 #elif defined(AMD64)
2263 strncpy(cpuinfo, "x86_64", length);
2264 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2265 strncpy(cpuinfo, "ARM", length);
2266 #elif defined(IA32)
2267 strncpy(cpuinfo, "x86_32", length);
2268 #elif defined(IA64)
2269 strncpy(cpuinfo, "IA64", length);
2270 #elif defined(PPC)
2271 strncpy(cpuinfo, "PPC64", length);
2272 #elif defined(S390)
2273 strncpy(cpuinfo, "S390", length);
2274 #elif defined(SPARC)
2275 strncpy(cpuinfo, "sparcv9", length);
2276 #elif defined(ZERO_LIBARCH)
2277 strncpy(cpuinfo, ZERO_LIBARCH, length);
2278 #else
2279 strncpy(cpuinfo, "unknown", length);
2280 #endif
2281 }
2282
2283 static void print_signal_handler(outputStream* st, int sig,
2284 char* buf, size_t buflen);
2285
2286 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2287 st->print_cr("Signal Handlers:");
2288 print_signal_handler(st, SIGSEGV, buf, buflen);
2289 print_signal_handler(st, SIGBUS , buf, buflen);
2290 print_signal_handler(st, SIGFPE , buf, buflen);
2291 print_signal_handler(st, SIGPIPE, buf, buflen);
2292 print_signal_handler(st, SIGXFSZ, buf, buflen);
2293 print_signal_handler(st, SIGILL , buf, buflen);
2294 print_signal_handler(st, SR_signum, buf, buflen);
2295 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2296 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2297 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2298 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2299 #if defined(PPC64)
2300 print_signal_handler(st, SIGTRAP, buf, buflen);
2301 #endif
2302 }
2303
2304 static char saved_jvm_path[MAXPATHLEN] = {0};
2305
2306 // Find the full path to the current module, libjvm.so
2307 void os::jvm_path(char *buf, jint buflen) {
2308 // Error checking.
2309 if (buflen < MAXPATHLEN) {
2310 assert(false, "must use a large-enough buffer");
2311 buf[0] = '\0';
2312 return;
2313 }
2314 // Lazy resolve the path to current module.
2315 if (saved_jvm_path[0] != 0) {
2316 strcpy(buf, saved_jvm_path);
2317 return;
2318 }
2319
2320 char dli_fname[MAXPATHLEN];
2321 bool ret = dll_address_to_library_name(
2322 CAST_FROM_FN_PTR(address, os::jvm_path),
2323 dli_fname, sizeof(dli_fname), NULL);
2324 assert(ret, "cannot locate libjvm");
2325 char *rp = NULL;
2326 if (ret && dli_fname[0] != '\0') {
2327 rp = os::Posix::realpath(dli_fname, buf, buflen);
2328 }
2329 if (rp == NULL) {
2330 return;
2331 }
2332
2333 if (Arguments::sun_java_launcher_is_altjvm()) {
2334 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2335 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2336 // If "/jre/lib/" appears at the right place in the string, then
2337 // assume we are installed in a JDK and we're done. Otherwise, check
2338 // for a JAVA_HOME environment variable and fix up the path so it
2339 // looks like libjvm.so is installed there (append a fake suffix
2340 // hotspot/libjvm.so).
2341 const char *p = buf + strlen(buf) - 1;
2342 for (int count = 0; p > buf && count < 5; ++count) {
2343 for (--p; p > buf && *p != '/'; --p)
2344 /* empty */ ;
2345 }
2346
2347 if (strncmp(p, "/jre/lib/", 9) != 0) {
2348 // Look for JAVA_HOME in the environment.
2349 char* java_home_var = ::getenv("JAVA_HOME");
2350 if (java_home_var != NULL && java_home_var[0] != 0) {
2351 char* jrelib_p;
2352 int len;
2353
2354 // Check the current module name "libjvm.so".
2355 p = strrchr(buf, '/');
2356 if (p == NULL) {
2357 return;
2358 }
2359 assert(strstr(p, "/libjvm") == p, "invalid library name");
2360
2361 rp = os::Posix::realpath(java_home_var, buf, buflen);
2362 if (rp == NULL) {
2363 return;
2364 }
2365
2366 // determine if this is a legacy image or modules image
2367 // modules image doesn't have "jre" subdirectory
2368 len = strlen(buf);
2369 assert(len < buflen, "Ran out of buffer room");
2370 jrelib_p = buf + len;
2371 snprintf(jrelib_p, buflen-len, "/jre/lib");
2372 if (0 != access(buf, F_OK)) {
2373 snprintf(jrelib_p, buflen-len, "/lib");
2374 }
2375
2376 if (0 == access(buf, F_OK)) {
2377 // Use current module name "libjvm.so"
2378 len = strlen(buf);
2379 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2380 } else {
2381 // Go back to path of .so
2382 rp = os::Posix::realpath(dli_fname, buf, buflen);
2383 if (rp == NULL) {
2384 return;
2385 }
2386 }
2387 }
2388 }
2389 }
2390
2391 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2392 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2393 }
2394
2395 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2396 // no prefix required, not even "_"
2397 }
2398
2399 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2400 // no suffix required
2401 }
2402
2403 ////////////////////////////////////////////////////////////////////////////////
2404 // sun.misc.Signal support
2405
2406 static volatile jint sigint_count = 0;
2407
2408 static void UserHandler(int sig, void *siginfo, void *context) {
2409 // 4511530 - sem_post is serialized and handled by the manager thread. When
2410 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2411 // don't want to flood the manager thread with sem_post requests.
2412 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2413 return;
2414 }
2415
2416 // Ctrl-C is pressed during error reporting, likely because the error
2417 // handler fails to abort. Let VM die immediately.
2418 if (sig == SIGINT && VMError::is_error_reported()) {
2419 os::die();
2420 }
2421
2422 os::signal_notify(sig);
2423 }
2424
2425 void* os::user_handler() {
2426 return CAST_FROM_FN_PTR(void*, UserHandler);
2427 }
2428
2429 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2430 struct timespec ts;
2431 // Semaphore's are always associated with CLOCK_REALTIME
2432 os::Posix::clock_gettime(CLOCK_REALTIME, &ts);
2433 // see os_posix.cpp for discussion on overflow checking
2434 if (sec >= MAX_SECS) {
2435 ts.tv_sec += MAX_SECS;
2436 ts.tv_nsec = 0;
2437 } else {
2438 ts.tv_sec += sec;
2439 ts.tv_nsec += nsec;
2440 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2441 ts.tv_nsec -= NANOSECS_PER_SEC;
2442 ++ts.tv_sec; // note: this must be <= max_secs
2443 }
2444 }
2445
2446 return ts;
2447 }
2448
2449 extern "C" {
2450 typedef void (*sa_handler_t)(int);
2451 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2452 }
2453
2454 void* os::signal(int signal_number, void* handler) {
2455 struct sigaction sigAct, oldSigAct;
2456
2457 sigfillset(&(sigAct.sa_mask));
2458 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2459 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2460
2461 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2462 // -1 means registration failed
2463 return (void *)-1;
2464 }
2465
2466 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2467 }
2468
2469 void os::signal_raise(int signal_number) {
2470 ::raise(signal_number);
2471 }
2472
2473 // The following code is moved from os.cpp for making this
2474 // code platform specific, which it is by its very nature.
2475
2476 // Will be modified when max signal is changed to be dynamic
2477 int os::sigexitnum_pd() {
2478 return NSIG;
2479 }
2480
2481 // a counter for each possible signal value
2482 static volatile jint pending_signals[NSIG+1] = { 0 };
2483
2484 // Linux(POSIX) specific hand shaking semaphore.
2485 static Semaphore* sig_sem = NULL;
2486 static PosixSemaphore sr_semaphore;
2487
2488 static void jdk_misc_signal_init() {
2489 // Initialize signal structures
2490 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2491
2492 // Initialize signal semaphore
2493 sig_sem = new Semaphore();
2494 }
2495
2496 void os::signal_notify(int sig) {
2497 if (sig_sem != NULL) {
2498 Atomic::inc(&pending_signals[sig]);
2499 sig_sem->signal();
2500 } else {
2501 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2502 // initialization isn't called.
2503 assert(ReduceSignalUsage, "signal semaphore should be created");
2504 }
2505 }
2506
2507 static int check_pending_signals() {
2508 Atomic::store(0, &sigint_count);
2509 for (;;) {
2510 for (int i = 0; i < NSIG + 1; i++) {
2511 jint n = pending_signals[i];
2512 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2513 return i;
2514 }
2515 }
2516 JavaThread *thread = JavaThread::current();
2517 ThreadBlockInVM tbivm(thread);
2518
2519 bool threadIsSuspended;
2520 do {
2521 thread->set_suspend_equivalent();
2522 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2523 sig_sem->wait();
2524
2525 // were we externally suspended while we were waiting?
2526 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2527 if (threadIsSuspended) {
2528 // The semaphore has been incremented, but while we were waiting
2529 // another thread suspended us. We don't want to continue running
2530 // while suspended because that would surprise the thread that
2531 // suspended us.
2532 sig_sem->signal();
2533
2534 thread->java_suspend_self();
2535 }
2536 } while (threadIsSuspended);
2537 }
2538 }
2539
2540 int os::signal_wait() {
2541 return check_pending_signals();
2542 }
2543
2544 ////////////////////////////////////////////////////////////////////////////////
2545 // Virtual Memory
2546
2547 int os::vm_page_size() {
2548 // Seems redundant as all get out
2549 assert(os::Linux::page_size() != -1, "must call os::init");
2550 return os::Linux::page_size();
2551 }
2552
2553 // Solaris allocates memory by pages.
2554 int os::vm_allocation_granularity() {
2555 assert(os::Linux::page_size() != -1, "must call os::init");
2556 return os::Linux::page_size();
2557 }
2558
2559 // Rationale behind this function:
2560 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2561 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2562 // samples for JITted code. Here we create private executable mapping over the code cache
2563 // and then we can use standard (well, almost, as mapping can change) way to provide
2564 // info for the reporting script by storing timestamp and location of symbol
2565 void linux_wrap_code(char* base, size_t size) {
2566 static volatile jint cnt = 0;
2567
2568 if (!UseOprofile) {
2569 return;
2570 }
2571
2572 char buf[PATH_MAX+1];
2573 int num = Atomic::add(1, &cnt);
2574
2575 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2576 os::get_temp_directory(), os::current_process_id(), num);
2577 unlink(buf);
2578
2579 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2580
2581 if (fd != -1) {
2582 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2583 if (rv != (off_t)-1) {
2584 if (::write(fd, "", 1) == 1) {
2585 mmap(base, size,
2586 PROT_READ|PROT_WRITE|PROT_EXEC,
2587 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2588 }
2589 }
2590 ::close(fd);
2591 unlink(buf);
2592 }
2593 }
2594
2595 static bool recoverable_mmap_error(int err) {
2596 // See if the error is one we can let the caller handle. This
2597 // list of errno values comes from JBS-6843484. I can't find a
2598 // Linux man page that documents this specific set of errno
2599 // values so while this list currently matches Solaris, it may
2600 // change as we gain experience with this failure mode.
2601 switch (err) {
2602 case EBADF:
2603 case EINVAL:
2604 case ENOTSUP:
2605 // let the caller deal with these errors
2606 return true;
2607
2608 default:
2609 // Any remaining errors on this OS can cause our reserved mapping
2610 // to be lost. That can cause confusion where different data
2611 // structures think they have the same memory mapped. The worst
2612 // scenario is if both the VM and a library think they have the
2613 // same memory mapped.
2614 return false;
2615 }
2616 }
2617
2618 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2619 int err) {
2620 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2621 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2622 os::strerror(err), err);
2623 }
2624
2625 static void warn_fail_commit_memory(char* addr, size_t size,
2626 size_t alignment_hint, bool exec,
2627 int err) {
2628 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2629 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2630 alignment_hint, exec, os::strerror(err), err);
2631 }
2632
2633 // NOTE: Linux kernel does not really reserve the pages for us.
2634 // All it does is to check if there are enough free pages
2635 // left at the time of mmap(). This could be a potential
2636 // problem.
2637 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2638 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2639 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2640 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2641 if (res != (uintptr_t) MAP_FAILED) {
2642 if (UseNUMAInterleaving) {
2643 numa_make_global(addr, size);
2644 }
2645 return 0;
2646 }
2647
2648 int err = errno; // save errno from mmap() call above
2649
2650 if (!recoverable_mmap_error(err)) {
2651 warn_fail_commit_memory(addr, size, exec, err);
2652 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2653 }
2654
2655 return err;
2656 }
2657
2658 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2659 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2660 }
2661
2662 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2663 const char* mesg) {
2664 assert(mesg != NULL, "mesg must be specified");
2665 int err = os::Linux::commit_memory_impl(addr, size, exec);
2666 if (err != 0) {
2667 // the caller wants all commit errors to exit with the specified mesg:
2668 warn_fail_commit_memory(addr, size, exec, err);
2669 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2670 }
2671 }
2672
2673 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2674 #ifndef MAP_HUGETLB
2675 #define MAP_HUGETLB 0x40000
2676 #endif
2677
2678 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2679 #ifndef MADV_HUGEPAGE
2680 #define MADV_HUGEPAGE 14
2681 #endif
2682
2683 int os::Linux::commit_memory_impl(char* addr, size_t size,
2684 size_t alignment_hint, bool exec) {
2685 int err = os::Linux::commit_memory_impl(addr, size, exec);
2686 if (err == 0) {
2687 realign_memory(addr, size, alignment_hint);
2688 }
2689 return err;
2690 }
2691
2692 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2693 bool exec) {
2694 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2695 }
2696
2697 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2698 size_t alignment_hint, bool exec,
2699 const char* mesg) {
2700 assert(mesg != NULL, "mesg must be specified");
2701 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2702 if (err != 0) {
2703 // the caller wants all commit errors to exit with the specified mesg:
2704 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2705 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2706 }
2707 }
2708
2709 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2710 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2711 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2712 // be supported or the memory may already be backed by huge pages.
2713 ::madvise(addr, bytes, MADV_HUGEPAGE);
2714 }
2715 }
2716
2717 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2718 // This method works by doing an mmap over an existing mmaping and effectively discarding
2719 // the existing pages. However it won't work for SHM-based large pages that cannot be
2720 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2721 // small pages on top of the SHM segment. This method always works for small pages, so we
2722 // allow that in any case.
2723 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2724 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2725 }
2726 }
2727
2728 void os::numa_make_global(char *addr, size_t bytes) {
2729 Linux::numa_interleave_memory(addr, bytes);
2730 }
2731
2732 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2733 // bind policy to MPOL_PREFERRED for the current thread.
2734 #define USE_MPOL_PREFERRED 0
2735
2736 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2737 // To make NUMA and large pages more robust when both enabled, we need to ease
2738 // the requirements on where the memory should be allocated. MPOL_BIND is the
2739 // default policy and it will force memory to be allocated on the specified
2740 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2741 // the specified node, but will not force it. Using this policy will prevent
2742 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2743 // free large pages.
2744 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2745 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2746 }
2747
2748 bool os::numa_topology_changed() { return false; }
2749
2750 size_t os::numa_get_groups_num() {
2751 // Return just the number of nodes in which it's possible to allocate memory
2752 // (in numa terminology, configured nodes).
2753 return Linux::numa_num_configured_nodes();
2754 }
2755
2756 int os::numa_get_group_id() {
2757 int cpu_id = Linux::sched_getcpu();
2758 if (cpu_id != -1) {
2759 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2760 if (lgrp_id != -1) {
2761 return lgrp_id;
2762 }
2763 }
2764 return 0;
2765 }
2766
2767 int os::Linux::get_existing_num_nodes() {
2768 int node;
2769 int highest_node_number = Linux::numa_max_node();
2770 int num_nodes = 0;
2771
2772 // Get the total number of nodes in the system including nodes without memory.
2773 for (node = 0; node <= highest_node_number; node++) {
2774 if (isnode_in_existing_nodes(node)) {
2775 num_nodes++;
2776 }
2777 }
2778 return num_nodes;
2779 }
2780
2781 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2782 int highest_node_number = Linux::numa_max_node();
2783 size_t i = 0;
2784
2785 // Map all node ids in which it is possible to allocate memory. Also nodes are
2786 // not always consecutively available, i.e. available from 0 to the highest
2787 // node number. If the nodes have been bound explicitly using numactl membind,
2788 // then allocate memory from those nodes only.
2789 for (int node = 0; node <= highest_node_number; node++) {
2790 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2791 ids[i++] = node;
2792 }
2793 }
2794 return i;
2795 }
2796
2797 bool os::get_page_info(char *start, page_info* info) {
2798 return false;
2799 }
2800
2801 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2802 page_info* page_found) {
2803 return end;
2804 }
2805
2806
2807 int os::Linux::sched_getcpu_syscall(void) {
2808 unsigned int cpu = 0;
2809 int retval = -1;
2810
2811 #if defined(IA32)
2812 #ifndef SYS_getcpu
2813 #define SYS_getcpu 318
2814 #endif
2815 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2816 #elif defined(AMD64)
2817 // Unfortunately we have to bring all these macros here from vsyscall.h
2818 // to be able to compile on old linuxes.
2819 #define __NR_vgetcpu 2
2820 #define VSYSCALL_START (-10UL << 20)
2821 #define VSYSCALL_SIZE 1024
2822 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2823 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2824 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2825 retval = vgetcpu(&cpu, NULL, NULL);
2826 #endif
2827
2828 return (retval == -1) ? retval : cpu;
2829 }
2830
2831 void os::Linux::sched_getcpu_init() {
2832 // sched_getcpu() should be in libc.
2833 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2834 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2835
2836 // If it's not, try a direct syscall.
2837 if (sched_getcpu() == -1) {
2838 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2839 (void*)&sched_getcpu_syscall));
2840 }
2841
2842 if (sched_getcpu() == -1) {
2843 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2844 }
2845 }
2846
2847 // Something to do with the numa-aware allocator needs these symbols
2848 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2849 extern "C" JNIEXPORT void numa_error(char *where) { }
2850
2851 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2852 // load symbol from base version instead.
2853 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2854 void *f = dlvsym(handle, name, "libnuma_1.1");
2855 if (f == NULL) {
2856 f = dlsym(handle, name);
2857 }
2858 return f;
2859 }
2860
2861 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2862 // Return NULL if the symbol is not defined in this particular version.
2863 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2864 return dlvsym(handle, name, "libnuma_1.2");
2865 }
2866
2867 bool os::Linux::libnuma_init() {
2868 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2869 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2870 if (handle != NULL) {
2871 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2872 libnuma_dlsym(handle, "numa_node_to_cpus")));
2873 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2874 libnuma_dlsym(handle, "numa_max_node")));
2875 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2876 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2877 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2878 libnuma_dlsym(handle, "numa_available")));
2879 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2880 libnuma_dlsym(handle, "numa_tonode_memory")));
2881 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2882 libnuma_dlsym(handle, "numa_interleave_memory")));
2883 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2884 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2885 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2886 libnuma_dlsym(handle, "numa_set_bind_policy")));
2887 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2888 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2889 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2890 libnuma_dlsym(handle, "numa_distance")));
2891 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2892 libnuma_v2_dlsym(handle, "numa_get_membind")));
2893
2894 if (numa_available() != -1) {
2895 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2896 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2897 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2898 // Create an index -> node mapping, since nodes are not always consecutive
2899 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2900 rebuild_nindex_to_node_map();
2901 // Create a cpu -> node mapping
2902 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2903 rebuild_cpu_to_node_map();
2904 return true;
2905 }
2906 }
2907 }
2908 return false;
2909 }
2910
2911 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2912 // Creating guard page is very expensive. Java thread has HotSpot
2913 // guard pages, only enable glibc guard page for non-Java threads.
2914 // (Remember: compiler thread is a Java thread, too!)
2915 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2916 }
2917
2918 void os::Linux::rebuild_nindex_to_node_map() {
2919 int highest_node_number = Linux::numa_max_node();
2920
2921 nindex_to_node()->clear();
2922 for (int node = 0; node <= highest_node_number; node++) {
2923 if (Linux::isnode_in_existing_nodes(node)) {
2924 nindex_to_node()->append(node);
2925 }
2926 }
2927 }
2928
2929 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2930 // The table is later used in get_node_by_cpu().
2931 void os::Linux::rebuild_cpu_to_node_map() {
2932 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2933 // in libnuma (possible values are starting from 16,
2934 // and continuing up with every other power of 2, but less
2935 // than the maximum number of CPUs supported by kernel), and
2936 // is a subject to change (in libnuma version 2 the requirements
2937 // are more reasonable) we'll just hardcode the number they use
2938 // in the library.
2939 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2940
2941 size_t cpu_num = processor_count();
2942 size_t cpu_map_size = NCPUS / BitsPerCLong;
2943 size_t cpu_map_valid_size =
2944 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2945
2946 cpu_to_node()->clear();
2947 cpu_to_node()->at_grow(cpu_num - 1);
2948
2949 size_t node_num = get_existing_num_nodes();
2950
2951 int distance = 0;
2952 int closest_distance = INT_MAX;
2953 int closest_node = 0;
2954 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2955 for (size_t i = 0; i < node_num; i++) {
2956 // Check if node is configured (not a memory-less node). If it is not, find
2957 // the closest configured node. Check also if node is bound, i.e. it's allowed
2958 // to allocate memory from the node. If it's not allowed, map cpus in that node
2959 // to the closest node from which memory allocation is allowed.
2960 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
2961 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
2962 closest_distance = INT_MAX;
2963 // Check distance from all remaining nodes in the system. Ignore distance
2964 // from itself, from another non-configured node, and from another non-bound
2965 // node.
2966 for (size_t m = 0; m < node_num; m++) {
2967 if (m != i &&
2968 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
2969 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
2970 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2971 // If a closest node is found, update. There is always at least one
2972 // configured and bound node in the system so there is always at least
2973 // one node close.
2974 if (distance != 0 && distance < closest_distance) {
2975 closest_distance = distance;
2976 closest_node = nindex_to_node()->at(m);
2977 }
2978 }
2979 }
2980 } else {
2981 // Current node is already a configured node.
2982 closest_node = nindex_to_node()->at(i);
2983 }
2984
2985 // Get cpus from the original node and map them to the closest node. If node
2986 // is a configured node (not a memory-less node), then original node and
2987 // closest node are the same.
2988 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2989 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2990 if (cpu_map[j] != 0) {
2991 for (size_t k = 0; k < BitsPerCLong; k++) {
2992 if (cpu_map[j] & (1UL << k)) {
2993 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2994 }
2995 }
2996 }
2997 }
2998 }
2999 }
3000 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3001 }
3002
3003 int os::Linux::get_node_by_cpu(int cpu_id) {
3004 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3005 return cpu_to_node()->at(cpu_id);
3006 }
3007 return -1;
3008 }
3009
3010 GrowableArray<int>* os::Linux::_cpu_to_node;
3011 GrowableArray<int>* os::Linux::_nindex_to_node;
3012 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3013 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3014 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3015 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3016 os::Linux::numa_available_func_t os::Linux::_numa_available;
3017 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3018 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3019 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3020 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3021 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3022 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3023 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3024 unsigned long* os::Linux::_numa_all_nodes;
3025 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3026 struct bitmask* os::Linux::_numa_nodes_ptr;
3027
3028 bool os::pd_uncommit_memory(char* addr, size_t size) {
3029 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3030 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3031 return res != (uintptr_t) MAP_FAILED;
3032 }
3033
3034 static address get_stack_commited_bottom(address bottom, size_t size) {
3035 address nbot = bottom;
3036 address ntop = bottom + size;
3037
3038 size_t page_sz = os::vm_page_size();
3039 unsigned pages = size / page_sz;
3040
3041 unsigned char vec[1];
3042 unsigned imin = 1, imax = pages + 1, imid;
3043 int mincore_return_value = 0;
3044
3045 assert(imin <= imax, "Unexpected page size");
3046
3047 while (imin < imax) {
3048 imid = (imax + imin) / 2;
3049 nbot = ntop - (imid * page_sz);
3050
3051 // Use a trick with mincore to check whether the page is mapped or not.
3052 // mincore sets vec to 1 if page resides in memory and to 0 if page
3053 // is swapped output but if page we are asking for is unmapped
3054 // it returns -1,ENOMEM
3055 mincore_return_value = mincore(nbot, page_sz, vec);
3056
3057 if (mincore_return_value == -1) {
3058 // Page is not mapped go up
3059 // to find first mapped page
3060 if (errno != EAGAIN) {
3061 assert(errno == ENOMEM, "Unexpected mincore errno");
3062 imax = imid;
3063 }
3064 } else {
3065 // Page is mapped go down
3066 // to find first not mapped page
3067 imin = imid + 1;
3068 }
3069 }
3070
3071 nbot = nbot + page_sz;
3072
3073 // Adjust stack bottom one page up if last checked page is not mapped
3074 if (mincore_return_value == -1) {
3075 nbot = nbot + page_sz;
3076 }
3077
3078 return nbot;
3079 }
3080
3081 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3082 int mincore_return_value;
3083 const size_t stripe = 1024; // query this many pages each time
3084 unsigned char vec[stripe + 1];
3085 // set a guard
3086 vec[stripe] = 'X';
3087
3088 const size_t page_sz = os::vm_page_size();
3089 size_t pages = size / page_sz;
3090
3091 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3092 assert(is_aligned(size, page_sz), "Size must be page aligned");
3093
3094 committed_start = NULL;
3095
3096 int loops = (pages + stripe - 1) / stripe;
3097 int committed_pages = 0;
3098 address loop_base = start;
3099 bool found_range = false;
3100
3101 for (int index = 0; index < loops && !found_range; index ++) {
3102 assert(pages > 0, "Nothing to do");
3103 int pages_to_query = (pages >= stripe) ? stripe : pages;
3104 pages -= pages_to_query;
3105
3106 // Get stable read
3107 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3108
3109 // During shutdown, some memory goes away without properly notifying NMT,
3110 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3111 // Bailout and return as not committed for now.
3112 if (mincore_return_value == -1 && errno == ENOMEM) {
3113 return false;
3114 }
3115
3116 assert(vec[stripe] == 'X', "overflow guard");
3117 assert(mincore_return_value == 0, "Range must be valid");
3118 // Process this stripe
3119 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3120 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3121 // End of current contiguous region
3122 if (committed_start != NULL) {
3123 found_range = true;
3124 break;
3125 }
3126 } else { // committed
3127 // Start of region
3128 if (committed_start == NULL) {
3129 committed_start = loop_base + page_sz * vecIdx;
3130 }
3131 committed_pages ++;
3132 }
3133 }
3134
3135 loop_base += pages_to_query * page_sz;
3136 }
3137
3138 if (committed_start != NULL) {
3139 assert(committed_pages > 0, "Must have committed region");
3140 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3141 assert(committed_start >= start && committed_start < start + size, "Out of range");
3142 committed_size = page_sz * committed_pages;
3143 return true;
3144 } else {
3145 assert(committed_pages == 0, "Should not have committed region");
3146 return false;
3147 }
3148 }
3149
3150
3151 // Linux uses a growable mapping for the stack, and if the mapping for
3152 // the stack guard pages is not removed when we detach a thread the
3153 // stack cannot grow beyond the pages where the stack guard was
3154 // mapped. If at some point later in the process the stack expands to
3155 // that point, the Linux kernel cannot expand the stack any further
3156 // because the guard pages are in the way, and a segfault occurs.
3157 //
3158 // However, it's essential not to split the stack region by unmapping
3159 // a region (leaving a hole) that's already part of the stack mapping,
3160 // so if the stack mapping has already grown beyond the guard pages at
3161 // the time we create them, we have to truncate the stack mapping.
3162 // So, we need to know the extent of the stack mapping when
3163 // create_stack_guard_pages() is called.
3164
3165 // We only need this for stacks that are growable: at the time of
3166 // writing thread stacks don't use growable mappings (i.e. those
3167 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3168 // only applies to the main thread.
3169
3170 // If the (growable) stack mapping already extends beyond the point
3171 // where we're going to put our guard pages, truncate the mapping at
3172 // that point by munmap()ping it. This ensures that when we later
3173 // munmap() the guard pages we don't leave a hole in the stack
3174 // mapping. This only affects the main/primordial thread
3175
3176 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3177 if (os::is_primordial_thread()) {
3178 // As we manually grow stack up to bottom inside create_attached_thread(),
3179 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3180 // we don't need to do anything special.
3181 // Check it first, before calling heavy function.
3182 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3183 unsigned char vec[1];
3184
3185 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3186 // Fallback to slow path on all errors, including EAGAIN
3187 stack_extent = (uintptr_t) get_stack_commited_bottom(
3188 os::Linux::initial_thread_stack_bottom(),
3189 (size_t)addr - stack_extent);
3190 }
3191
3192 if (stack_extent < (uintptr_t)addr) {
3193 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3194 }
3195 }
3196
3197 return os::commit_memory(addr, size, !ExecMem);
3198 }
3199
3200 // If this is a growable mapping, remove the guard pages entirely by
3201 // munmap()ping them. If not, just call uncommit_memory(). This only
3202 // affects the main/primordial thread, but guard against future OS changes.
3203 // It's safe to always unmap guard pages for primordial thread because we
3204 // always place it right after end of the mapped region.
3205
3206 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3207 uintptr_t stack_extent, stack_base;
3208
3209 if (os::is_primordial_thread()) {
3210 return ::munmap(addr, size) == 0;
3211 }
3212
3213 return os::uncommit_memory(addr, size);
3214 }
3215
3216 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3217 // at 'requested_addr'. If there are existing memory mappings at the same
3218 // location, however, they will be overwritten. If 'fixed' is false,
3219 // 'requested_addr' is only treated as a hint, the return value may or
3220 // may not start from the requested address. Unlike Linux mmap(), this
3221 // function returns NULL to indicate failure.
3222 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3223 char * addr;
3224 int flags;
3225
3226 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3227 if (fixed) {
3228 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3229 flags |= MAP_FIXED;
3230 }
3231
3232 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3233 // touch an uncommitted page. Otherwise, the read/write might
3234 // succeed if we have enough swap space to back the physical page.
3235 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3236 flags, -1, 0);
3237
3238 return addr == MAP_FAILED ? NULL : addr;
3239 }
3240
3241 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3242 // (req_addr != NULL) or with a given alignment.
3243 // - bytes shall be a multiple of alignment.
3244 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3245 // - alignment sets the alignment at which memory shall be allocated.
3246 // It must be a multiple of allocation granularity.
3247 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3248 // req_addr or NULL.
3249 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3250
3251 size_t extra_size = bytes;
3252 if (req_addr == NULL && alignment > 0) {
3253 extra_size += alignment;
3254 }
3255
3256 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3257 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3258 -1, 0);
3259 if (start == MAP_FAILED) {
3260 start = NULL;
3261 } else {
3262 if (req_addr != NULL) {
3263 if (start != req_addr) {
3264 ::munmap(start, extra_size);
3265 start = NULL;
3266 }
3267 } else {
3268 char* const start_aligned = align_up(start, alignment);
3269 char* const end_aligned = start_aligned + bytes;
3270 char* const end = start + extra_size;
3271 if (start_aligned > start) {
3272 ::munmap(start, start_aligned - start);
3273 }
3274 if (end_aligned < end) {
3275 ::munmap(end_aligned, end - end_aligned);
3276 }
3277 start = start_aligned;
3278 }
3279 }
3280 return start;
3281 }
3282
3283 static int anon_munmap(char * addr, size_t size) {
3284 return ::munmap(addr, size) == 0;
3285 }
3286
3287 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3288 size_t alignment_hint) {
3289 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3290 }
3291
3292 bool os::pd_release_memory(char* addr, size_t size) {
3293 return anon_munmap(addr, size);
3294 }
3295
3296 static bool linux_mprotect(char* addr, size_t size, int prot) {
3297 // Linux wants the mprotect address argument to be page aligned.
3298 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3299
3300 // According to SUSv3, mprotect() should only be used with mappings
3301 // established by mmap(), and mmap() always maps whole pages. Unaligned
3302 // 'addr' likely indicates problem in the VM (e.g. trying to change
3303 // protection of malloc'ed or statically allocated memory). Check the
3304 // caller if you hit this assert.
3305 assert(addr == bottom, "sanity check");
3306
3307 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3308 return ::mprotect(bottom, size, prot) == 0;
3309 }
3310
3311 // Set protections specified
3312 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3313 bool is_committed) {
3314 unsigned int p = 0;
3315 switch (prot) {
3316 case MEM_PROT_NONE: p = PROT_NONE; break;
3317 case MEM_PROT_READ: p = PROT_READ; break;
3318 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3319 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3320 default:
3321 ShouldNotReachHere();
3322 }
3323 // is_committed is unused.
3324 return linux_mprotect(addr, bytes, p);
3325 }
3326
3327 bool os::guard_memory(char* addr, size_t size) {
3328 return linux_mprotect(addr, size, PROT_NONE);
3329 }
3330
3331 bool os::unguard_memory(char* addr, size_t size) {
3332 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3333 }
3334
3335 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3336 size_t page_size) {
3337 bool result = false;
3338 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3339 MAP_ANONYMOUS|MAP_PRIVATE,
3340 -1, 0);
3341 if (p != MAP_FAILED) {
3342 void *aligned_p = align_up(p, page_size);
3343
3344 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3345
3346 munmap(p, page_size * 2);
3347 }
3348
3349 if (warn && !result) {
3350 warning("TransparentHugePages is not supported by the operating system.");
3351 }
3352
3353 return result;
3354 }
3355
3356 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3357 bool result = false;
3358 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3359 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3360 -1, 0);
3361
3362 if (p != MAP_FAILED) {
3363 // We don't know if this really is a huge page or not.
3364 FILE *fp = fopen("/proc/self/maps", "r");
3365 if (fp) {
3366 while (!feof(fp)) {
3367 char chars[257];
3368 long x = 0;
3369 if (fgets(chars, sizeof(chars), fp)) {
3370 if (sscanf(chars, "%lx-%*x", &x) == 1
3371 && x == (long)p) {
3372 if (strstr (chars, "hugepage")) {
3373 result = true;
3374 break;
3375 }
3376 }
3377 }
3378 }
3379 fclose(fp);
3380 }
3381 munmap(p, page_size);
3382 }
3383
3384 if (warn && !result) {
3385 warning("HugeTLBFS is not supported by the operating system.");
3386 }
3387
3388 return result;
3389 }
3390
3391 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3392 //
3393 // From the coredump_filter documentation:
3394 //
3395 // - (bit 0) anonymous private memory
3396 // - (bit 1) anonymous shared memory
3397 // - (bit 2) file-backed private memory
3398 // - (bit 3) file-backed shared memory
3399 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3400 // effective only if the bit 2 is cleared)
3401 // - (bit 5) hugetlb private memory
3402 // - (bit 6) hugetlb shared memory
3403 // - (bit 7) dax private memory
3404 // - (bit 8) dax shared memory
3405 //
3406 static void set_coredump_filter(CoredumpFilterBit bit) {
3407 FILE *f;
3408 long cdm;
3409
3410 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3411 return;
3412 }
3413
3414 if (fscanf(f, "%lx", &cdm) != 1) {
3415 fclose(f);
3416 return;
3417 }
3418
3419 long saved_cdm = cdm;
3420 rewind(f);
3421 cdm |= bit;
3422
3423 if (cdm != saved_cdm) {
3424 fprintf(f, "%#lx", cdm);
3425 }
3426
3427 fclose(f);
3428 }
3429
3430 // Large page support
3431
3432 static size_t _large_page_size = 0;
3433
3434 size_t os::Linux::find_large_page_size() {
3435 size_t large_page_size = 0;
3436
3437 // large_page_size on Linux is used to round up heap size. x86 uses either
3438 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3439 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3440 // page as large as 256M.
3441 //
3442 // Here we try to figure out page size by parsing /proc/meminfo and looking
3443 // for a line with the following format:
3444 // Hugepagesize: 2048 kB
3445 //
3446 // If we can't determine the value (e.g. /proc is not mounted, or the text
3447 // format has been changed), we'll use the largest page size supported by
3448 // the processor.
3449
3450 #ifndef ZERO
3451 large_page_size =
3452 AARCH64_ONLY(2 * M)
3453 AMD64_ONLY(2 * M)
3454 ARM32_ONLY(2 * M)
3455 IA32_ONLY(4 * M)
3456 IA64_ONLY(256 * M)
3457 PPC_ONLY(4 * M)
3458 S390_ONLY(1 * M)
3459 SPARC_ONLY(4 * M);
3460 #endif // ZERO
3461
3462 FILE *fp = fopen("/proc/meminfo", "r");
3463 if (fp) {
3464 while (!feof(fp)) {
3465 int x = 0;
3466 char buf[16];
3467 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3468 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3469 large_page_size = x * K;
3470 break;
3471 }
3472 } else {
3473 // skip to next line
3474 for (;;) {
3475 int ch = fgetc(fp);
3476 if (ch == EOF || ch == (int)'\n') break;
3477 }
3478 }
3479 }
3480 fclose(fp);
3481 }
3482
3483 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3484 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3485 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3486 proper_unit_for_byte_size(large_page_size));
3487 }
3488
3489 return large_page_size;
3490 }
3491
3492 size_t os::Linux::setup_large_page_size() {
3493 _large_page_size = Linux::find_large_page_size();
3494 const size_t default_page_size = (size_t)Linux::page_size();
3495 if (_large_page_size > default_page_size) {
3496 _page_sizes[0] = _large_page_size;
3497 _page_sizes[1] = default_page_size;
3498 _page_sizes[2] = 0;
3499 }
3500
3501 return _large_page_size;
3502 }
3503
3504 bool os::Linux::setup_large_page_type(size_t page_size) {
3505 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3506 FLAG_IS_DEFAULT(UseSHM) &&
3507 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3508
3509 // The type of large pages has not been specified by the user.
3510
3511 // Try UseHugeTLBFS and then UseSHM.
3512 UseHugeTLBFS = UseSHM = true;
3513
3514 // Don't try UseTransparentHugePages since there are known
3515 // performance issues with it turned on. This might change in the future.
3516 UseTransparentHugePages = false;
3517 }
3518
3519 if (UseTransparentHugePages) {
3520 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3521 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3522 UseHugeTLBFS = false;
3523 UseSHM = false;
3524 return true;
3525 }
3526 UseTransparentHugePages = false;
3527 }
3528
3529 if (UseHugeTLBFS) {
3530 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3531 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3532 UseSHM = false;
3533 return true;
3534 }
3535 UseHugeTLBFS = false;
3536 }
3537
3538 return UseSHM;
3539 }
3540
3541 void os::large_page_init() {
3542 if (!UseLargePages &&
3543 !UseTransparentHugePages &&
3544 !UseHugeTLBFS &&
3545 !UseSHM) {
3546 // Not using large pages.
3547 return;
3548 }
3549
3550 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3551 // The user explicitly turned off large pages.
3552 // Ignore the rest of the large pages flags.
3553 UseTransparentHugePages = false;
3554 UseHugeTLBFS = false;
3555 UseSHM = false;
3556 return;
3557 }
3558
3559 size_t large_page_size = Linux::setup_large_page_size();
3560 UseLargePages = Linux::setup_large_page_type(large_page_size);
3561
3562 set_coredump_filter(LARGEPAGES_BIT);
3563 }
3564
3565 #ifndef SHM_HUGETLB
3566 #define SHM_HUGETLB 04000
3567 #endif
3568
3569 #define shm_warning_format(format, ...) \
3570 do { \
3571 if (UseLargePages && \
3572 (!FLAG_IS_DEFAULT(UseLargePages) || \
3573 !FLAG_IS_DEFAULT(UseSHM) || \
3574 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3575 warning(format, __VA_ARGS__); \
3576 } \
3577 } while (0)
3578
3579 #define shm_warning(str) shm_warning_format("%s", str)
3580
3581 #define shm_warning_with_errno(str) \
3582 do { \
3583 int err = errno; \
3584 shm_warning_format(str " (error = %d)", err); \
3585 } while (0)
3586
3587 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3588 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3589
3590 if (!is_aligned(alignment, SHMLBA)) {
3591 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3592 return NULL;
3593 }
3594
3595 // To ensure that we get 'alignment' aligned memory from shmat,
3596 // we pre-reserve aligned virtual memory and then attach to that.
3597
3598 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3599 if (pre_reserved_addr == NULL) {
3600 // Couldn't pre-reserve aligned memory.
3601 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3602 return NULL;
3603 }
3604
3605 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3606 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3607
3608 if ((intptr_t)addr == -1) {
3609 int err = errno;
3610 shm_warning_with_errno("Failed to attach shared memory.");
3611
3612 assert(err != EACCES, "Unexpected error");
3613 assert(err != EIDRM, "Unexpected error");
3614 assert(err != EINVAL, "Unexpected error");
3615
3616 // Since we don't know if the kernel unmapped the pre-reserved memory area
3617 // we can't unmap it, since that would potentially unmap memory that was
3618 // mapped from other threads.
3619 return NULL;
3620 }
3621
3622 return addr;
3623 }
3624
3625 static char* shmat_at_address(int shmid, char* req_addr) {
3626 if (!is_aligned(req_addr, SHMLBA)) {
3627 assert(false, "Requested address needs to be SHMLBA aligned");
3628 return NULL;
3629 }
3630
3631 char* addr = (char*)shmat(shmid, req_addr, 0);
3632
3633 if ((intptr_t)addr == -1) {
3634 shm_warning_with_errno("Failed to attach shared memory.");
3635 return NULL;
3636 }
3637
3638 return addr;
3639 }
3640
3641 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3642 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3643 if (req_addr != NULL) {
3644 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3645 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3646 return shmat_at_address(shmid, req_addr);
3647 }
3648
3649 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3650 // return large page size aligned memory addresses when req_addr == NULL.
3651 // However, if the alignment is larger than the large page size, we have
3652 // to manually ensure that the memory returned is 'alignment' aligned.
3653 if (alignment > os::large_page_size()) {
3654 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3655 return shmat_with_alignment(shmid, bytes, alignment);
3656 } else {
3657 return shmat_at_address(shmid, NULL);
3658 }
3659 }
3660
3661 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3662 char* req_addr, bool exec) {
3663 // "exec" is passed in but not used. Creating the shared image for
3664 // the code cache doesn't have an SHM_X executable permission to check.
3665 assert(UseLargePages && UseSHM, "only for SHM large pages");
3666 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3667 assert(is_aligned(req_addr, alignment), "Unaligned address");
3668
3669 if (!is_aligned(bytes, os::large_page_size())) {
3670 return NULL; // Fallback to small pages.
3671 }
3672
3673 // Create a large shared memory region to attach to based on size.
3674 // Currently, size is the total size of the heap.
3675 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3676 if (shmid == -1) {
3677 // Possible reasons for shmget failure:
3678 // 1. shmmax is too small for Java heap.
3679 // > check shmmax value: cat /proc/sys/kernel/shmmax
3680 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3681 // 2. not enough large page memory.
3682 // > check available large pages: cat /proc/meminfo
3683 // > increase amount of large pages:
3684 // echo new_value > /proc/sys/vm/nr_hugepages
3685 // Note 1: different Linux may use different name for this property,
3686 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3687 // Note 2: it's possible there's enough physical memory available but
3688 // they are so fragmented after a long run that they can't
3689 // coalesce into large pages. Try to reserve large pages when
3690 // the system is still "fresh".
3691 shm_warning_with_errno("Failed to reserve shared memory.");
3692 return NULL;
3693 }
3694
3695 // Attach to the region.
3696 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3697
3698 // Remove shmid. If shmat() is successful, the actual shared memory segment
3699 // will be deleted when it's detached by shmdt() or when the process
3700 // terminates. If shmat() is not successful this will remove the shared
3701 // segment immediately.
3702 shmctl(shmid, IPC_RMID, NULL);
3703
3704 return addr;
3705 }
3706
3707 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3708 int error) {
3709 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3710
3711 bool warn_on_failure = UseLargePages &&
3712 (!FLAG_IS_DEFAULT(UseLargePages) ||
3713 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3714 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3715
3716 if (warn_on_failure) {
3717 char msg[128];
3718 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3719 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3720 warning("%s", msg);
3721 }
3722 }
3723
3724 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3725 char* req_addr,
3726 bool exec) {
3727 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3728 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3729 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3730
3731 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3732 char* addr = (char*)::mmap(req_addr, bytes, prot,
3733 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3734 -1, 0);
3735
3736 if (addr == MAP_FAILED) {
3737 warn_on_large_pages_failure(req_addr, bytes, errno);
3738 return NULL;
3739 }
3740
3741 assert(is_aligned(addr, os::large_page_size()), "Must be");
3742
3743 return addr;
3744 }
3745
3746 // Reserve memory using mmap(MAP_HUGETLB).
3747 // - bytes shall be a multiple of alignment.
3748 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3749 // - alignment sets the alignment at which memory shall be allocated.
3750 // It must be a multiple of allocation granularity.
3751 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3752 // req_addr or NULL.
3753 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3754 size_t alignment,
3755 char* req_addr,
3756 bool exec) {
3757 size_t large_page_size = os::large_page_size();
3758 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3759
3760 assert(is_aligned(req_addr, alignment), "Must be");
3761 assert(is_aligned(bytes, alignment), "Must be");
3762
3763 // First reserve - but not commit - the address range in small pages.
3764 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3765
3766 if (start == NULL) {
3767 return NULL;
3768 }
3769
3770 assert(is_aligned(start, alignment), "Must be");
3771
3772 char* end = start + bytes;
3773
3774 // Find the regions of the allocated chunk that can be promoted to large pages.
3775 char* lp_start = align_up(start, large_page_size);
3776 char* lp_end = align_down(end, large_page_size);
3777
3778 size_t lp_bytes = lp_end - lp_start;
3779
3780 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3781
3782 if (lp_bytes == 0) {
3783 // The mapped region doesn't even span the start and the end of a large page.
3784 // Fall back to allocate a non-special area.
3785 ::munmap(start, end - start);
3786 return NULL;
3787 }
3788
3789 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3790
3791 void* result;
3792
3793 // Commit small-paged leading area.
3794 if (start != lp_start) {
3795 result = ::mmap(start, lp_start - start, prot,
3796 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3797 -1, 0);
3798 if (result == MAP_FAILED) {
3799 ::munmap(lp_start, end - lp_start);
3800 return NULL;
3801 }
3802 }
3803
3804 // Commit large-paged area.
3805 result = ::mmap(lp_start, lp_bytes, prot,
3806 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3807 -1, 0);
3808 if (result == MAP_FAILED) {
3809 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3810 // If the mmap above fails, the large pages region will be unmapped and we
3811 // have regions before and after with small pages. Release these regions.
3812 //
3813 // | mapped | unmapped | mapped |
3814 // ^ ^ ^ ^
3815 // start lp_start lp_end end
3816 //
3817 ::munmap(start, lp_start - start);
3818 ::munmap(lp_end, end - lp_end);
3819 return NULL;
3820 }
3821
3822 // Commit small-paged trailing area.
3823 if (lp_end != end) {
3824 result = ::mmap(lp_end, end - lp_end, prot,
3825 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3826 -1, 0);
3827 if (result == MAP_FAILED) {
3828 ::munmap(start, lp_end - start);
3829 return NULL;
3830 }
3831 }
3832
3833 return start;
3834 }
3835
3836 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3837 size_t alignment,
3838 char* req_addr,
3839 bool exec) {
3840 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3841 assert(is_aligned(req_addr, alignment), "Must be");
3842 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3843 assert(is_power_of_2(os::large_page_size()), "Must be");
3844 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3845
3846 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3847 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3848 } else {
3849 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3850 }
3851 }
3852
3853 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3854 char* req_addr, bool exec) {
3855 assert(UseLargePages, "only for large pages");
3856
3857 char* addr;
3858 if (UseSHM) {
3859 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3860 } else {
3861 assert(UseHugeTLBFS, "must be");
3862 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3863 }
3864
3865 if (addr != NULL) {
3866 if (UseNUMAInterleaving) {
3867 numa_make_global(addr, bytes);
3868 }
3869
3870 // The memory is committed
3871 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3872 }
3873
3874 return addr;
3875 }
3876
3877 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3878 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3879 return shmdt(base) == 0;
3880 }
3881
3882 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3883 return pd_release_memory(base, bytes);
3884 }
3885
3886 bool os::release_memory_special(char* base, size_t bytes) {
3887 bool res;
3888 if (MemTracker::tracking_level() > NMT_minimal) {
3889 Tracker tkr(Tracker::release);
3890 res = os::Linux::release_memory_special_impl(base, bytes);
3891 if (res) {
3892 tkr.record((address)base, bytes);
3893 }
3894
3895 } else {
3896 res = os::Linux::release_memory_special_impl(base, bytes);
3897 }
3898 return res;
3899 }
3900
3901 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3902 assert(UseLargePages, "only for large pages");
3903 bool res;
3904
3905 if (UseSHM) {
3906 res = os::Linux::release_memory_special_shm(base, bytes);
3907 } else {
3908 assert(UseHugeTLBFS, "must be");
3909 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3910 }
3911 return res;
3912 }
3913
3914 size_t os::large_page_size() {
3915 return _large_page_size;
3916 }
3917
3918 // With SysV SHM the entire memory region must be allocated as shared
3919 // memory.
3920 // HugeTLBFS allows application to commit large page memory on demand.
3921 // However, when committing memory with HugeTLBFS fails, the region
3922 // that was supposed to be committed will lose the old reservation
3923 // and allow other threads to steal that memory region. Because of this
3924 // behavior we can't commit HugeTLBFS memory.
3925 bool os::can_commit_large_page_memory() {
3926 return UseTransparentHugePages;
3927 }
3928
3929 bool os::can_execute_large_page_memory() {
3930 return UseTransparentHugePages || UseHugeTLBFS;
3931 }
3932
3933 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3934 assert(file_desc >= 0, "file_desc is not valid");
3935 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3936 if (result != NULL) {
3937 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3938 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3939 }
3940 }
3941 return result;
3942 }
3943
3944 // Reserve memory at an arbitrary address, only if that area is
3945 // available (and not reserved for something else).
3946
3947 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3948 const int max_tries = 10;
3949 char* base[max_tries];
3950 size_t size[max_tries];
3951 const size_t gap = 0x000000;
3952
3953 // Assert only that the size is a multiple of the page size, since
3954 // that's all that mmap requires, and since that's all we really know
3955 // about at this low abstraction level. If we need higher alignment,
3956 // we can either pass an alignment to this method or verify alignment
3957 // in one of the methods further up the call chain. See bug 5044738.
3958 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3959
3960 // Repeatedly allocate blocks until the block is allocated at the
3961 // right spot.
3962
3963 // Linux mmap allows caller to pass an address as hint; give it a try first,
3964 // if kernel honors the hint then we can return immediately.
3965 char * addr = anon_mmap(requested_addr, bytes, false);
3966 if (addr == requested_addr) {
3967 return requested_addr;
3968 }
3969
3970 if (addr != NULL) {
3971 // mmap() is successful but it fails to reserve at the requested address
3972 anon_munmap(addr, bytes);
3973 }
3974
3975 int i;
3976 for (i = 0; i < max_tries; ++i) {
3977 base[i] = reserve_memory(bytes);
3978
3979 if (base[i] != NULL) {
3980 // Is this the block we wanted?
3981 if (base[i] == requested_addr) {
3982 size[i] = bytes;
3983 break;
3984 }
3985
3986 // Does this overlap the block we wanted? Give back the overlapped
3987 // parts and try again.
3988
3989 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3990 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3991 unmap_memory(base[i], top_overlap);
3992 base[i] += top_overlap;
3993 size[i] = bytes - top_overlap;
3994 } else {
3995 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3996 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3997 unmap_memory(requested_addr, bottom_overlap);
3998 size[i] = bytes - bottom_overlap;
3999 } else {
4000 size[i] = bytes;
4001 }
4002 }
4003 }
4004 }
4005
4006 // Give back the unused reserved pieces.
4007
4008 for (int j = 0; j < i; ++j) {
4009 if (base[j] != NULL) {
4010 unmap_memory(base[j], size[j]);
4011 }
4012 }
4013
4014 if (i < max_tries) {
4015 return requested_addr;
4016 } else {
4017 return NULL;
4018 }
4019 }
4020
4021 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4022 return ::read(fd, buf, nBytes);
4023 }
4024
4025 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4026 return ::pread(fd, buf, nBytes, offset);
4027 }
4028
4029 // Short sleep, direct OS call.
4030 //
4031 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4032 // sched_yield(2) will actually give up the CPU:
4033 //
4034 // * Alone on this pariticular CPU, keeps running.
4035 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4036 // (pre 2.6.39).
4037 //
4038 // So calling this with 0 is an alternative.
4039 //
4040 void os::naked_short_sleep(jlong ms) {
4041 struct timespec req;
4042
4043 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4044 req.tv_sec = 0;
4045 if (ms > 0) {
4046 req.tv_nsec = (ms % 1000) * 1000000;
4047 } else {
4048 req.tv_nsec = 1;
4049 }
4050
4051 nanosleep(&req, NULL);
4052
4053 return;
4054 }
4055
4056 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4057 void os::infinite_sleep() {
4058 while (true) { // sleep forever ...
4059 ::sleep(100); // ... 100 seconds at a time
4060 }
4061 }
4062
4063 // Used to convert frequent JVM_Yield() to nops
4064 bool os::dont_yield() {
4065 return DontYieldALot;
4066 }
4067
4068 void os::naked_yield() {
4069 sched_yield();
4070 }
4071
4072 ////////////////////////////////////////////////////////////////////////////////
4073 // thread priority support
4074
4075 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4076 // only supports dynamic priority, static priority must be zero. For real-time
4077 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4078 // However, for large multi-threaded applications, SCHED_RR is not only slower
4079 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4080 // of 5 runs - Sep 2005).
4081 //
4082 // The following code actually changes the niceness of kernel-thread/LWP. It
4083 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4084 // not the entire user process, and user level threads are 1:1 mapped to kernel
4085 // threads. It has always been the case, but could change in the future. For
4086 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4087 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4088
4089 int os::java_to_os_priority[CriticalPriority + 1] = {
4090 19, // 0 Entry should never be used
4091
4092 4, // 1 MinPriority
4093 3, // 2
4094 2, // 3
4095
4096 1, // 4
4097 0, // 5 NormPriority
4098 -1, // 6
4099
4100 -2, // 7
4101 -3, // 8
4102 -4, // 9 NearMaxPriority
4103
4104 -5, // 10 MaxPriority
4105
4106 -5 // 11 CriticalPriority
4107 };
4108
4109 static int prio_init() {
4110 if (ThreadPriorityPolicy == 1) {
4111 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4112 // if effective uid is not root. Perhaps, a more elegant way of doing
4113 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4114 if (geteuid() != 0) {
4115 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4116 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4117 }
4118 ThreadPriorityPolicy = 0;
4119 }
4120 }
4121 if (UseCriticalJavaThreadPriority) {
4122 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4123 }
4124 return 0;
4125 }
4126
4127 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4128 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4129
4130 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4131 return (ret == 0) ? OS_OK : OS_ERR;
4132 }
4133
4134 OSReturn os::get_native_priority(const Thread* const thread,
4135 int *priority_ptr) {
4136 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4137 *priority_ptr = java_to_os_priority[NormPriority];
4138 return OS_OK;
4139 }
4140
4141 errno = 0;
4142 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4143 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4144 }
4145
4146 ////////////////////////////////////////////////////////////////////////////////
4147 // suspend/resume support
4148
4149 // The low-level signal-based suspend/resume support is a remnant from the
4150 // old VM-suspension that used to be for java-suspension, safepoints etc,
4151 // within hotspot. Currently used by JFR's OSThreadSampler
4152 //
4153 // The remaining code is greatly simplified from the more general suspension
4154 // code that used to be used.
4155 //
4156 // The protocol is quite simple:
4157 // - suspend:
4158 // - sends a signal to the target thread
4159 // - polls the suspend state of the osthread using a yield loop
4160 // - target thread signal handler (SR_handler) sets suspend state
4161 // and blocks in sigsuspend until continued
4162 // - resume:
4163 // - sets target osthread state to continue
4164 // - sends signal to end the sigsuspend loop in the SR_handler
4165 //
4166 // Note that the SR_lock plays no role in this suspend/resume protocol,
4167 // but is checked for NULL in SR_handler as a thread termination indicator.
4168 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4169 //
4170 // Note that resume_clear_context() and suspend_save_context() are needed
4171 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4172 // which in part is used by:
4173 // - Forte Analyzer: AsyncGetCallTrace()
4174 // - StackBanging: get_frame_at_stack_banging_point()
4175
4176 static void resume_clear_context(OSThread *osthread) {
4177 osthread->set_ucontext(NULL);
4178 osthread->set_siginfo(NULL);
4179 }
4180
4181 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4182 ucontext_t* context) {
4183 osthread->set_ucontext(context);
4184 osthread->set_siginfo(siginfo);
4185 }
4186
4187 // Handler function invoked when a thread's execution is suspended or
4188 // resumed. We have to be careful that only async-safe functions are
4189 // called here (Note: most pthread functions are not async safe and
4190 // should be avoided.)
4191 //
4192 // Note: sigwait() is a more natural fit than sigsuspend() from an
4193 // interface point of view, but sigwait() prevents the signal hander
4194 // from being run. libpthread would get very confused by not having
4195 // its signal handlers run and prevents sigwait()'s use with the
4196 // mutex granting granting signal.
4197 //
4198 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4199 //
4200 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4201 // Save and restore errno to avoid confusing native code with EINTR
4202 // after sigsuspend.
4203 int old_errno = errno;
4204
4205 Thread* thread = Thread::current_or_null_safe();
4206 assert(thread != NULL, "Missing current thread in SR_handler");
4207
4208 // On some systems we have seen signal delivery get "stuck" until the signal
4209 // mask is changed as part of thread termination. Check that the current thread
4210 // has not already terminated (via SR_lock()) - else the following assertion
4211 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4212 // destructor has completed.
4213
4214 if (thread->SR_lock() == NULL) {
4215 return;
4216 }
4217
4218 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4219
4220 OSThread* osthread = thread->osthread();
4221
4222 os::SuspendResume::State current = osthread->sr.state();
4223 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4224 suspend_save_context(osthread, siginfo, context);
4225
4226 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4227 os::SuspendResume::State state = osthread->sr.suspended();
4228 if (state == os::SuspendResume::SR_SUSPENDED) {
4229 sigset_t suspend_set; // signals for sigsuspend()
4230 sigemptyset(&suspend_set);
4231 // get current set of blocked signals and unblock resume signal
4232 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4233 sigdelset(&suspend_set, SR_signum);
4234
4235 sr_semaphore.signal();
4236 // wait here until we are resumed
4237 while (1) {
4238 sigsuspend(&suspend_set);
4239
4240 os::SuspendResume::State result = osthread->sr.running();
4241 if (result == os::SuspendResume::SR_RUNNING) {
4242 sr_semaphore.signal();
4243 break;
4244 }
4245 }
4246
4247 } else if (state == os::SuspendResume::SR_RUNNING) {
4248 // request was cancelled, continue
4249 } else {
4250 ShouldNotReachHere();
4251 }
4252
4253 resume_clear_context(osthread);
4254 } else if (current == os::SuspendResume::SR_RUNNING) {
4255 // request was cancelled, continue
4256 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4257 // ignore
4258 } else {
4259 // ignore
4260 }
4261
4262 errno = old_errno;
4263 }
4264
4265 static int SR_initialize() {
4266 struct sigaction act;
4267 char *s;
4268
4269 // Get signal number to use for suspend/resume
4270 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4271 int sig = ::strtol(s, 0, 10);
4272 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4273 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4274 SR_signum = sig;
4275 } else {
4276 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4277 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4278 }
4279 }
4280
4281 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4282 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4283
4284 sigemptyset(&SR_sigset);
4285 sigaddset(&SR_sigset, SR_signum);
4286
4287 // Set up signal handler for suspend/resume
4288 act.sa_flags = SA_RESTART|SA_SIGINFO;
4289 act.sa_handler = (void (*)(int)) SR_handler;
4290
4291 // SR_signum is blocked by default.
4292 // 4528190 - We also need to block pthread restart signal (32 on all
4293 // supported Linux platforms). Note that LinuxThreads need to block
4294 // this signal for all threads to work properly. So we don't have
4295 // to use hard-coded signal number when setting up the mask.
4296 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4297
4298 if (sigaction(SR_signum, &act, 0) == -1) {
4299 return -1;
4300 }
4301
4302 // Save signal flag
4303 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4304 return 0;
4305 }
4306
4307 static int sr_notify(OSThread* osthread) {
4308 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4309 assert_status(status == 0, status, "pthread_kill");
4310 return status;
4311 }
4312
4313 // "Randomly" selected value for how long we want to spin
4314 // before bailing out on suspending a thread, also how often
4315 // we send a signal to a thread we want to resume
4316 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4317 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4318
4319 // returns true on success and false on error - really an error is fatal
4320 // but this seems the normal response to library errors
4321 static bool do_suspend(OSThread* osthread) {
4322 assert(osthread->sr.is_running(), "thread should be running");
4323 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4324
4325 // mark as suspended and send signal
4326 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4327 // failed to switch, state wasn't running?
4328 ShouldNotReachHere();
4329 return false;
4330 }
4331
4332 if (sr_notify(osthread) != 0) {
4333 ShouldNotReachHere();
4334 }
4335
4336 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4337 while (true) {
4338 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4339 break;
4340 } else {
4341 // timeout
4342 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4343 if (cancelled == os::SuspendResume::SR_RUNNING) {
4344 return false;
4345 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4346 // make sure that we consume the signal on the semaphore as well
4347 sr_semaphore.wait();
4348 break;
4349 } else {
4350 ShouldNotReachHere();
4351 return false;
4352 }
4353 }
4354 }
4355
4356 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4357 return true;
4358 }
4359
4360 static void do_resume(OSThread* osthread) {
4361 assert(osthread->sr.is_suspended(), "thread should be suspended");
4362 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4363
4364 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4365 // failed to switch to WAKEUP_REQUEST
4366 ShouldNotReachHere();
4367 return;
4368 }
4369
4370 while (true) {
4371 if (sr_notify(osthread) == 0) {
4372 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4373 if (osthread->sr.is_running()) {
4374 return;
4375 }
4376 }
4377 } else {
4378 ShouldNotReachHere();
4379 }
4380 }
4381
4382 guarantee(osthread->sr.is_running(), "Must be running!");
4383 }
4384
4385 ///////////////////////////////////////////////////////////////////////////////////
4386 // signal handling (except suspend/resume)
4387
4388 // This routine may be used by user applications as a "hook" to catch signals.
4389 // The user-defined signal handler must pass unrecognized signals to this
4390 // routine, and if it returns true (non-zero), then the signal handler must
4391 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4392 // routine will never retun false (zero), but instead will execute a VM panic
4393 // routine kill the process.
4394 //
4395 // If this routine returns false, it is OK to call it again. This allows
4396 // the user-defined signal handler to perform checks either before or after
4397 // the VM performs its own checks. Naturally, the user code would be making
4398 // a serious error if it tried to handle an exception (such as a null check
4399 // or breakpoint) that the VM was generating for its own correct operation.
4400 //
4401 // This routine may recognize any of the following kinds of signals:
4402 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4403 // It should be consulted by handlers for any of those signals.
4404 //
4405 // The caller of this routine must pass in the three arguments supplied
4406 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4407 // field of the structure passed to sigaction(). This routine assumes that
4408 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4409 //
4410 // Note that the VM will print warnings if it detects conflicting signal
4411 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4412 //
4413 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4414 siginfo_t* siginfo,
4415 void* ucontext,
4416 int abort_if_unrecognized);
4417
4418 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4419 assert(info != NULL && uc != NULL, "it must be old kernel");
4420 int orig_errno = errno; // Preserve errno value over signal handler.
4421 JVM_handle_linux_signal(sig, info, uc, true);
4422 errno = orig_errno;
4423 }
4424
4425
4426 // This boolean allows users to forward their own non-matching signals
4427 // to JVM_handle_linux_signal, harmlessly.
4428 bool os::Linux::signal_handlers_are_installed = false;
4429
4430 // For signal-chaining
4431 struct sigaction sigact[NSIG];
4432 uint64_t sigs = 0;
4433 #if (64 < NSIG-1)
4434 #error "Not all signals can be encoded in sigs. Adapt its type!"
4435 #endif
4436 bool os::Linux::libjsig_is_loaded = false;
4437 typedef struct sigaction *(*get_signal_t)(int);
4438 get_signal_t os::Linux::get_signal_action = NULL;
4439
4440 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4441 struct sigaction *actp = NULL;
4442
4443 if (libjsig_is_loaded) {
4444 // Retrieve the old signal handler from libjsig
4445 actp = (*get_signal_action)(sig);
4446 }
4447 if (actp == NULL) {
4448 // Retrieve the preinstalled signal handler from jvm
4449 actp = get_preinstalled_handler(sig);
4450 }
4451
4452 return actp;
4453 }
4454
4455 static bool call_chained_handler(struct sigaction *actp, int sig,
4456 siginfo_t *siginfo, void *context) {
4457 // Call the old signal handler
4458 if (actp->sa_handler == SIG_DFL) {
4459 // It's more reasonable to let jvm treat it as an unexpected exception
4460 // instead of taking the default action.
4461 return false;
4462 } else if (actp->sa_handler != SIG_IGN) {
4463 if ((actp->sa_flags & SA_NODEFER) == 0) {
4464 // automaticlly block the signal
4465 sigaddset(&(actp->sa_mask), sig);
4466 }
4467
4468 sa_handler_t hand = NULL;
4469 sa_sigaction_t sa = NULL;
4470 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4471 // retrieve the chained handler
4472 if (siginfo_flag_set) {
4473 sa = actp->sa_sigaction;
4474 } else {
4475 hand = actp->sa_handler;
4476 }
4477
4478 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4479 actp->sa_handler = SIG_DFL;
4480 }
4481
4482 // try to honor the signal mask
4483 sigset_t oset;
4484 sigemptyset(&oset);
4485 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4486
4487 // call into the chained handler
4488 if (siginfo_flag_set) {
4489 (*sa)(sig, siginfo, context);
4490 } else {
4491 (*hand)(sig);
4492 }
4493
4494 // restore the signal mask
4495 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4496 }
4497 // Tell jvm's signal handler the signal is taken care of.
4498 return true;
4499 }
4500
4501 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4502 bool chained = false;
4503 // signal-chaining
4504 if (UseSignalChaining) {
4505 struct sigaction *actp = get_chained_signal_action(sig);
4506 if (actp != NULL) {
4507 chained = call_chained_handler(actp, sig, siginfo, context);
4508 }
4509 }
4510 return chained;
4511 }
4512
4513 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4514 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4515 return &sigact[sig];
4516 }
4517 return NULL;
4518 }
4519
4520 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4521 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4522 sigact[sig] = oldAct;
4523 sigs |= (uint64_t)1 << (sig-1);
4524 }
4525
4526 // for diagnostic
4527 int sigflags[NSIG];
4528
4529 int os::Linux::get_our_sigflags(int sig) {
4530 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4531 return sigflags[sig];
4532 }
4533
4534 void os::Linux::set_our_sigflags(int sig, int flags) {
4535 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4536 if (sig > 0 && sig < NSIG) {
4537 sigflags[sig] = flags;
4538 }
4539 }
4540
4541 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4542 // Check for overwrite.
4543 struct sigaction oldAct;
4544 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4545
4546 void* oldhand = oldAct.sa_sigaction
4547 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4548 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4549 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4550 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4551 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4552 if (AllowUserSignalHandlers || !set_installed) {
4553 // Do not overwrite; user takes responsibility to forward to us.
4554 return;
4555 } else if (UseSignalChaining) {
4556 // save the old handler in jvm
4557 save_preinstalled_handler(sig, oldAct);
4558 // libjsig also interposes the sigaction() call below and saves the
4559 // old sigaction on it own.
4560 } else {
4561 fatal("Encountered unexpected pre-existing sigaction handler "
4562 "%#lx for signal %d.", (long)oldhand, sig);
4563 }
4564 }
4565
4566 struct sigaction sigAct;
4567 sigfillset(&(sigAct.sa_mask));
4568 sigAct.sa_handler = SIG_DFL;
4569 if (!set_installed) {
4570 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4571 } else {
4572 sigAct.sa_sigaction = signalHandler;
4573 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4574 }
4575 // Save flags, which are set by ours
4576 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4577 sigflags[sig] = sigAct.sa_flags;
4578
4579 int ret = sigaction(sig, &sigAct, &oldAct);
4580 assert(ret == 0, "check");
4581
4582 void* oldhand2 = oldAct.sa_sigaction
4583 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4584 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4585 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4586 }
4587
4588 // install signal handlers for signals that HotSpot needs to
4589 // handle in order to support Java-level exception handling.
4590
4591 void os::Linux::install_signal_handlers() {
4592 if (!signal_handlers_are_installed) {
4593 signal_handlers_are_installed = true;
4594
4595 // signal-chaining
4596 typedef void (*signal_setting_t)();
4597 signal_setting_t begin_signal_setting = NULL;
4598 signal_setting_t end_signal_setting = NULL;
4599 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4600 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4601 if (begin_signal_setting != NULL) {
4602 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4603 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4604 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4605 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4606 libjsig_is_loaded = true;
4607 assert(UseSignalChaining, "should enable signal-chaining");
4608 }
4609 if (libjsig_is_loaded) {
4610 // Tell libjsig jvm is setting signal handlers
4611 (*begin_signal_setting)();
4612 }
4613
4614 set_signal_handler(SIGSEGV, true);
4615 set_signal_handler(SIGPIPE, true);
4616 set_signal_handler(SIGBUS, true);
4617 set_signal_handler(SIGILL, true);
4618 set_signal_handler(SIGFPE, true);
4619 #if defined(PPC64)
4620 set_signal_handler(SIGTRAP, true);
4621 #endif
4622 set_signal_handler(SIGXFSZ, true);
4623
4624 if (libjsig_is_loaded) {
4625 // Tell libjsig jvm finishes setting signal handlers
4626 (*end_signal_setting)();
4627 }
4628
4629 // We don't activate signal checker if libjsig is in place, we trust ourselves
4630 // and if UserSignalHandler is installed all bets are off.
4631 // Log that signal checking is off only if -verbose:jni is specified.
4632 if (CheckJNICalls) {
4633 if (libjsig_is_loaded) {
4634 if (PrintJNIResolving) {
4635 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4636 }
4637 check_signals = false;
4638 }
4639 if (AllowUserSignalHandlers) {
4640 if (PrintJNIResolving) {
4641 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4642 }
4643 check_signals = false;
4644 }
4645 }
4646 }
4647 }
4648
4649 // This is the fastest way to get thread cpu time on Linux.
4650 // Returns cpu time (user+sys) for any thread, not only for current.
4651 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4652 // It might work on 2.6.10+ with a special kernel/glibc patch.
4653 // For reference, please, see IEEE Std 1003.1-2004:
4654 // http://www.unix.org/single_unix_specification
4655
4656 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4657 struct timespec tp;
4658 int rc = os::Posix::clock_gettime(clockid, &tp);
4659 assert(rc == 0, "clock_gettime is expected to return 0 code");
4660
4661 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4662 }
4663
4664 void os::Linux::initialize_os_info() {
4665 assert(_os_version == 0, "OS info already initialized");
4666
4667 struct utsname _uname;
4668
4669 uint32_t major;
4670 uint32_t minor;
4671 uint32_t fix;
4672
4673 int rc;
4674
4675 // Kernel version is unknown if
4676 // verification below fails.
4677 _os_version = 0x01000000;
4678
4679 rc = uname(&_uname);
4680 if (rc != -1) {
4681
4682 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4683 if (rc == 3) {
4684
4685 if (major < 256 && minor < 256 && fix < 256) {
4686 // Kernel version format is as expected,
4687 // set it overriding unknown state.
4688 _os_version = (major << 16) |
4689 (minor << 8 ) |
4690 (fix << 0 ) ;
4691 }
4692 }
4693 }
4694 }
4695
4696 uint32_t os::Linux::os_version() {
4697 assert(_os_version != 0, "not initialized");
4698 return _os_version & 0x00FFFFFF;
4699 }
4700
4701 bool os::Linux::os_version_is_known() {
4702 assert(_os_version != 0, "not initialized");
4703 return _os_version & 0x01000000 ? false : true;
4704 }
4705
4706 /////
4707 // glibc on Linux platform uses non-documented flag
4708 // to indicate, that some special sort of signal
4709 // trampoline is used.
4710 // We will never set this flag, and we should
4711 // ignore this flag in our diagnostic
4712 #ifdef SIGNIFICANT_SIGNAL_MASK
4713 #undef SIGNIFICANT_SIGNAL_MASK
4714 #endif
4715 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4716
4717 static const char* get_signal_handler_name(address handler,
4718 char* buf, int buflen) {
4719 int offset = 0;
4720 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4721 if (found) {
4722 // skip directory names
4723 const char *p1, *p2;
4724 p1 = buf;
4725 size_t len = strlen(os::file_separator());
4726 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4727 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4728 } else {
4729 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4730 }
4731 return buf;
4732 }
4733
4734 static void print_signal_handler(outputStream* st, int sig,
4735 char* buf, size_t buflen) {
4736 struct sigaction sa;
4737
4738 sigaction(sig, NULL, &sa);
4739
4740 // See comment for SIGNIFICANT_SIGNAL_MASK define
4741 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4742
4743 st->print("%s: ", os::exception_name(sig, buf, buflen));
4744
4745 address handler = (sa.sa_flags & SA_SIGINFO)
4746 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4747 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4748
4749 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4750 st->print("SIG_DFL");
4751 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4752 st->print("SIG_IGN");
4753 } else {
4754 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4755 }
4756
4757 st->print(", sa_mask[0]=");
4758 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4759
4760 address rh = VMError::get_resetted_sighandler(sig);
4761 // May be, handler was resetted by VMError?
4762 if (rh != NULL) {
4763 handler = rh;
4764 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4765 }
4766
4767 st->print(", sa_flags=");
4768 os::Posix::print_sa_flags(st, sa.sa_flags);
4769
4770 // Check: is it our handler?
4771 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4772 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4773 // It is our signal handler
4774 // check for flags, reset system-used one!
4775 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4776 st->print(
4777 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4778 os::Linux::get_our_sigflags(sig));
4779 }
4780 }
4781 st->cr();
4782 }
4783
4784
4785 #define DO_SIGNAL_CHECK(sig) \
4786 do { \
4787 if (!sigismember(&check_signal_done, sig)) { \
4788 os::Linux::check_signal_handler(sig); \
4789 } \
4790 } while (0)
4791
4792 // This method is a periodic task to check for misbehaving JNI applications
4793 // under CheckJNI, we can add any periodic checks here
4794
4795 void os::run_periodic_checks() {
4796 if (check_signals == false) return;
4797
4798 // SEGV and BUS if overridden could potentially prevent
4799 // generation of hs*.log in the event of a crash, debugging
4800 // such a case can be very challenging, so we absolutely
4801 // check the following for a good measure:
4802 DO_SIGNAL_CHECK(SIGSEGV);
4803 DO_SIGNAL_CHECK(SIGILL);
4804 DO_SIGNAL_CHECK(SIGFPE);
4805 DO_SIGNAL_CHECK(SIGBUS);
4806 DO_SIGNAL_CHECK(SIGPIPE);
4807 DO_SIGNAL_CHECK(SIGXFSZ);
4808 #if defined(PPC64)
4809 DO_SIGNAL_CHECK(SIGTRAP);
4810 #endif
4811
4812 // ReduceSignalUsage allows the user to override these handlers
4813 // see comments at the very top and jvm_md.h
4814 if (!ReduceSignalUsage) {
4815 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4816 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4817 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4818 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4819 }
4820
4821 DO_SIGNAL_CHECK(SR_signum);
4822 }
4823
4824 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4825
4826 static os_sigaction_t os_sigaction = NULL;
4827
4828 void os::Linux::check_signal_handler(int sig) {
4829 char buf[O_BUFLEN];
4830 address jvmHandler = NULL;
4831
4832
4833 struct sigaction act;
4834 if (os_sigaction == NULL) {
4835 // only trust the default sigaction, in case it has been interposed
4836 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4837 if (os_sigaction == NULL) return;
4838 }
4839
4840 os_sigaction(sig, (struct sigaction*)NULL, &act);
4841
4842
4843 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4844
4845 address thisHandler = (act.sa_flags & SA_SIGINFO)
4846 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4847 : CAST_FROM_FN_PTR(address, act.sa_handler);
4848
4849
4850 switch (sig) {
4851 case SIGSEGV:
4852 case SIGBUS:
4853 case SIGFPE:
4854 case SIGPIPE:
4855 case SIGILL:
4856 case SIGXFSZ:
4857 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4858 break;
4859
4860 case SHUTDOWN1_SIGNAL:
4861 case SHUTDOWN2_SIGNAL:
4862 case SHUTDOWN3_SIGNAL:
4863 case BREAK_SIGNAL:
4864 jvmHandler = (address)user_handler();
4865 break;
4866
4867 default:
4868 if (sig == SR_signum) {
4869 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4870 } else {
4871 return;
4872 }
4873 break;
4874 }
4875
4876 if (thisHandler != jvmHandler) {
4877 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4878 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4879 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4880 // No need to check this sig any longer
4881 sigaddset(&check_signal_done, sig);
4882 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4883 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4884 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4885 exception_name(sig, buf, O_BUFLEN));
4886 }
4887 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4888 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4889 tty->print("expected:");
4890 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4891 tty->cr();
4892 tty->print(" found:");
4893 os::Posix::print_sa_flags(tty, act.sa_flags);
4894 tty->cr();
4895 // No need to check this sig any longer
4896 sigaddset(&check_signal_done, sig);
4897 }
4898
4899 // Dump all the signal
4900 if (sigismember(&check_signal_done, sig)) {
4901 print_signal_handlers(tty, buf, O_BUFLEN);
4902 }
4903 }
4904
4905 extern void report_error(char* file_name, int line_no, char* title,
4906 char* format, ...);
4907
4908 // this is called _before_ most of the global arguments have been parsed
4909 void os::init(void) {
4910 char dummy; // used to get a guess on initial stack address
4911
4912 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4913
4914 init_random(1234567);
4915
4916 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4917 if (Linux::page_size() == -1) {
4918 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4919 os::strerror(errno));
4920 }
4921 init_page_sizes((size_t) Linux::page_size());
4922
4923 Linux::initialize_system_info();
4924
4925 Linux::initialize_os_info();
4926
4927 // _main_thread points to the thread that created/loaded the JVM.
4928 Linux::_main_thread = pthread_self();
4929
4930 // retrieve entry point for pthread_setname_np
4931 Linux::_pthread_setname_np =
4932 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4933
4934 os::Posix::init();
4935
4936 initial_time_count = javaTimeNanos();
4937
4938 // Always warn if no monotonic clock available
4939 if (!os::Posix::supports_monotonic_clock()) {
4940 warning("No monotonic clock was available - timed services may " \
4941 "be adversely affected if the time-of-day clock changes");
4942 }
4943 }
4944
4945 // To install functions for atexit system call
4946 extern "C" {
4947 static void perfMemory_exit_helper() {
4948 perfMemory_exit();
4949 }
4950 }
4951
4952 void os::pd_init_container_support() {
4953 OSContainer::init();
4954 }
4955
4956 // this is called _after_ the global arguments have been parsed
4957 jint os::init_2(void) {
4958
4959 os::Posix::init_2();
4960
4961 Linux::fast_thread_clock_init();
4962
4963 // initialize suspend/resume support - must do this before signal_sets_init()
4964 if (SR_initialize() != 0) {
4965 perror("SR_initialize failed");
4966 return JNI_ERR;
4967 }
4968
4969 Linux::signal_sets_init();
4970 Linux::install_signal_handlers();
4971 // Initialize data for jdk.internal.misc.Signal
4972 if (!ReduceSignalUsage) {
4973 jdk_misc_signal_init();
4974 }
4975
4976 // Check and sets minimum stack sizes against command line options
4977 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4978 return JNI_ERR;
4979 }
4980
4981 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
4982 if (!suppress_primordial_thread_resolution) {
4983 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4984 }
4985
4986 #if defined(IA32)
4987 workaround_expand_exec_shield_cs_limit();
4988 #endif
4989
4990 Linux::libpthread_init();
4991 Linux::sched_getcpu_init();
4992 log_info(os)("HotSpot is running with %s, %s",
4993 Linux::glibc_version(), Linux::libpthread_version());
4994
4995 if (UseNUMA) {
4996 if (!Linux::libnuma_init()) {
4997 UseNUMA = false;
4998 } else {
4999 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5000 // If there's only one node (they start from 0) or if the process
5001 // is bound explicitly to a single node using membind, disable NUMA.
5002 UseNUMA = false;
5003 }
5004 }
5005
5006 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5007 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5008 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5009 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5010 // and disable adaptive resizing.
5011 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5012 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5013 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5014 UseAdaptiveSizePolicy = false;
5015 UseAdaptiveNUMAChunkSizing = false;
5016 }
5017 }
5018
5019 if (!UseNUMA && ForceNUMA) {
5020 UseNUMA = true;
5021 }
5022 }
5023
5024 if (MaxFDLimit) {
5025 // set the number of file descriptors to max. print out error
5026 // if getrlimit/setrlimit fails but continue regardless.
5027 struct rlimit nbr_files;
5028 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5029 if (status != 0) {
5030 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5031 } else {
5032 nbr_files.rlim_cur = nbr_files.rlim_max;
5033 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5034 if (status != 0) {
5035 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5036 }
5037 }
5038 }
5039
5040 // Initialize lock used to serialize thread creation (see os::create_thread)
5041 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5042
5043 // at-exit methods are called in the reverse order of their registration.
5044 // atexit functions are called on return from main or as a result of a
5045 // call to exit(3C). There can be only 32 of these functions registered
5046 // and atexit() does not set errno.
5047
5048 if (PerfAllowAtExitRegistration) {
5049 // only register atexit functions if PerfAllowAtExitRegistration is set.
5050 // atexit functions can be delayed until process exit time, which
5051 // can be problematic for embedded VM situations. Embedded VMs should
5052 // call DestroyJavaVM() to assure that VM resources are released.
5053
5054 // note: perfMemory_exit_helper atexit function may be removed in
5055 // the future if the appropriate cleanup code can be added to the
5056 // VM_Exit VMOperation's doit method.
5057 if (atexit(perfMemory_exit_helper) != 0) {
5058 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5059 }
5060 }
5061
5062 // initialize thread priority policy
5063 prio_init();
5064
5065 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5066 set_coredump_filter(DAX_SHARED_BIT);
5067 }
5068
5069 if (UseSharedSpaces) {
5070 set_coredump_filter(FILE_BACKED_PVT_BIT);
5071 }
5072
5073 return JNI_OK;
5074 }
5075
5076 // Mark the polling page as unreadable
5077 void os::make_polling_page_unreadable(void) {
5078 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5079 fatal("Could not disable polling page");
5080 }
5081 }
5082
5083 // Mark the polling page as readable
5084 void os::make_polling_page_readable(void) {
5085 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5086 fatal("Could not enable polling page");
5087 }
5088 }
5089
5090 // older glibc versions don't have this macro (which expands to
5091 // an optimized bit-counting function) so we have to roll our own
5092 #ifndef CPU_COUNT
5093
5094 static int _cpu_count(const cpu_set_t* cpus) {
5095 int count = 0;
5096 // only look up to the number of configured processors
5097 for (int i = 0; i < os::processor_count(); i++) {
5098 if (CPU_ISSET(i, cpus)) {
5099 count++;
5100 }
5101 }
5102 return count;
5103 }
5104
5105 #define CPU_COUNT(cpus) _cpu_count(cpus)
5106
5107 #endif // CPU_COUNT
5108
5109 // Get the current number of available processors for this process.
5110 // This value can change at any time during a process's lifetime.
5111 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5112 // If it appears there may be more than 1024 processors then we do a
5113 // dynamic check - see 6515172 for details.
5114 // If anything goes wrong we fallback to returning the number of online
5115 // processors - which can be greater than the number available to the process.
5116 int os::Linux::active_processor_count() {
5117 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5118 cpu_set_t* cpus_p = &cpus;
5119 int cpus_size = sizeof(cpu_set_t);
5120
5121 int configured_cpus = os::processor_count(); // upper bound on available cpus
5122 int cpu_count = 0;
5123
5124 // old build platforms may not support dynamic cpu sets
5125 #ifdef CPU_ALLOC
5126
5127 // To enable easy testing of the dynamic path on different platforms we
5128 // introduce a diagnostic flag: UseCpuAllocPath
5129 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5130 // kernel may use a mask bigger than cpu_set_t
5131 log_trace(os)("active_processor_count: using dynamic path %s"
5132 "- configured processors: %d",
5133 UseCpuAllocPath ? "(forced) " : "",
5134 configured_cpus);
5135 cpus_p = CPU_ALLOC(configured_cpus);
5136 if (cpus_p != NULL) {
5137 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5138 // zero it just to be safe
5139 CPU_ZERO_S(cpus_size, cpus_p);
5140 }
5141 else {
5142 // failed to allocate so fallback to online cpus
5143 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5144 log_trace(os)("active_processor_count: "
5145 "CPU_ALLOC failed (%s) - using "
5146 "online processor count: %d",
5147 os::strerror(errno), online_cpus);
5148 return online_cpus;
5149 }
5150 }
5151 else {
5152 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5153 configured_cpus);
5154 }
5155 #else // CPU_ALLOC
5156 // these stubs won't be executed
5157 #define CPU_COUNT_S(size, cpus) -1
5158 #define CPU_FREE(cpus)
5159
5160 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5161 configured_cpus);
5162 #endif // CPU_ALLOC
5163
5164 // pid 0 means the current thread - which we have to assume represents the process
5165 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5166 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5167 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5168 }
5169 else {
5170 cpu_count = CPU_COUNT(cpus_p);
5171 }
5172 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5173 }
5174 else {
5175 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5176 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5177 "which may exceed available processors", os::strerror(errno), cpu_count);
5178 }
5179
5180 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5181 CPU_FREE(cpus_p);
5182 }
5183
5184 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5185 return cpu_count;
5186 }
5187
5188 // Determine the active processor count from one of
5189 // three different sources:
5190 //
5191 // 1. User option -XX:ActiveProcessorCount
5192 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5193 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5194 //
5195 // Option 1, if specified, will always override.
5196 // If the cgroup subsystem is active and configured, we
5197 // will return the min of the cgroup and option 2 results.
5198 // This is required since tools, such as numactl, that
5199 // alter cpu affinity do not update cgroup subsystem
5200 // cpuset configuration files.
5201 int os::active_processor_count() {
5202 // User has overridden the number of active processors
5203 if (ActiveProcessorCount > 0) {
5204 log_trace(os)("active_processor_count: "
5205 "active processor count set by user : %d",
5206 ActiveProcessorCount);
5207 return ActiveProcessorCount;
5208 }
5209
5210 int active_cpus;
5211 if (OSContainer::is_containerized()) {
5212 active_cpus = OSContainer::active_processor_count();
5213 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5214 active_cpus);
5215 } else {
5216 active_cpus = os::Linux::active_processor_count();
5217 }
5218
5219 return active_cpus;
5220 }
5221
5222 uint os::processor_id() {
5223 const int id = Linux::sched_getcpu();
5224 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5225 return (uint)id;
5226 }
5227
5228 void os::set_native_thread_name(const char *name) {
5229 if (Linux::_pthread_setname_np) {
5230 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5231 snprintf(buf, sizeof(buf), "%s", name);
5232 buf[sizeof(buf) - 1] = '\0';
5233 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5234 // ERANGE should not happen; all other errors should just be ignored.
5235 assert(rc != ERANGE, "pthread_setname_np failed");
5236 }
5237 }
5238
5239 bool os::distribute_processes(uint length, uint* distribution) {
5240 // Not yet implemented.
5241 return false;
5242 }
5243
5244 bool os::bind_to_processor(uint processor_id) {
5245 // Not yet implemented.
5246 return false;
5247 }
5248
5249 ///
5250
5251 void os::SuspendedThreadTask::internal_do_task() {
5252 if (do_suspend(_thread->osthread())) {
5253 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5254 do_task(context);
5255 do_resume(_thread->osthread());
5256 }
5257 }
5258
5259 ////////////////////////////////////////////////////////////////////////////////
5260 // debug support
5261
5262 bool os::find(address addr, outputStream* st) {
5263 Dl_info dlinfo;
5264 memset(&dlinfo, 0, sizeof(dlinfo));
5265 if (dladdr(addr, &dlinfo) != 0) {
5266 st->print(PTR_FORMAT ": ", p2i(addr));
5267 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5268 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5269 p2i(addr) - p2i(dlinfo.dli_saddr));
5270 } else if (dlinfo.dli_fbase != NULL) {
5271 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5272 } else {
5273 st->print("<absolute address>");
5274 }
5275 if (dlinfo.dli_fname != NULL) {
5276 st->print(" in %s", dlinfo.dli_fname);
5277 }
5278 if (dlinfo.dli_fbase != NULL) {
5279 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5280 }
5281 st->cr();
5282
5283 if (Verbose) {
5284 // decode some bytes around the PC
5285 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5286 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5287 address lowest = (address) dlinfo.dli_sname;
5288 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5289 if (begin < lowest) begin = lowest;
5290 Dl_info dlinfo2;
5291 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5292 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5293 end = (address) dlinfo2.dli_saddr;
5294 }
5295 Disassembler::decode(begin, end, st);
5296 }
5297 return true;
5298 }
5299 return false;
5300 }
5301
5302 ////////////////////////////////////////////////////////////////////////////////
5303 // misc
5304
5305 // This does not do anything on Linux. This is basically a hook for being
5306 // able to use structured exception handling (thread-local exception filters)
5307 // on, e.g., Win32.
5308 void
5309 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5310 JavaCallArguments* args, Thread* thread) {
5311 f(value, method, args, thread);
5312 }
5313
5314 void os::print_statistics() {
5315 }
5316
5317 bool os::message_box(const char* title, const char* message) {
5318 int i;
5319 fdStream err(defaultStream::error_fd());
5320 for (i = 0; i < 78; i++) err.print_raw("=");
5321 err.cr();
5322 err.print_raw_cr(title);
5323 for (i = 0; i < 78; i++) err.print_raw("-");
5324 err.cr();
5325 err.print_raw_cr(message);
5326 for (i = 0; i < 78; i++) err.print_raw("=");
5327 err.cr();
5328
5329 char buf[16];
5330 // Prevent process from exiting upon "read error" without consuming all CPU
5331 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5332
5333 return buf[0] == 'y' || buf[0] == 'Y';
5334 }
5335
5336 // Is a (classpath) directory empty?
5337 bool os::dir_is_empty(const char* path) {
5338 DIR *dir = NULL;
5339 struct dirent *ptr;
5340
5341 dir = opendir(path);
5342 if (dir == NULL) return true;
5343
5344 // Scan the directory
5345 bool result = true;
5346 while (result && (ptr = readdir(dir)) != NULL) {
5347 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5348 result = false;
5349 }
5350 }
5351 closedir(dir);
5352 return result;
5353 }
5354
5355 // This code originates from JDK's sysOpen and open64_w
5356 // from src/solaris/hpi/src/system_md.c
5357
5358 int os::open(const char *path, int oflag, int mode) {
5359 if (strlen(path) > MAX_PATH - 1) {
5360 errno = ENAMETOOLONG;
5361 return -1;
5362 }
5363
5364 // All file descriptors that are opened in the Java process and not
5365 // specifically destined for a subprocess should have the close-on-exec
5366 // flag set. If we don't set it, then careless 3rd party native code
5367 // might fork and exec without closing all appropriate file descriptors
5368 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5369 // turn might:
5370 //
5371 // - cause end-of-file to fail to be detected on some file
5372 // descriptors, resulting in mysterious hangs, or
5373 //
5374 // - might cause an fopen in the subprocess to fail on a system
5375 // suffering from bug 1085341.
5376 //
5377 // (Yes, the default setting of the close-on-exec flag is a Unix
5378 // design flaw)
5379 //
5380 // See:
5381 // 1085341: 32-bit stdio routines should support file descriptors >255
5382 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5383 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5384 //
5385 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5386 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5387 // because it saves a system call and removes a small window where the flag
5388 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5389 // and we fall back to using FD_CLOEXEC (see below).
5390 #ifdef O_CLOEXEC
5391 oflag |= O_CLOEXEC;
5392 #endif
5393
5394 int fd = ::open64(path, oflag, mode);
5395 if (fd == -1) return -1;
5396
5397 //If the open succeeded, the file might still be a directory
5398 {
5399 struct stat64 buf64;
5400 int ret = ::fstat64(fd, &buf64);
5401 int st_mode = buf64.st_mode;
5402
5403 if (ret != -1) {
5404 if ((st_mode & S_IFMT) == S_IFDIR) {
5405 errno = EISDIR;
5406 ::close(fd);
5407 return -1;
5408 }
5409 } else {
5410 ::close(fd);
5411 return -1;
5412 }
5413 }
5414
5415 #ifdef FD_CLOEXEC
5416 // Validate that the use of the O_CLOEXEC flag on open above worked.
5417 // With recent kernels, we will perform this check exactly once.
5418 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5419 if (!O_CLOEXEC_is_known_to_work) {
5420 int flags = ::fcntl(fd, F_GETFD);
5421 if (flags != -1) {
5422 if ((flags & FD_CLOEXEC) != 0)
5423 O_CLOEXEC_is_known_to_work = 1;
5424 else
5425 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5426 }
5427 }
5428 #endif
5429
5430 return fd;
5431 }
5432
5433
5434 // create binary file, rewriting existing file if required
5435 int os::create_binary_file(const char* path, bool rewrite_existing) {
5436 int oflags = O_WRONLY | O_CREAT;
5437 if (!rewrite_existing) {
5438 oflags |= O_EXCL;
5439 }
5440 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5441 }
5442
5443 // return current position of file pointer
5444 jlong os::current_file_offset(int fd) {
5445 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5446 }
5447
5448 // move file pointer to the specified offset
5449 jlong os::seek_to_file_offset(int fd, jlong offset) {
5450 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5451 }
5452
5453 // This code originates from JDK's sysAvailable
5454 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5455
5456 int os::available(int fd, jlong *bytes) {
5457 jlong cur, end;
5458 int mode;
5459 struct stat64 buf64;
5460
5461 if (::fstat64(fd, &buf64) >= 0) {
5462 mode = buf64.st_mode;
5463 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5464 int n;
5465 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5466 *bytes = n;
5467 return 1;
5468 }
5469 }
5470 }
5471 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5472 return 0;
5473 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5474 return 0;
5475 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5476 return 0;
5477 }
5478 *bytes = end - cur;
5479 return 1;
5480 }
5481
5482 // Map a block of memory.
5483 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5484 char *addr, size_t bytes, bool read_only,
5485 bool allow_exec) {
5486 int prot;
5487 int flags = MAP_PRIVATE;
5488
5489 if (read_only) {
5490 prot = PROT_READ;
5491 } else {
5492 prot = PROT_READ | PROT_WRITE;
5493 }
5494
5495 if (allow_exec) {
5496 prot |= PROT_EXEC;
5497 }
5498
5499 if (addr != NULL) {
5500 flags |= MAP_FIXED;
5501 }
5502
5503 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5504 fd, file_offset);
5505 if (mapped_address == MAP_FAILED) {
5506 return NULL;
5507 }
5508 return mapped_address;
5509 }
5510
5511
5512 // Remap a block of memory.
5513 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5514 char *addr, size_t bytes, bool read_only,
5515 bool allow_exec) {
5516 // same as map_memory() on this OS
5517 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5518 allow_exec);
5519 }
5520
5521
5522 // Unmap a block of memory.
5523 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5524 return munmap(addr, bytes) == 0;
5525 }
5526
5527 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5528
5529 static jlong fast_cpu_time(Thread *thread) {
5530 clockid_t clockid;
5531 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5532 &clockid);
5533 if (rc == 0) {
5534 return os::Linux::fast_thread_cpu_time(clockid);
5535 } else {
5536 // It's possible to encounter a terminated native thread that failed
5537 // to detach itself from the VM - which should result in ESRCH.
5538 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5539 return -1;
5540 }
5541 }
5542
5543 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5544 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5545 // of a thread.
5546 //
5547 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5548 // the fast estimate available on the platform.
5549
5550 jlong os::current_thread_cpu_time() {
5551 if (os::Linux::supports_fast_thread_cpu_time()) {
5552 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5553 } else {
5554 // return user + sys since the cost is the same
5555 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5556 }
5557 }
5558
5559 jlong os::thread_cpu_time(Thread* thread) {
5560 // consistent with what current_thread_cpu_time() returns
5561 if (os::Linux::supports_fast_thread_cpu_time()) {
5562 return fast_cpu_time(thread);
5563 } else {
5564 return slow_thread_cpu_time(thread, true /* user + sys */);
5565 }
5566 }
5567
5568 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5569 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5570 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5571 } else {
5572 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5573 }
5574 }
5575
5576 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5577 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5578 return fast_cpu_time(thread);
5579 } else {
5580 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5581 }
5582 }
5583
5584 // -1 on error.
5585 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5586 pid_t tid = thread->osthread()->thread_id();
5587 char *s;
5588 char stat[2048];
5589 int statlen;
5590 char proc_name[64];
5591 int count;
5592 long sys_time, user_time;
5593 char cdummy;
5594 int idummy;
5595 long ldummy;
5596 FILE *fp;
5597
5598 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5599 fp = fopen(proc_name, "r");
5600 if (fp == NULL) return -1;
5601 statlen = fread(stat, 1, 2047, fp);
5602 stat[statlen] = '\0';
5603 fclose(fp);
5604
5605 // Skip pid and the command string. Note that we could be dealing with
5606 // weird command names, e.g. user could decide to rename java launcher
5607 // to "java 1.4.2 :)", then the stat file would look like
5608 // 1234 (java 1.4.2 :)) R ... ...
5609 // We don't really need to know the command string, just find the last
5610 // occurrence of ")" and then start parsing from there. See bug 4726580.
5611 s = strrchr(stat, ')');
5612 if (s == NULL) return -1;
5613
5614 // Skip blank chars
5615 do { s++; } while (s && isspace(*s));
5616
5617 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5618 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5619 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5620 &user_time, &sys_time);
5621 if (count != 13) return -1;
5622 if (user_sys_cpu_time) {
5623 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5624 } else {
5625 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5626 }
5627 }
5628
5629 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5630 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5631 info_ptr->may_skip_backward = false; // elapsed time not wall time
5632 info_ptr->may_skip_forward = false; // elapsed time not wall time
5633 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5634 }
5635
5636 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5637 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5638 info_ptr->may_skip_backward = false; // elapsed time not wall time
5639 info_ptr->may_skip_forward = false; // elapsed time not wall time
5640 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5641 }
5642
5643 bool os::is_thread_cpu_time_supported() {
5644 return true;
5645 }
5646
5647 // System loadavg support. Returns -1 if load average cannot be obtained.
5648 // Linux doesn't yet have a (official) notion of processor sets,
5649 // so just return the system wide load average.
5650 int os::loadavg(double loadavg[], int nelem) {
5651 return ::getloadavg(loadavg, nelem);
5652 }
5653
5654 void os::pause() {
5655 char filename[MAX_PATH];
5656 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5657 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5658 } else {
5659 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5660 }
5661
5662 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5663 if (fd != -1) {
5664 struct stat buf;
5665 ::close(fd);
5666 while (::stat(filename, &buf) == 0) {
5667 (void)::poll(NULL, 0, 100);
5668 }
5669 } else {
5670 jio_fprintf(stderr,
5671 "Could not open pause file '%s', continuing immediately.\n", filename);
5672 }
5673 }
5674
5675 extern char** environ;
5676
5677 // Run the specified command in a separate process. Return its exit value,
5678 // or -1 on failure (e.g. can't fork a new process).
5679 // Unlike system(), this function can be called from signal handler. It
5680 // doesn't block SIGINT et al.
5681 int os::fork_and_exec(char* cmd) {
5682 const char * argv[4] = {"sh", "-c", cmd, NULL};
5683
5684 pid_t pid = fork();
5685
5686 if (pid < 0) {
5687 // fork failed
5688 return -1;
5689
5690 } else if (pid == 0) {
5691 // child process
5692
5693 execve("/bin/sh", (char* const*)argv, environ);
5694
5695 // execve failed
5696 _exit(-1);
5697
5698 } else {
5699 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5700 // care about the actual exit code, for now.
5701
5702 int status;
5703
5704 // Wait for the child process to exit. This returns immediately if
5705 // the child has already exited. */
5706 while (waitpid(pid, &status, 0) < 0) {
5707 switch (errno) {
5708 case ECHILD: return 0;
5709 case EINTR: break;
5710 default: return -1;
5711 }
5712 }
5713
5714 if (WIFEXITED(status)) {
5715 // The child exited normally; get its exit code.
5716 return WEXITSTATUS(status);
5717 } else if (WIFSIGNALED(status)) {
5718 // The child exited because of a signal
5719 // The best value to return is 0x80 + signal number,
5720 // because that is what all Unix shells do, and because
5721 // it allows callers to distinguish between process exit and
5722 // process death by signal.
5723 return 0x80 + WTERMSIG(status);
5724 } else {
5725 // Unknown exit code; pass it through
5726 return status;
5727 }
5728 }
5729 }
5730
5731 // Get the default path to the core file
5732 // Returns the length of the string
5733 int os::get_core_path(char* buffer, size_t bufferSize) {
5734 /*
5735 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5736 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5737 */
5738 const int core_pattern_len = 129;
5739 char core_pattern[core_pattern_len] = {0};
5740
5741 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5742 if (core_pattern_file == -1) {
5743 return -1;
5744 }
5745
5746 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5747 ::close(core_pattern_file);
5748 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5749 return -1;
5750 }
5751 if (core_pattern[ret-1] == '\n') {
5752 core_pattern[ret-1] = '\0';
5753 } else {
5754 core_pattern[ret] = '\0';
5755 }
5756
5757 // Replace the %p in the core pattern with the process id. NOTE: we do this
5758 // only if the pattern doesn't start with "|", and we support only one %p in
5759 // the pattern.
5760 char *pid_pos = strstr(core_pattern, "%p");
5761 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5762 int written;
5763
5764 if (core_pattern[0] == '/') {
5765 if (pid_pos != NULL) {
5766 *pid_pos = '\0';
5767 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5768 current_process_id(), tail);
5769 } else {
5770 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5771 }
5772 } else {
5773 char cwd[PATH_MAX];
5774
5775 const char* p = get_current_directory(cwd, PATH_MAX);
5776 if (p == NULL) {
5777 return -1;
5778 }
5779
5780 if (core_pattern[0] == '|') {
5781 written = jio_snprintf(buffer, bufferSize,
5782 "\"%s\" (or dumping to %s/core.%d)",
5783 &core_pattern[1], p, current_process_id());
5784 } else if (pid_pos != NULL) {
5785 *pid_pos = '\0';
5786 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5787 current_process_id(), tail);
5788 } else {
5789 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5790 }
5791 }
5792
5793 if (written < 0) {
5794 return -1;
5795 }
5796
5797 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5798 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5799
5800 if (core_uses_pid_file != -1) {
5801 char core_uses_pid = 0;
5802 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5803 ::close(core_uses_pid_file);
5804
5805 if (core_uses_pid == '1') {
5806 jio_snprintf(buffer + written, bufferSize - written,
5807 ".%d", current_process_id());
5808 }
5809 }
5810 }
5811
5812 return strlen(buffer);
5813 }
5814
5815 bool os::start_debugging(char *buf, int buflen) {
5816 int len = (int)strlen(buf);
5817 char *p = &buf[len];
5818
5819 jio_snprintf(p, buflen-len,
5820 "\n\n"
5821 "Do you want to debug the problem?\n\n"
5822 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5823 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5824 "Otherwise, press RETURN to abort...",
5825 os::current_process_id(), os::current_process_id(),
5826 os::current_thread_id(), os::current_thread_id());
5827
5828 bool yes = os::message_box("Unexpected Error", buf);
5829
5830 if (yes) {
5831 // yes, user asked VM to launch debugger
5832 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5833 os::current_process_id(), os::current_process_id());
5834
5835 os::fork_and_exec(buf);
5836 yes = false;
5837 }
5838 return yes;
5839 }
5840
5841
5842 // Java/Compiler thread:
5843 //
5844 // Low memory addresses
5845 // P0 +------------------------+
5846 // | |\ Java thread created by VM does not have glibc
5847 // | glibc guard page | - guard page, attached Java thread usually has
5848 // | |/ 1 glibc guard page.
5849 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5850 // | |\
5851 // | HotSpot Guard Pages | - red, yellow and reserved pages
5852 // | |/
5853 // +------------------------+ JavaThread::stack_reserved_zone_base()
5854 // | |\
5855 // | Normal Stack | -
5856 // | |/
5857 // P2 +------------------------+ Thread::stack_base()
5858 //
5859 // Non-Java thread:
5860 //
5861 // Low memory addresses
5862 // P0 +------------------------+
5863 // | |\
5864 // | glibc guard page | - usually 1 page
5865 // | |/
5866 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5867 // | |\
5868 // | Normal Stack | -
5869 // | |/
5870 // P2 +------------------------+ Thread::stack_base()
5871 //
5872 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5873 // returned from pthread_attr_getstack().
5874 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5875 // of the stack size given in pthread_attr. We work around this for
5876 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5877 //
5878 #ifndef ZERO
5879 static void current_stack_region(address * bottom, size_t * size) {
5880 if (os::is_primordial_thread()) {
5881 // primordial thread needs special handling because pthread_getattr_np()
5882 // may return bogus value.
5883 *bottom = os::Linux::initial_thread_stack_bottom();
5884 *size = os::Linux::initial_thread_stack_size();
5885 } else {
5886 pthread_attr_t attr;
5887
5888 int rslt = pthread_getattr_np(pthread_self(), &attr);
5889
5890 // JVM needs to know exact stack location, abort if it fails
5891 if (rslt != 0) {
5892 if (rslt == ENOMEM) {
5893 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5894 } else {
5895 fatal("pthread_getattr_np failed with error = %d", rslt);
5896 }
5897 }
5898
5899 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5900 fatal("Cannot locate current stack attributes!");
5901 }
5902
5903 // Work around NPTL stack guard error.
5904 size_t guard_size = 0;
5905 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5906 if (rslt != 0) {
5907 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5908 }
5909 *bottom += guard_size;
5910 *size -= guard_size;
5911
5912 pthread_attr_destroy(&attr);
5913
5914 }
5915 assert(os::current_stack_pointer() >= *bottom &&
5916 os::current_stack_pointer() < *bottom + *size, "just checking");
5917 }
5918
5919 address os::current_stack_base() {
5920 address bottom;
5921 size_t size;
5922 current_stack_region(&bottom, &size);
5923 return (bottom + size);
5924 }
5925
5926 size_t os::current_stack_size() {
5927 // This stack size includes the usable stack and HotSpot guard pages
5928 // (for the threads that have Hotspot guard pages).
5929 address bottom;
5930 size_t size;
5931 current_stack_region(&bottom, &size);
5932 return size;
5933 }
5934 #endif
5935
5936 static inline struct timespec get_mtime(const char* filename) {
5937 struct stat st;
5938 int ret = os::stat(filename, &st);
5939 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5940 return st.st_mtim;
5941 }
5942
5943 int os::compare_file_modified_times(const char* file1, const char* file2) {
5944 struct timespec filetime1 = get_mtime(file1);
5945 struct timespec filetime2 = get_mtime(file2);
5946 int diff = filetime1.tv_sec - filetime2.tv_sec;
5947 if (diff == 0) {
5948 return filetime1.tv_nsec - filetime2.tv_nsec;
5949 }
5950 return diff;
5951 }
5952
5953 /////////////// Unit tests ///////////////
5954
5955 #ifndef PRODUCT
5956
5957 #define test_log(...) \
5958 do { \
5959 if (VerboseInternalVMTests) { \
5960 tty->print_cr(__VA_ARGS__); \
5961 tty->flush(); \
5962 } \
5963 } while (false)
5964
5965 class TestReserveMemorySpecial : AllStatic {
5966 public:
5967 static void small_page_write(void* addr, size_t size) {
5968 size_t page_size = os::vm_page_size();
5969
5970 char* end = (char*)addr + size;
5971 for (char* p = (char*)addr; p < end; p += page_size) {
5972 *p = 1;
5973 }
5974 }
5975
5976 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5977 if (!UseHugeTLBFS) {
5978 return;
5979 }
5980
5981 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5982
5983 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5984
5985 if (addr != NULL) {
5986 small_page_write(addr, size);
5987
5988 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5989 }
5990 }
5991
5992 static void test_reserve_memory_special_huge_tlbfs_only() {
5993 if (!UseHugeTLBFS) {
5994 return;
5995 }
5996
5997 size_t lp = os::large_page_size();
5998
5999 for (size_t size = lp; size <= lp * 10; size += lp) {
6000 test_reserve_memory_special_huge_tlbfs_only(size);
6001 }
6002 }
6003
6004 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6005 size_t lp = os::large_page_size();
6006 size_t ag = os::vm_allocation_granularity();
6007
6008 // sizes to test
6009 const size_t sizes[] = {
6010 lp, lp + ag, lp + lp / 2, lp * 2,
6011 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6012 lp * 10, lp * 10 + lp / 2
6013 };
6014 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6015
6016 // For each size/alignment combination, we test three scenarios:
6017 // 1) with req_addr == NULL
6018 // 2) with a non-null req_addr at which we expect to successfully allocate
6019 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6020 // expect the allocation to either fail or to ignore req_addr
6021
6022 // Pre-allocate two areas; they shall be as large as the largest allocation
6023 // and aligned to the largest alignment we will be testing.
6024 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6025 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6026 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6027 -1, 0);
6028 assert(mapping1 != MAP_FAILED, "should work");
6029
6030 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6031 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6032 -1, 0);
6033 assert(mapping2 != MAP_FAILED, "should work");
6034
6035 // Unmap the first mapping, but leave the second mapping intact: the first
6036 // mapping will serve as a value for a "good" req_addr (case 2). The second
6037 // mapping, still intact, as "bad" req_addr (case 3).
6038 ::munmap(mapping1, mapping_size);
6039
6040 // Case 1
6041 test_log("%s, req_addr NULL:", __FUNCTION__);
6042 test_log("size align result");
6043
6044 for (int i = 0; i < num_sizes; i++) {
6045 const size_t size = sizes[i];
6046 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6047 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6048 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6049 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6050 if (p != NULL) {
6051 assert(is_aligned(p, alignment), "must be");
6052 small_page_write(p, size);
6053 os::Linux::release_memory_special_huge_tlbfs(p, size);
6054 }
6055 }
6056 }
6057
6058 // Case 2
6059 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6060 test_log("size align req_addr result");
6061
6062 for (int i = 0; i < num_sizes; i++) {
6063 const size_t size = sizes[i];
6064 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6065 char* const req_addr = align_up(mapping1, alignment);
6066 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6067 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6068 size, alignment, p2i(req_addr), p2i(p),
6069 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6070 if (p != NULL) {
6071 assert(p == req_addr, "must be");
6072 small_page_write(p, size);
6073 os::Linux::release_memory_special_huge_tlbfs(p, size);
6074 }
6075 }
6076 }
6077
6078 // Case 3
6079 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6080 test_log("size align req_addr result");
6081
6082 for (int i = 0; i < num_sizes; i++) {
6083 const size_t size = sizes[i];
6084 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6085 char* const req_addr = align_up(mapping2, alignment);
6086 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6087 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6088 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6089 // as the area around req_addr contains already existing mappings, the API should always
6090 // return NULL (as per contract, it cannot return another address)
6091 assert(p == NULL, "must be");
6092 }
6093 }
6094
6095 ::munmap(mapping2, mapping_size);
6096
6097 }
6098
6099 static void test_reserve_memory_special_huge_tlbfs() {
6100 if (!UseHugeTLBFS) {
6101 return;
6102 }
6103
6104 test_reserve_memory_special_huge_tlbfs_only();
6105 test_reserve_memory_special_huge_tlbfs_mixed();
6106 }
6107
6108 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6109 if (!UseSHM) {
6110 return;
6111 }
6112
6113 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6114
6115 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6116
6117 if (addr != NULL) {
6118 assert(is_aligned(addr, alignment), "Check");
6119 assert(is_aligned(addr, os::large_page_size()), "Check");
6120
6121 small_page_write(addr, size);
6122
6123 os::Linux::release_memory_special_shm(addr, size);
6124 }
6125 }
6126
6127 static void test_reserve_memory_special_shm() {
6128 size_t lp = os::large_page_size();
6129 size_t ag = os::vm_allocation_granularity();
6130
6131 for (size_t size = ag; size < lp * 3; size += ag) {
6132 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6133 test_reserve_memory_special_shm(size, alignment);
6134 }
6135 }
6136 }
6137
6138 static void test() {
6139 test_reserve_memory_special_huge_tlbfs();
6140 test_reserve_memory_special_shm();
6141 }
6142 };
6143
6144 void TestReserveMemorySpecial_test() {
6145 TestReserveMemorySpecial::test();
6146 }
6147
6148 #endif