1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.inline.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/threadSMR.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/align.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/vmError.hpp"
76
77 // put OS-includes here
78 # include <sys/types.h>
79 # include <sys/mman.h>
80 # include <sys/stat.h>
81 # include <sys/select.h>
82 # include <pthread.h>
83 # include <signal.h>
84 # include <errno.h>
85 # include <dlfcn.h>
86 # include <stdio.h>
87 # include <unistd.h>
88 # include <sys/resource.h>
89 # include <pthread.h>
90 # include <sys/stat.h>
91 # include <sys/time.h>
92 # include <sys/times.h>
93 # include <sys/utsname.h>
94 # include <sys/socket.h>
95 # include <sys/wait.h>
96 # include <pwd.h>
97 # include <poll.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 enum CoredumpFilterBit {
132 FILE_BACKED_PVT_BIT = 1 << 2,
133 LARGEPAGES_BIT = 1 << 6,
134 DAX_SHARED_BIT = 1 << 8
135 };
136
137 ////////////////////////////////////////////////////////////////////////////////
138 // global variables
139 julong os::Linux::_physical_memory = 0;
140
141 address os::Linux::_initial_thread_stack_bottom = NULL;
142 uintptr_t os::Linux::_initial_thread_stack_size = 0;
143
144 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
145 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
146 Mutex* os::Linux::_createThread_lock = NULL;
147 pthread_t os::Linux::_main_thread;
148 int os::Linux::_page_size = -1;
149 bool os::Linux::_supports_fast_thread_cpu_time = false;
150 uint32_t os::Linux::_os_version = 0;
151 const char * os::Linux::_glibc_version = NULL;
152 const char * os::Linux::_libpthread_version = NULL;
153
154 static jlong initial_time_count=0;
155
156 static int clock_tics_per_sec = 100;
157
158 // If the VM might have been created on the primordial thread, we need to resolve the
159 // primordial thread stack bounds and check if the current thread might be the
160 // primordial thread in places. If we know that the primordial thread is never used,
161 // such as when the VM was created by one of the standard java launchers, we can
162 // avoid this
163 static bool suppress_primordial_thread_resolution = false;
164
165 // For diagnostics to print a message once. see run_periodic_checks
166 static sigset_t check_signal_done;
167 static bool check_signals = true;
168
169 // Signal number used to suspend/resume a thread
170
171 // do not use any signal number less than SIGSEGV, see 4355769
172 static int SR_signum = SIGUSR2;
173 sigset_t SR_sigset;
174
175 // utility functions
176
177 static int SR_initialize();
178
179 julong os::available_memory() {
180 return Linux::available_memory();
181 }
182
183 julong os::Linux::available_memory() {
184 // values in struct sysinfo are "unsigned long"
185 struct sysinfo si;
186 julong avail_mem;
187
188 if (OSContainer::is_containerized()) {
189 jlong mem_limit, mem_usage;
190 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
191 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
192 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
193 }
194 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
195 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
196 }
197 if (mem_limit > 0 && mem_usage > 0 ) {
198 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
199 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
200 return avail_mem;
201 }
202 }
203
204 sysinfo(&si);
205 avail_mem = (julong)si.freeram * si.mem_unit;
206 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
207 return avail_mem;
208 }
209
210 julong os::physical_memory() {
211 jlong phys_mem = 0;
212 if (OSContainer::is_containerized()) {
213 jlong mem_limit;
214 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
215 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
216 return mem_limit;
217 }
218 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
219 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
220 }
221
222 phys_mem = Linux::physical_memory();
223 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
224 return phys_mem;
225 }
226
227 // Return true if user is running as root.
228
229 bool os::have_special_privileges() {
230 static bool init = false;
231 static bool privileges = false;
232 if (!init) {
233 privileges = (getuid() != geteuid()) || (getgid() != getegid());
234 init = true;
235 }
236 return privileges;
237 }
238
239
240 #ifndef SYS_gettid
241 // i386: 224, ia64: 1105, amd64: 186, sparc 143
242 #ifdef __ia64__
243 #define SYS_gettid 1105
244 #else
245 #ifdef __i386__
246 #define SYS_gettid 224
247 #else
248 #ifdef __amd64__
249 #define SYS_gettid 186
250 #else
251 #ifdef __sparc__
252 #define SYS_gettid 143
253 #else
254 #error define gettid for the arch
255 #endif
256 #endif
257 #endif
258 #endif
259 #endif
260
261
262 // pid_t gettid()
263 //
264 // Returns the kernel thread id of the currently running thread. Kernel
265 // thread id is used to access /proc.
266 pid_t os::Linux::gettid() {
267 int rslt = syscall(SYS_gettid);
268 assert(rslt != -1, "must be."); // old linuxthreads implementation?
269 return (pid_t)rslt;
270 }
271
272 // Most versions of linux have a bug where the number of processors are
273 // determined by looking at the /proc file system. In a chroot environment,
274 // the system call returns 1.
275 static bool unsafe_chroot_detected = false;
276 static const char *unstable_chroot_error = "/proc file system not found.\n"
277 "Java may be unstable running multithreaded in a chroot "
278 "environment on Linux when /proc filesystem is not mounted.";
279
280 void os::Linux::initialize_system_info() {
281 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
282 if (processor_count() == 1) {
283 pid_t pid = os::Linux::gettid();
284 char fname[32];
285 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
286 FILE *fp = fopen(fname, "r");
287 if (fp == NULL) {
288 unsafe_chroot_detected = true;
289 } else {
290 fclose(fp);
291 }
292 }
293 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
294 assert(processor_count() > 0, "linux error");
295 }
296
297 void os::init_system_properties_values() {
298 // The next steps are taken in the product version:
299 //
300 // Obtain the JAVA_HOME value from the location of libjvm.so.
301 // This library should be located at:
302 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
303 //
304 // If "/jre/lib/" appears at the right place in the path, then we
305 // assume libjvm.so is installed in a JDK and we use this path.
306 //
307 // Otherwise exit with message: "Could not create the Java virtual machine."
308 //
309 // The following extra steps are taken in the debugging version:
310 //
311 // If "/jre/lib/" does NOT appear at the right place in the path
312 // instead of exit check for $JAVA_HOME environment variable.
313 //
314 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
315 // then we append a fake suffix "hotspot/libjvm.so" to this path so
316 // it looks like libjvm.so is installed there
317 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
318 //
319 // Otherwise exit.
320 //
321 // Important note: if the location of libjvm.so changes this
322 // code needs to be changed accordingly.
323
324 // See ld(1):
325 // The linker uses the following search paths to locate required
326 // shared libraries:
327 // 1: ...
328 // ...
329 // 7: The default directories, normally /lib and /usr/lib.
330 #ifndef OVERRIDE_LIBPATH
331 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
332 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
333 #else
334 #define DEFAULT_LIBPATH "/lib:/usr/lib"
335 #endif
336 #else
337 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
338 #endif
339
340 // Base path of extensions installed on the system.
341 #define SYS_EXT_DIR "/usr/java/packages"
342 #define EXTENSIONS_DIR "/lib/ext"
343
344 // Buffer that fits several sprintfs.
345 // Note that the space for the colon and the trailing null are provided
346 // by the nulls included by the sizeof operator.
347 const size_t bufsize =
348 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
349 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
350 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
351
352 // sysclasspath, java_home, dll_dir
353 {
354 char *pslash;
355 os::jvm_path(buf, bufsize);
356
357 // Found the full path to libjvm.so.
358 // Now cut the path to <java_home>/jre if we can.
359 pslash = strrchr(buf, '/');
360 if (pslash != NULL) {
361 *pslash = '\0'; // Get rid of /libjvm.so.
362 }
363 pslash = strrchr(buf, '/');
364 if (pslash != NULL) {
365 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
366 }
367 Arguments::set_dll_dir(buf);
368
369 if (pslash != NULL) {
370 pslash = strrchr(buf, '/');
371 if (pslash != NULL) {
372 *pslash = '\0'; // Get rid of /lib.
373 }
374 }
375 Arguments::set_java_home(buf);
376 if (!set_boot_path('/', ':')) {
377 vm_exit_during_initialization("Failed setting boot class path.", NULL);
378 }
379 }
380
381 // Where to look for native libraries.
382 //
383 // Note: Due to a legacy implementation, most of the library path
384 // is set in the launcher. This was to accomodate linking restrictions
385 // on legacy Linux implementations (which are no longer supported).
386 // Eventually, all the library path setting will be done here.
387 //
388 // However, to prevent the proliferation of improperly built native
389 // libraries, the new path component /usr/java/packages is added here.
390 // Eventually, all the library path setting will be done here.
391 {
392 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
393 // should always exist (until the legacy problem cited above is
394 // addressed).
395 const char *v = ::getenv("LD_LIBRARY_PATH");
396 const char *v_colon = ":";
397 if (v == NULL) { v = ""; v_colon = ""; }
398 // That's +1 for the colon and +1 for the trailing '\0'.
399 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
400 strlen(v) + 1 +
401 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
402 mtInternal);
403 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
404 Arguments::set_library_path(ld_library_path);
405 FREE_C_HEAP_ARRAY(char, ld_library_path);
406 }
407
408 // Extensions directories.
409 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
410 Arguments::set_ext_dirs(buf);
411
412 FREE_C_HEAP_ARRAY(char, buf);
413
414 #undef DEFAULT_LIBPATH
415 #undef SYS_EXT_DIR
416 #undef EXTENSIONS_DIR
417 }
418
419 ////////////////////////////////////////////////////////////////////////////////
420 // breakpoint support
421
422 void os::breakpoint() {
423 BREAKPOINT;
424 }
425
426 extern "C" void breakpoint() {
427 // use debugger to set breakpoint here
428 }
429
430 ////////////////////////////////////////////////////////////////////////////////
431 // signal support
432
433 debug_only(static bool signal_sets_initialized = false);
434 static sigset_t unblocked_sigs, vm_sigs;
435
436 void os::Linux::signal_sets_init() {
437 // Should also have an assertion stating we are still single-threaded.
438 assert(!signal_sets_initialized, "Already initialized");
439 // Fill in signals that are necessarily unblocked for all threads in
440 // the VM. Currently, we unblock the following signals:
441 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
442 // by -Xrs (=ReduceSignalUsage));
443 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
444 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
445 // the dispositions or masks wrt these signals.
446 // Programs embedding the VM that want to use the above signals for their
447 // own purposes must, at this time, use the "-Xrs" option to prevent
448 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
449 // (See bug 4345157, and other related bugs).
450 // In reality, though, unblocking these signals is really a nop, since
451 // these signals are not blocked by default.
452 sigemptyset(&unblocked_sigs);
453 sigaddset(&unblocked_sigs, SIGILL);
454 sigaddset(&unblocked_sigs, SIGSEGV);
455 sigaddset(&unblocked_sigs, SIGBUS);
456 sigaddset(&unblocked_sigs, SIGFPE);
457 #if defined(PPC64)
458 sigaddset(&unblocked_sigs, SIGTRAP);
459 #endif
460 sigaddset(&unblocked_sigs, SR_signum);
461
462 if (!ReduceSignalUsage) {
463 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
464 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
465 }
466 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
468 }
469 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
470 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
471 }
472 }
473 // Fill in signals that are blocked by all but the VM thread.
474 sigemptyset(&vm_sigs);
475 if (!ReduceSignalUsage) {
476 sigaddset(&vm_sigs, BREAK_SIGNAL);
477 }
478 debug_only(signal_sets_initialized = true);
479
480 }
481
482 // These are signals that are unblocked while a thread is running Java.
483 // (For some reason, they get blocked by default.)
484 sigset_t* os::Linux::unblocked_signals() {
485 assert(signal_sets_initialized, "Not initialized");
486 return &unblocked_sigs;
487 }
488
489 // These are the signals that are blocked while a (non-VM) thread is
490 // running Java. Only the VM thread handles these signals.
491 sigset_t* os::Linux::vm_signals() {
492 assert(signal_sets_initialized, "Not initialized");
493 return &vm_sigs;
494 }
495
496 void os::Linux::hotspot_sigmask(Thread* thread) {
497
498 //Save caller's signal mask before setting VM signal mask
499 sigset_t caller_sigmask;
500 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
501
502 OSThread* osthread = thread->osthread();
503 osthread->set_caller_sigmask(caller_sigmask);
504
505 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
506
507 if (!ReduceSignalUsage) {
508 if (thread->is_VM_thread()) {
509 // Only the VM thread handles BREAK_SIGNAL ...
510 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
511 } else {
512 // ... all other threads block BREAK_SIGNAL
513 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
514 }
515 }
516 }
517
518 //////////////////////////////////////////////////////////////////////////////
519 // detecting pthread library
520
521 void os::Linux::libpthread_init() {
522 // Save glibc and pthread version strings.
523 #if !defined(_CS_GNU_LIBC_VERSION) || \
524 !defined(_CS_GNU_LIBPTHREAD_VERSION)
525 #error "glibc too old (< 2.3.2)"
526 #endif
527
528 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
529 assert(n > 0, "cannot retrieve glibc version");
530 char *str = (char *)malloc(n, mtInternal);
531 confstr(_CS_GNU_LIBC_VERSION, str, n);
532 os::Linux::set_glibc_version(str);
533
534 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
535 assert(n > 0, "cannot retrieve pthread version");
536 str = (char *)malloc(n, mtInternal);
537 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
538 os::Linux::set_libpthread_version(str);
539 }
540
541 /////////////////////////////////////////////////////////////////////////////
542 // thread stack expansion
543
544 // os::Linux::manually_expand_stack() takes care of expanding the thread
545 // stack. Note that this is normally not needed: pthread stacks allocate
546 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
547 // committed. Therefore it is not necessary to expand the stack manually.
548 //
549 // Manually expanding the stack was historically needed on LinuxThreads
550 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
551 // it is kept to deal with very rare corner cases:
552 //
553 // For one, user may run the VM on an own implementation of threads
554 // whose stacks are - like the old LinuxThreads - implemented using
555 // mmap(MAP_GROWSDOWN).
556 //
557 // Also, this coding may be needed if the VM is running on the primordial
558 // thread. Normally we avoid running on the primordial thread; however,
559 // user may still invoke the VM on the primordial thread.
560 //
561 // The following historical comment describes the details about running
562 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
563
564
565 // Force Linux kernel to expand current thread stack. If "bottom" is close
566 // to the stack guard, caller should block all signals.
567 //
568 // MAP_GROWSDOWN:
569 // A special mmap() flag that is used to implement thread stacks. It tells
570 // kernel that the memory region should extend downwards when needed. This
571 // allows early versions of LinuxThreads to only mmap the first few pages
572 // when creating a new thread. Linux kernel will automatically expand thread
573 // stack as needed (on page faults).
574 //
575 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
576 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
577 // region, it's hard to tell if the fault is due to a legitimate stack
578 // access or because of reading/writing non-exist memory (e.g. buffer
579 // overrun). As a rule, if the fault happens below current stack pointer,
580 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
581 // application (see Linux kernel fault.c).
582 //
583 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
584 // stack overflow detection.
585 //
586 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
587 // not use MAP_GROWSDOWN.
588 //
589 // To get around the problem and allow stack banging on Linux, we need to
590 // manually expand thread stack after receiving the SIGSEGV.
591 //
592 // There are two ways to expand thread stack to address "bottom", we used
593 // both of them in JVM before 1.5:
594 // 1. adjust stack pointer first so that it is below "bottom", and then
595 // touch "bottom"
596 // 2. mmap() the page in question
597 //
598 // Now alternate signal stack is gone, it's harder to use 2. For instance,
599 // if current sp is already near the lower end of page 101, and we need to
600 // call mmap() to map page 100, it is possible that part of the mmap() frame
601 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
602 // That will destroy the mmap() frame and cause VM to crash.
603 //
604 // The following code works by adjusting sp first, then accessing the "bottom"
605 // page to force a page fault. Linux kernel will then automatically expand the
606 // stack mapping.
607 //
608 // _expand_stack_to() assumes its frame size is less than page size, which
609 // should always be true if the function is not inlined.
610
611 static void NOINLINE _expand_stack_to(address bottom) {
612 address sp;
613 size_t size;
614 volatile char *p;
615
616 // Adjust bottom to point to the largest address within the same page, it
617 // gives us a one-page buffer if alloca() allocates slightly more memory.
618 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
619 bottom += os::Linux::page_size() - 1;
620
621 // sp might be slightly above current stack pointer; if that's the case, we
622 // will alloca() a little more space than necessary, which is OK. Don't use
623 // os::current_stack_pointer(), as its result can be slightly below current
624 // stack pointer, causing us to not alloca enough to reach "bottom".
625 sp = (address)&sp;
626
627 if (sp > bottom) {
628 size = sp - bottom;
629 p = (volatile char *)alloca(size);
630 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
631 p[0] = '\0';
632 }
633 }
634
635 void os::Linux::expand_stack_to(address bottom) {
636 _expand_stack_to(bottom);
637 }
638
639 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
640 assert(t!=NULL, "just checking");
641 assert(t->osthread()->expanding_stack(), "expand should be set");
642 assert(t->stack_base() != NULL, "stack_base was not initialized");
643
644 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
645 sigset_t mask_all, old_sigset;
646 sigfillset(&mask_all);
647 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
648 _expand_stack_to(addr);
649 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
650 return true;
651 }
652 return false;
653 }
654
655 //////////////////////////////////////////////////////////////////////////////
656 // create new thread
657
658 // Thread start routine for all newly created threads
659 static void *thread_native_entry(Thread *thread) {
660
661 thread->record_stack_base_and_size();
662
663 // Try to randomize the cache line index of hot stack frames.
664 // This helps when threads of the same stack traces evict each other's
665 // cache lines. The threads can be either from the same JVM instance, or
666 // from different JVM instances. The benefit is especially true for
667 // processors with hyperthreading technology.
668 static int counter = 0;
669 int pid = os::current_process_id();
670 alloca(((pid ^ counter++) & 7) * 128);
671
672 thread->initialize_thread_current();
673
674 OSThread* osthread = thread->osthread();
675 Monitor* sync = osthread->startThread_lock();
676
677 osthread->set_thread_id(os::current_thread_id());
678
679 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
680 os::current_thread_id(), (uintx) pthread_self());
681
682 if (UseNUMA) {
683 int lgrp_id = os::numa_get_group_id();
684 if (lgrp_id != -1) {
685 thread->set_lgrp_id(lgrp_id);
686 }
687 }
688 // initialize signal mask for this thread
689 os::Linux::hotspot_sigmask(thread);
690
691 // initialize floating point control register
692 os::Linux::init_thread_fpu_state();
693
694 // handshaking with parent thread
695 {
696 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
697
698 // notify parent thread
699 osthread->set_state(INITIALIZED);
700 sync->notify_all();
701
702 // wait until os::start_thread()
703 while (osthread->get_state() == INITIALIZED) {
704 sync->wait(Mutex::_no_safepoint_check_flag);
705 }
706 }
707
708 // call one more level start routine
709 thread->call_run();
710
711 // Note: at this point the thread object may already have deleted itself.
712 // Prevent dereferencing it from here on out.
713 thread = NULL;
714
715 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
716 os::current_thread_id(), (uintx) pthread_self());
717
718 return 0;
719 }
720
721 bool os::create_thread(Thread* thread, ThreadType thr_type,
722 size_t req_stack_size) {
723 assert(thread->osthread() == NULL, "caller responsible");
724
725 // Allocate the OSThread object
726 OSThread* osthread = new OSThread(NULL, NULL);
727 if (osthread == NULL) {
728 return false;
729 }
730
731 // set the correct thread state
732 osthread->set_thread_type(thr_type);
733
734 // Initial state is ALLOCATED but not INITIALIZED
735 osthread->set_state(ALLOCATED);
736
737 thread->set_osthread(osthread);
738
739 // init thread attributes
740 pthread_attr_t attr;
741 pthread_attr_init(&attr);
742 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
743
744 // Calculate stack size if it's not specified by caller.
745 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
746 // In the Linux NPTL pthread implementation the guard size mechanism
747 // is not implemented properly. The posix standard requires adding
748 // the size of the guard pages to the stack size, instead Linux
749 // takes the space out of 'stacksize'. Thus we adapt the requested
750 // stack_size by the size of the guard pages to mimick proper
751 // behaviour. However, be careful not to end up with a size
752 // of zero due to overflow. Don't add the guard page in that case.
753 size_t guard_size = os::Linux::default_guard_size(thr_type);
754 if (stack_size <= SIZE_MAX - guard_size) {
755 stack_size += guard_size;
756 }
757 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
758
759 int status = pthread_attr_setstacksize(&attr, stack_size);
760 assert_status(status == 0, status, "pthread_attr_setstacksize");
761
762 // Configure glibc guard page.
763 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
764
765 ThreadState state;
766
767 {
768 pthread_t tid;
769 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
770
771 char buf[64];
772 if (ret == 0) {
773 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
774 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
775 } else {
776 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
777 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
778 }
779
780 pthread_attr_destroy(&attr);
781
782 if (ret != 0) {
783 // Need to clean up stuff we've allocated so far
784 thread->set_osthread(NULL);
785 delete osthread;
786 return false;
787 }
788
789 // Store pthread info into the OSThread
790 osthread->set_pthread_id(tid);
791
792 // Wait until child thread is either initialized or aborted
793 {
794 Monitor* sync_with_child = osthread->startThread_lock();
795 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
796 while ((state = osthread->get_state()) == ALLOCATED) {
797 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
798 }
799 }
800 }
801
802 // Aborted due to thread limit being reached
803 if (state == ZOMBIE) {
804 thread->set_osthread(NULL);
805 delete osthread;
806 return false;
807 }
808
809 // The thread is returned suspended (in state INITIALIZED),
810 // and is started higher up in the call chain
811 assert(state == INITIALIZED, "race condition");
812 return true;
813 }
814
815 /////////////////////////////////////////////////////////////////////////////
816 // attach existing thread
817
818 // bootstrap the main thread
819 bool os::create_main_thread(JavaThread* thread) {
820 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
821 return create_attached_thread(thread);
822 }
823
824 bool os::create_attached_thread(JavaThread* thread) {
825 #ifdef ASSERT
826 thread->verify_not_published();
827 #endif
828
829 // Allocate the OSThread object
830 OSThread* osthread = new OSThread(NULL, NULL);
831
832 if (osthread == NULL) {
833 return false;
834 }
835
836 // Store pthread info into the OSThread
837 osthread->set_thread_id(os::Linux::gettid());
838 osthread->set_pthread_id(::pthread_self());
839
840 // initialize floating point control register
841 os::Linux::init_thread_fpu_state();
842
843 // Initial thread state is RUNNABLE
844 osthread->set_state(RUNNABLE);
845
846 thread->set_osthread(osthread);
847
848 if (UseNUMA) {
849 int lgrp_id = os::numa_get_group_id();
850 if (lgrp_id != -1) {
851 thread->set_lgrp_id(lgrp_id);
852 }
853 }
854
855 if (os::is_primordial_thread()) {
856 // If current thread is primordial thread, its stack is mapped on demand,
857 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
858 // the entire stack region to avoid SEGV in stack banging.
859 // It is also useful to get around the heap-stack-gap problem on SuSE
860 // kernel (see 4821821 for details). We first expand stack to the top
861 // of yellow zone, then enable stack yellow zone (order is significant,
862 // enabling yellow zone first will crash JVM on SuSE Linux), so there
863 // is no gap between the last two virtual memory regions.
864
865 JavaThread *jt = (JavaThread *)thread;
866 address addr = jt->stack_reserved_zone_base();
867 assert(addr != NULL, "initialization problem?");
868 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
869
870 osthread->set_expanding_stack();
871 os::Linux::manually_expand_stack(jt, addr);
872 osthread->clear_expanding_stack();
873 }
874
875 // initialize signal mask for this thread
876 // and save the caller's signal mask
877 os::Linux::hotspot_sigmask(thread);
878
879 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
880 os::current_thread_id(), (uintx) pthread_self());
881
882 return true;
883 }
884
885 void os::pd_start_thread(Thread* thread) {
886 OSThread * osthread = thread->osthread();
887 assert(osthread->get_state() != INITIALIZED, "just checking");
888 Monitor* sync_with_child = osthread->startThread_lock();
889 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
890 sync_with_child->notify();
891 }
892
893 // Free Linux resources related to the OSThread
894 void os::free_thread(OSThread* osthread) {
895 assert(osthread != NULL, "osthread not set");
896
897 // We are told to free resources of the argument thread,
898 // but we can only really operate on the current thread.
899 assert(Thread::current()->osthread() == osthread,
900 "os::free_thread but not current thread");
901
902 #ifdef ASSERT
903 sigset_t current;
904 sigemptyset(¤t);
905 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
906 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
907 #endif
908
909 // Restore caller's signal mask
910 sigset_t sigmask = osthread->caller_sigmask();
911 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
912
913 delete osthread;
914 }
915
916 //////////////////////////////////////////////////////////////////////////////
917 // primordial thread
918
919 // Check if current thread is the primordial thread, similar to Solaris thr_main.
920 bool os::is_primordial_thread(void) {
921 if (suppress_primordial_thread_resolution) {
922 return false;
923 }
924 char dummy;
925 // If called before init complete, thread stack bottom will be null.
926 // Can be called if fatal error occurs before initialization.
927 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
928 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
929 os::Linux::initial_thread_stack_size() != 0,
930 "os::init did not locate primordial thread's stack region");
931 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
932 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
933 os::Linux::initial_thread_stack_size()) {
934 return true;
935 } else {
936 return false;
937 }
938 }
939
940 // Find the virtual memory area that contains addr
941 static bool find_vma(address addr, address* vma_low, address* vma_high) {
942 FILE *fp = fopen("/proc/self/maps", "r");
943 if (fp) {
944 address low, high;
945 while (!feof(fp)) {
946 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
947 if (low <= addr && addr < high) {
948 if (vma_low) *vma_low = low;
949 if (vma_high) *vma_high = high;
950 fclose(fp);
951 return true;
952 }
953 }
954 for (;;) {
955 int ch = fgetc(fp);
956 if (ch == EOF || ch == (int)'\n') break;
957 }
958 }
959 fclose(fp);
960 }
961 return false;
962 }
963
964 // Locate primordial thread stack. This special handling of primordial thread stack
965 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
966 // bogus value for the primordial process thread. While the launcher has created
967 // the VM in a new thread since JDK 6, we still have to allow for the use of the
968 // JNI invocation API from a primordial thread.
969 void os::Linux::capture_initial_stack(size_t max_size) {
970
971 // max_size is either 0 (which means accept OS default for thread stacks) or
972 // a user-specified value known to be at least the minimum needed. If we
973 // are actually on the primordial thread we can make it appear that we have a
974 // smaller max_size stack by inserting the guard pages at that location. But we
975 // cannot do anything to emulate a larger stack than what has been provided by
976 // the OS or threading library. In fact if we try to use a stack greater than
977 // what is set by rlimit then we will crash the hosting process.
978
979 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
980 // If this is "unlimited" then it will be a huge value.
981 struct rlimit rlim;
982 getrlimit(RLIMIT_STACK, &rlim);
983 size_t stack_size = rlim.rlim_cur;
984
985 // 6308388: a bug in ld.so will relocate its own .data section to the
986 // lower end of primordial stack; reduce ulimit -s value a little bit
987 // so we won't install guard page on ld.so's data section.
988 // But ensure we don't underflow the stack size - allow 1 page spare
989 if (stack_size >= (size_t)(3 * page_size())) {
990 stack_size -= 2 * page_size();
991 }
992
993 // Try to figure out where the stack base (top) is. This is harder.
994 //
995 // When an application is started, glibc saves the initial stack pointer in
996 // a global variable "__libc_stack_end", which is then used by system
997 // libraries. __libc_stack_end should be pretty close to stack top. The
998 // variable is available since the very early days. However, because it is
999 // a private interface, it could disappear in the future.
1000 //
1001 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1002 // to __libc_stack_end, it is very close to stack top, but isn't the real
1003 // stack top. Note that /proc may not exist if VM is running as a chroot
1004 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1005 // /proc/<pid>/stat could change in the future (though unlikely).
1006 //
1007 // We try __libc_stack_end first. If that doesn't work, look for
1008 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1009 // as a hint, which should work well in most cases.
1010
1011 uintptr_t stack_start;
1012
1013 // try __libc_stack_end first
1014 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1015 if (p && *p) {
1016 stack_start = *p;
1017 } else {
1018 // see if we can get the start_stack field from /proc/self/stat
1019 FILE *fp;
1020 int pid;
1021 char state;
1022 int ppid;
1023 int pgrp;
1024 int session;
1025 int nr;
1026 int tpgrp;
1027 unsigned long flags;
1028 unsigned long minflt;
1029 unsigned long cminflt;
1030 unsigned long majflt;
1031 unsigned long cmajflt;
1032 unsigned long utime;
1033 unsigned long stime;
1034 long cutime;
1035 long cstime;
1036 long prio;
1037 long nice;
1038 long junk;
1039 long it_real;
1040 uintptr_t start;
1041 uintptr_t vsize;
1042 intptr_t rss;
1043 uintptr_t rsslim;
1044 uintptr_t scodes;
1045 uintptr_t ecode;
1046 int i;
1047
1048 // Figure what the primordial thread stack base is. Code is inspired
1049 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1050 // followed by command name surrounded by parentheses, state, etc.
1051 char stat[2048];
1052 int statlen;
1053
1054 fp = fopen("/proc/self/stat", "r");
1055 if (fp) {
1056 statlen = fread(stat, 1, 2047, fp);
1057 stat[statlen] = '\0';
1058 fclose(fp);
1059
1060 // Skip pid and the command string. Note that we could be dealing with
1061 // weird command names, e.g. user could decide to rename java launcher
1062 // to "java 1.4.2 :)", then the stat file would look like
1063 // 1234 (java 1.4.2 :)) R ... ...
1064 // We don't really need to know the command string, just find the last
1065 // occurrence of ")" and then start parsing from there. See bug 4726580.
1066 char * s = strrchr(stat, ')');
1067
1068 i = 0;
1069 if (s) {
1070 // Skip blank chars
1071 do { s++; } while (s && isspace(*s));
1072
1073 #define _UFM UINTX_FORMAT
1074 #define _DFM INTX_FORMAT
1075
1076 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1077 // 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
1078 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1079 &state, // 3 %c
1080 &ppid, // 4 %d
1081 &pgrp, // 5 %d
1082 &session, // 6 %d
1083 &nr, // 7 %d
1084 &tpgrp, // 8 %d
1085 &flags, // 9 %lu
1086 &minflt, // 10 %lu
1087 &cminflt, // 11 %lu
1088 &majflt, // 12 %lu
1089 &cmajflt, // 13 %lu
1090 &utime, // 14 %lu
1091 &stime, // 15 %lu
1092 &cutime, // 16 %ld
1093 &cstime, // 17 %ld
1094 &prio, // 18 %ld
1095 &nice, // 19 %ld
1096 &junk, // 20 %ld
1097 &it_real, // 21 %ld
1098 &start, // 22 UINTX_FORMAT
1099 &vsize, // 23 UINTX_FORMAT
1100 &rss, // 24 INTX_FORMAT
1101 &rsslim, // 25 UINTX_FORMAT
1102 &scodes, // 26 UINTX_FORMAT
1103 &ecode, // 27 UINTX_FORMAT
1104 &stack_start); // 28 UINTX_FORMAT
1105 }
1106
1107 #undef _UFM
1108 #undef _DFM
1109
1110 if (i != 28 - 2) {
1111 assert(false, "Bad conversion from /proc/self/stat");
1112 // product mode - assume we are the primordial thread, good luck in the
1113 // embedded case.
1114 warning("Can't detect primordial thread stack location - bad conversion");
1115 stack_start = (uintptr_t) &rlim;
1116 }
1117 } else {
1118 // For some reason we can't open /proc/self/stat (for example, running on
1119 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1120 // most cases, so don't abort:
1121 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1122 stack_start = (uintptr_t) &rlim;
1123 }
1124 }
1125
1126 // Now we have a pointer (stack_start) very close to the stack top, the
1127 // next thing to do is to figure out the exact location of stack top. We
1128 // can find out the virtual memory area that contains stack_start by
1129 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1130 // and its upper limit is the real stack top. (again, this would fail if
1131 // running inside chroot, because /proc may not exist.)
1132
1133 uintptr_t stack_top;
1134 address low, high;
1135 if (find_vma((address)stack_start, &low, &high)) {
1136 // success, "high" is the true stack top. (ignore "low", because initial
1137 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1138 stack_top = (uintptr_t)high;
1139 } else {
1140 // failed, likely because /proc/self/maps does not exist
1141 warning("Can't detect primordial thread stack location - find_vma failed");
1142 // best effort: stack_start is normally within a few pages below the real
1143 // stack top, use it as stack top, and reduce stack size so we won't put
1144 // guard page outside stack.
1145 stack_top = stack_start;
1146 stack_size -= 16 * page_size();
1147 }
1148
1149 // stack_top could be partially down the page so align it
1150 stack_top = align_up(stack_top, page_size());
1151
1152 // Allowed stack value is minimum of max_size and what we derived from rlimit
1153 if (max_size > 0) {
1154 _initial_thread_stack_size = MIN2(max_size, stack_size);
1155 } else {
1156 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1157 // clamp it at 8MB as we do on Solaris
1158 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1159 }
1160 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1161 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1162
1163 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1164
1165 if (log_is_enabled(Info, os, thread)) {
1166 // See if we seem to be on primordial process thread
1167 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1168 uintptr_t(&rlim) < stack_top;
1169
1170 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1171 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1172 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1173 stack_top, intptr_t(_initial_thread_stack_bottom));
1174 }
1175 }
1176
1177 ////////////////////////////////////////////////////////////////////////////////
1178 // time support
1179
1180 #ifndef SUPPORTS_CLOCK_MONOTONIC
1181 #error "Build platform doesn't support clock_gettime and related functionality"
1182 #endif
1183
1184 // Time since start-up in seconds to a fine granularity.
1185 // Used by VMSelfDestructTimer and the MemProfiler.
1186 double os::elapsedTime() {
1187
1188 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1189 }
1190
1191 jlong os::elapsed_counter() {
1192 return javaTimeNanos() - initial_time_count;
1193 }
1194
1195 jlong os::elapsed_frequency() {
1196 return NANOSECS_PER_SEC; // nanosecond resolution
1197 }
1198
1199 bool os::supports_vtime() { return true; }
1200 bool os::enable_vtime() { return false; }
1201 bool os::vtime_enabled() { return false; }
1202
1203 double os::elapsedVTime() {
1204 struct rusage usage;
1205 int retval = getrusage(RUSAGE_THREAD, &usage);
1206 if (retval == 0) {
1207 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1208 } else {
1209 // better than nothing, but not much
1210 return elapsedTime();
1211 }
1212 }
1213
1214 jlong os::javaTimeMillis() {
1215 timeval time;
1216 int status = gettimeofday(&time, NULL);
1217 assert(status != -1, "linux error");
1218 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1219 }
1220
1221 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1222 timeval time;
1223 int status = gettimeofday(&time, NULL);
1224 assert(status != -1, "linux error");
1225 seconds = jlong(time.tv_sec);
1226 nanos = jlong(time.tv_usec) * 1000;
1227 }
1228
1229 void os::Linux::fast_thread_clock_init() {
1230 if (!UseLinuxPosixThreadCPUClocks) {
1231 return;
1232 }
1233 clockid_t clockid;
1234 struct timespec tp;
1235 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1236 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1237
1238 // Switch to using fast clocks for thread cpu time if
1239 // the clock_getres() returns 0 error code.
1240 // Note, that some kernels may support the current thread
1241 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1242 // returned by the pthread_getcpuclockid().
1243 // If the fast Posix clocks are supported then the clock_getres()
1244 // must return at least tp.tv_sec == 0 which means a resolution
1245 // better than 1 sec. This is extra check for reliability.
1246
1247 if (pthread_getcpuclockid_func &&
1248 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1249 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1250 _supports_fast_thread_cpu_time = true;
1251 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1252 }
1253 }
1254
1255 jlong os::javaTimeNanos() {
1256 if (os::supports_monotonic_clock()) {
1257 struct timespec tp;
1258 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1259 assert(status == 0, "gettime error");
1260 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1261 return result;
1262 } else {
1263 timeval time;
1264 int status = gettimeofday(&time, NULL);
1265 assert(status != -1, "linux error");
1266 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1267 return 1000 * usecs;
1268 }
1269 }
1270
1271 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1272 if (os::supports_monotonic_clock()) {
1273 info_ptr->max_value = ALL_64_BITS;
1274
1275 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1276 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1277 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1278 } else {
1279 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1280 info_ptr->max_value = ALL_64_BITS;
1281
1282 // gettimeofday is a real time clock so it skips
1283 info_ptr->may_skip_backward = true;
1284 info_ptr->may_skip_forward = true;
1285 }
1286
1287 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1288 }
1289
1290 // Return the real, user, and system times in seconds from an
1291 // arbitrary fixed point in the past.
1292 bool os::getTimesSecs(double* process_real_time,
1293 double* process_user_time,
1294 double* process_system_time) {
1295 struct tms ticks;
1296 clock_t real_ticks = times(&ticks);
1297
1298 if (real_ticks == (clock_t) (-1)) {
1299 return false;
1300 } else {
1301 double ticks_per_second = (double) clock_tics_per_sec;
1302 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1303 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1304 *process_real_time = ((double) real_ticks) / ticks_per_second;
1305
1306 return true;
1307 }
1308 }
1309
1310
1311 char * os::local_time_string(char *buf, size_t buflen) {
1312 struct tm t;
1313 time_t long_time;
1314 time(&long_time);
1315 localtime_r(&long_time, &t);
1316 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1317 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1318 t.tm_hour, t.tm_min, t.tm_sec);
1319 return buf;
1320 }
1321
1322 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1323 return localtime_r(clock, res);
1324 }
1325
1326 ////////////////////////////////////////////////////////////////////////////////
1327 // runtime exit support
1328
1329 // Note: os::shutdown() might be called very early during initialization, or
1330 // called from signal handler. Before adding something to os::shutdown(), make
1331 // sure it is async-safe and can handle partially initialized VM.
1332 void os::shutdown() {
1333
1334 // allow PerfMemory to attempt cleanup of any persistent resources
1335 perfMemory_exit();
1336
1337 // needs to remove object in file system
1338 AttachListener::abort();
1339
1340 // flush buffered output, finish log files
1341 ostream_abort();
1342
1343 // Check for abort hook
1344 abort_hook_t abort_hook = Arguments::abort_hook();
1345 if (abort_hook != NULL) {
1346 abort_hook();
1347 }
1348
1349 }
1350
1351 // Note: os::abort() might be called very early during initialization, or
1352 // called from signal handler. Before adding something to os::abort(), make
1353 // sure it is async-safe and can handle partially initialized VM.
1354 void os::abort(bool dump_core, void* siginfo, const void* context) {
1355 os::shutdown();
1356 if (dump_core) {
1357 if (UseSharedSpaces && DumpPrivateMappingsInCore) {
1358 ClassLoader::close_jrt_image();
1359 }
1360 #ifndef PRODUCT
1361 fdStream out(defaultStream::output_fd());
1362 out.print_raw("Current thread is ");
1363 char buf[16];
1364 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1365 out.print_raw_cr(buf);
1366 out.print_raw_cr("Dumping core ...");
1367 #endif
1368 ::abort(); // dump core
1369 }
1370
1371 ::exit(1);
1372 }
1373
1374 // Die immediately, no exit hook, no abort hook, no cleanup.
1375 void os::die() {
1376 ::abort();
1377 }
1378
1379 // thread_id is kernel thread id (similar to Solaris LWP id)
1380 intx os::current_thread_id() { return os::Linux::gettid(); }
1381 int os::current_process_id() {
1382 return ::getpid();
1383 }
1384
1385 // DLL functions
1386
1387 const char* os::dll_file_extension() { return ".so"; }
1388
1389 // This must be hard coded because it's the system's temporary
1390 // directory not the java application's temp directory, ala java.io.tmpdir.
1391 const char* os::get_temp_directory() { return "/tmp"; }
1392
1393 static bool file_exists(const char* filename) {
1394 struct stat statbuf;
1395 if (filename == NULL || strlen(filename) == 0) {
1396 return false;
1397 }
1398 return os::stat(filename, &statbuf) == 0;
1399 }
1400
1401 // check if addr is inside libjvm.so
1402 bool os::address_is_in_vm(address addr) {
1403 static address libjvm_base_addr;
1404 Dl_info dlinfo;
1405
1406 if (libjvm_base_addr == NULL) {
1407 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1408 libjvm_base_addr = (address)dlinfo.dli_fbase;
1409 }
1410 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1411 }
1412
1413 if (dladdr((void *)addr, &dlinfo) != 0) {
1414 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1415 }
1416
1417 return false;
1418 }
1419
1420 bool os::dll_address_to_function_name(address addr, char *buf,
1421 int buflen, int *offset,
1422 bool demangle) {
1423 // buf is not optional, but offset is optional
1424 assert(buf != NULL, "sanity check");
1425
1426 Dl_info dlinfo;
1427
1428 if (dladdr((void*)addr, &dlinfo) != 0) {
1429 // see if we have a matching symbol
1430 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1431 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1432 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1433 }
1434 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1435 return true;
1436 }
1437 // no matching symbol so try for just file info
1438 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1439 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1440 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1441 return true;
1442 }
1443 }
1444 }
1445
1446 buf[0] = '\0';
1447 if (offset != NULL) *offset = -1;
1448 return false;
1449 }
1450
1451 struct _address_to_library_name {
1452 address addr; // input : memory address
1453 size_t buflen; // size of fname
1454 char* fname; // output: library name
1455 address base; // library base addr
1456 };
1457
1458 static int address_to_library_name_callback(struct dl_phdr_info *info,
1459 size_t size, void *data) {
1460 int i;
1461 bool found = false;
1462 address libbase = NULL;
1463 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1464
1465 // iterate through all loadable segments
1466 for (i = 0; i < info->dlpi_phnum; i++) {
1467 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1468 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1469 // base address of a library is the lowest address of its loaded
1470 // segments.
1471 if (libbase == NULL || libbase > segbase) {
1472 libbase = segbase;
1473 }
1474 // see if 'addr' is within current segment
1475 if (segbase <= d->addr &&
1476 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1477 found = true;
1478 }
1479 }
1480 }
1481
1482 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1483 // so dll_address_to_library_name() can fall through to use dladdr() which
1484 // can figure out executable name from argv[0].
1485 if (found && info->dlpi_name && info->dlpi_name[0]) {
1486 d->base = libbase;
1487 if (d->fname) {
1488 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1489 }
1490 return 1;
1491 }
1492 return 0;
1493 }
1494
1495 bool os::dll_address_to_library_name(address addr, char* buf,
1496 int buflen, int* offset) {
1497 // buf is not optional, but offset is optional
1498 assert(buf != NULL, "sanity check");
1499
1500 Dl_info dlinfo;
1501 struct _address_to_library_name data;
1502
1503 // There is a bug in old glibc dladdr() implementation that it could resolve
1504 // to wrong library name if the .so file has a base address != NULL. Here
1505 // we iterate through the program headers of all loaded libraries to find
1506 // out which library 'addr' really belongs to. This workaround can be
1507 // removed once the minimum requirement for glibc is moved to 2.3.x.
1508 data.addr = addr;
1509 data.fname = buf;
1510 data.buflen = buflen;
1511 data.base = NULL;
1512 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1513
1514 if (rslt) {
1515 // buf already contains library name
1516 if (offset) *offset = addr - data.base;
1517 return true;
1518 }
1519 if (dladdr((void*)addr, &dlinfo) != 0) {
1520 if (dlinfo.dli_fname != NULL) {
1521 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1522 }
1523 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1524 *offset = addr - (address)dlinfo.dli_fbase;
1525 }
1526 return true;
1527 }
1528
1529 buf[0] = '\0';
1530 if (offset) *offset = -1;
1531 return false;
1532 }
1533
1534 // Loads .dll/.so and
1535 // in case of error it checks if .dll/.so was built for the
1536 // same architecture as Hotspot is running on
1537
1538
1539 // Remember the stack's state. The Linux dynamic linker will change
1540 // the stack to 'executable' at most once, so we must safepoint only once.
1541 bool os::Linux::_stack_is_executable = false;
1542
1543 // VM operation that loads a library. This is necessary if stack protection
1544 // of the Java stacks can be lost during loading the library. If we
1545 // do not stop the Java threads, they can stack overflow before the stacks
1546 // are protected again.
1547 class VM_LinuxDllLoad: public VM_Operation {
1548 private:
1549 const char *_filename;
1550 char *_ebuf;
1551 int _ebuflen;
1552 void *_lib;
1553 public:
1554 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1555 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1556 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1557 void doit() {
1558 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1559 os::Linux::_stack_is_executable = true;
1560 }
1561 void* loaded_library() { return _lib; }
1562 };
1563
1564 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1565 void * result = NULL;
1566 bool load_attempted = false;
1567
1568 // Check whether the library to load might change execution rights
1569 // of the stack. If they are changed, the protection of the stack
1570 // guard pages will be lost. We need a safepoint to fix this.
1571 //
1572 // See Linux man page execstack(8) for more info.
1573 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1574 if (!ElfFile::specifies_noexecstack(filename)) {
1575 if (!is_init_completed()) {
1576 os::Linux::_stack_is_executable = true;
1577 // This is OK - No Java threads have been created yet, and hence no
1578 // stack guard pages to fix.
1579 //
1580 // Dynamic loader will make all stacks executable after
1581 // this function returns, and will not do that again.
1582 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1583 } else {
1584 warning("You have loaded library %s which might have disabled stack guard. "
1585 "The VM will try to fix the stack guard now.\n"
1586 "It's highly recommended that you fix the library with "
1587 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1588 filename);
1589
1590 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1591 JavaThread *jt = JavaThread::current();
1592 if (jt->thread_state() != _thread_in_native) {
1593 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1594 // that requires ExecStack. Cannot enter safe point. Let's give up.
1595 warning("Unable to fix stack guard. Giving up.");
1596 } else {
1597 if (!LoadExecStackDllInVMThread) {
1598 // This is for the case where the DLL has an static
1599 // constructor function that executes JNI code. We cannot
1600 // load such DLLs in the VMThread.
1601 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1602 }
1603
1604 ThreadInVMfromNative tiv(jt);
1605 debug_only(VMNativeEntryWrapper vew;)
1606
1607 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1608 VMThread::execute(&op);
1609 if (LoadExecStackDllInVMThread) {
1610 result = op.loaded_library();
1611 }
1612 load_attempted = true;
1613 }
1614 }
1615 }
1616 }
1617
1618 if (!load_attempted) {
1619 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1620 }
1621
1622 if (result != NULL) {
1623 // Successful loading
1624 return result;
1625 }
1626
1627 Elf32_Ehdr elf_head;
1628 int diag_msg_max_length=ebuflen-strlen(ebuf);
1629 char* diag_msg_buf=ebuf+strlen(ebuf);
1630
1631 if (diag_msg_max_length==0) {
1632 // No more space in ebuf for additional diagnostics message
1633 return NULL;
1634 }
1635
1636
1637 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1638
1639 if (file_descriptor < 0) {
1640 // Can't open library, report dlerror() message
1641 return NULL;
1642 }
1643
1644 bool failed_to_read_elf_head=
1645 (sizeof(elf_head)!=
1646 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1647
1648 ::close(file_descriptor);
1649 if (failed_to_read_elf_head) {
1650 // file i/o error - report dlerror() msg
1651 return NULL;
1652 }
1653
1654 typedef struct {
1655 Elf32_Half code; // Actual value as defined in elf.h
1656 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1657 unsigned char elf_class; // 32 or 64 bit
1658 unsigned char endianess; // MSB or LSB
1659 char* name; // String representation
1660 } arch_t;
1661
1662 #ifndef EM_486
1663 #define EM_486 6 /* Intel 80486 */
1664 #endif
1665 #ifndef EM_AARCH64
1666 #define EM_AARCH64 183 /* ARM AARCH64 */
1667 #endif
1668
1669 static const arch_t arch_array[]={
1670 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1671 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1672 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1673 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1674 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1675 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1676 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1677 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1678 #if defined(VM_LITTLE_ENDIAN)
1679 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1680 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1681 #else
1682 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1683 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1684 #endif
1685 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1686 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1687 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1688 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1689 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1690 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1691 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1692 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1693 };
1694
1695 #if (defined IA32)
1696 static Elf32_Half running_arch_code=EM_386;
1697 #elif (defined AMD64) || (defined X32)
1698 static Elf32_Half running_arch_code=EM_X86_64;
1699 #elif (defined IA64)
1700 static Elf32_Half running_arch_code=EM_IA_64;
1701 #elif (defined __sparc) && (defined _LP64)
1702 static Elf32_Half running_arch_code=EM_SPARCV9;
1703 #elif (defined __sparc) && (!defined _LP64)
1704 static Elf32_Half running_arch_code=EM_SPARC;
1705 #elif (defined __powerpc64__)
1706 static Elf32_Half running_arch_code=EM_PPC64;
1707 #elif (defined __powerpc__)
1708 static Elf32_Half running_arch_code=EM_PPC;
1709 #elif (defined AARCH64)
1710 static Elf32_Half running_arch_code=EM_AARCH64;
1711 #elif (defined ARM)
1712 static Elf32_Half running_arch_code=EM_ARM;
1713 #elif (defined S390)
1714 static Elf32_Half running_arch_code=EM_S390;
1715 #elif (defined ALPHA)
1716 static Elf32_Half running_arch_code=EM_ALPHA;
1717 #elif (defined MIPSEL)
1718 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1719 #elif (defined PARISC)
1720 static Elf32_Half running_arch_code=EM_PARISC;
1721 #elif (defined MIPS)
1722 static Elf32_Half running_arch_code=EM_MIPS;
1723 #elif (defined M68K)
1724 static Elf32_Half running_arch_code=EM_68K;
1725 #elif (defined SH)
1726 static Elf32_Half running_arch_code=EM_SH;
1727 #else
1728 #error Method os::dll_load requires that one of following is defined:\
1729 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1730 #endif
1731
1732 // Identify compatability class for VM's architecture and library's architecture
1733 // Obtain string descriptions for architectures
1734
1735 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1736 int running_arch_index=-1;
1737
1738 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1739 if (running_arch_code == arch_array[i].code) {
1740 running_arch_index = i;
1741 }
1742 if (lib_arch.code == arch_array[i].code) {
1743 lib_arch.compat_class = arch_array[i].compat_class;
1744 lib_arch.name = arch_array[i].name;
1745 }
1746 }
1747
1748 assert(running_arch_index != -1,
1749 "Didn't find running architecture code (running_arch_code) in arch_array");
1750 if (running_arch_index == -1) {
1751 // Even though running architecture detection failed
1752 // we may still continue with reporting dlerror() message
1753 return NULL;
1754 }
1755
1756 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1757 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1758 return NULL;
1759 }
1760
1761 #ifndef S390
1762 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1763 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1764 return NULL;
1765 }
1766 #endif // !S390
1767
1768 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1769 if (lib_arch.name!=NULL) {
1770 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1771 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1772 lib_arch.name, arch_array[running_arch_index].name);
1773 } else {
1774 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1775 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1776 lib_arch.code,
1777 arch_array[running_arch_index].name);
1778 }
1779 }
1780
1781 return NULL;
1782 }
1783
1784 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1785 int ebuflen) {
1786 void * result = ::dlopen(filename, RTLD_LAZY);
1787 if (result == NULL) {
1788 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1789 ebuf[ebuflen-1] = '\0';
1790 }
1791 return result;
1792 }
1793
1794 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1795 int ebuflen) {
1796 void * result = NULL;
1797 if (LoadExecStackDllInVMThread) {
1798 result = dlopen_helper(filename, ebuf, ebuflen);
1799 }
1800
1801 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1802 // library that requires an executable stack, or which does not have this
1803 // stack attribute set, dlopen changes the stack attribute to executable. The
1804 // read protection of the guard pages gets lost.
1805 //
1806 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1807 // may have been queued at the same time.
1808
1809 if (!_stack_is_executable) {
1810 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1811 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1812 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1813 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1814 warning("Attempt to reguard stack yellow zone failed.");
1815 }
1816 }
1817 }
1818 }
1819
1820 return result;
1821 }
1822
1823 void* os::dll_lookup(void* handle, const char* name) {
1824 void* res = dlsym(handle, name);
1825 return res;
1826 }
1827
1828 void* os::get_default_process_handle() {
1829 return (void*)::dlopen(NULL, RTLD_LAZY);
1830 }
1831
1832 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1833 int fd = ::open(filename, O_RDONLY);
1834 if (fd == -1) {
1835 return false;
1836 }
1837
1838 if (hdr != NULL) {
1839 st->print_cr("%s", hdr);
1840 }
1841
1842 char buf[33];
1843 int bytes;
1844 buf[32] = '\0';
1845 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1846 st->print_raw(buf, bytes);
1847 }
1848
1849 ::close(fd);
1850
1851 return true;
1852 }
1853
1854 void os::print_dll_info(outputStream *st) {
1855 st->print_cr("Dynamic libraries:");
1856
1857 char fname[32];
1858 pid_t pid = os::Linux::gettid();
1859
1860 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1861
1862 if (!_print_ascii_file(fname, st)) {
1863 st->print("Can not get library information for pid = %d\n", pid);
1864 }
1865 }
1866
1867 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1868 FILE *procmapsFile = NULL;
1869
1870 // Open the procfs maps file for the current process
1871 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1872 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1873 char line[PATH_MAX + 100];
1874
1875 // Read line by line from 'file'
1876 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1877 u8 base, top, offset, inode;
1878 char permissions[5];
1879 char device[6];
1880 char name[PATH_MAX + 1];
1881
1882 // Parse fields from line
1883 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1884 &base, &top, permissions, &offset, device, &inode, name);
1885
1886 // Filter by device id '00:00' so that we only get file system mapped files.
1887 if (strcmp(device, "00:00") != 0) {
1888
1889 // Call callback with the fields of interest
1890 if(callback(name, (address)base, (address)top, param)) {
1891 // Oops abort, callback aborted
1892 fclose(procmapsFile);
1893 return 1;
1894 }
1895 }
1896 }
1897 fclose(procmapsFile);
1898 }
1899 return 0;
1900 }
1901
1902 void os::print_os_info_brief(outputStream* st) {
1903 os::Linux::print_distro_info(st);
1904
1905 os::Posix::print_uname_info(st);
1906
1907 os::Linux::print_libversion_info(st);
1908
1909 }
1910
1911 void os::print_os_info(outputStream* st) {
1912 st->print("OS:");
1913
1914 os::Linux::print_distro_info(st);
1915
1916 os::Posix::print_uname_info(st);
1917
1918 // Print warning if unsafe chroot environment detected
1919 if (unsafe_chroot_detected) {
1920 st->print("WARNING!! ");
1921 st->print_cr("%s", unstable_chroot_error);
1922 }
1923
1924 os::Linux::print_libversion_info(st);
1925
1926 os::Posix::print_rlimit_info(st);
1927
1928 os::Posix::print_load_average(st);
1929
1930 os::Linux::print_full_memory_info(st);
1931
1932 os::Linux::print_proc_sys_info(st);
1933
1934 os::Linux::print_ld_preload_file(st);
1935
1936 os::Linux::print_container_info(st);
1937 }
1938
1939 // Try to identify popular distros.
1940 // Most Linux distributions have a /etc/XXX-release file, which contains
1941 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1942 // file that also contains the OS version string. Some have more than one
1943 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1944 // /etc/redhat-release.), so the order is important.
1945 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1946 // their own specific XXX-release file as well as a redhat-release file.
1947 // Because of this the XXX-release file needs to be searched for before the
1948 // redhat-release file.
1949 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1950 // search for redhat-release / SuSE-release needs to be before lsb-release.
1951 // Since the lsb-release file is the new standard it needs to be searched
1952 // before the older style release files.
1953 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1954 // next to last resort. The os-release file is a new standard that contains
1955 // distribution information and the system-release file seems to be an old
1956 // standard that has been replaced by the lsb-release and os-release files.
1957 // Searching for the debian_version file is the last resort. It contains
1958 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1959 // "Debian " is printed before the contents of the debian_version file.
1960
1961 const char* distro_files[] = {
1962 "/etc/oracle-release",
1963 "/etc/mandriva-release",
1964 "/etc/mandrake-release",
1965 "/etc/sun-release",
1966 "/etc/redhat-release",
1967 "/etc/SuSE-release",
1968 "/etc/lsb-release",
1969 "/etc/turbolinux-release",
1970 "/etc/gentoo-release",
1971 "/etc/ltib-release",
1972 "/etc/angstrom-version",
1973 "/etc/system-release",
1974 "/etc/os-release",
1975 NULL };
1976
1977 void os::Linux::print_distro_info(outputStream* st) {
1978 for (int i = 0;; i++) {
1979 const char* file = distro_files[i];
1980 if (file == NULL) {
1981 break; // done
1982 }
1983 // If file prints, we found it.
1984 if (_print_ascii_file(file, st)) {
1985 return;
1986 }
1987 }
1988
1989 if (file_exists("/etc/debian_version")) {
1990 st->print("Debian ");
1991 _print_ascii_file("/etc/debian_version", st);
1992 } else {
1993 st->print("Linux");
1994 }
1995 st->cr();
1996 }
1997
1998 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
1999 char buf[256];
2000 while (fgets(buf, sizeof(buf), fp)) {
2001 // Edit out extra stuff in expected format
2002 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2003 char* ptr = strstr(buf, "\""); // the name is in quotes
2004 if (ptr != NULL) {
2005 ptr++; // go beyond first quote
2006 char* nl = strchr(ptr, '\"');
2007 if (nl != NULL) *nl = '\0';
2008 strncpy(distro, ptr, length);
2009 } else {
2010 ptr = strstr(buf, "=");
2011 ptr++; // go beyond equals then
2012 char* nl = strchr(ptr, '\n');
2013 if (nl != NULL) *nl = '\0';
2014 strncpy(distro, ptr, length);
2015 }
2016 return;
2017 } else if (get_first_line) {
2018 char* nl = strchr(buf, '\n');
2019 if (nl != NULL) *nl = '\0';
2020 strncpy(distro, buf, length);
2021 return;
2022 }
2023 }
2024 // print last line and close
2025 char* nl = strchr(buf, '\n');
2026 if (nl != NULL) *nl = '\0';
2027 strncpy(distro, buf, length);
2028 }
2029
2030 static void parse_os_info(char* distro, size_t length, const char* file) {
2031 FILE* fp = fopen(file, "r");
2032 if (fp != NULL) {
2033 // if suse format, print out first line
2034 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2035 parse_os_info_helper(fp, distro, length, get_first_line);
2036 fclose(fp);
2037 }
2038 }
2039
2040 void os::get_summary_os_info(char* buf, size_t buflen) {
2041 for (int i = 0;; i++) {
2042 const char* file = distro_files[i];
2043 if (file == NULL) {
2044 break; // ran out of distro_files
2045 }
2046 if (file_exists(file)) {
2047 parse_os_info(buf, buflen, file);
2048 return;
2049 }
2050 }
2051 // special case for debian
2052 if (file_exists("/etc/debian_version")) {
2053 strncpy(buf, "Debian ", buflen);
2054 if (buflen > 7) {
2055 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2056 }
2057 } else {
2058 strncpy(buf, "Linux", buflen);
2059 }
2060 }
2061
2062 void os::Linux::print_libversion_info(outputStream* st) {
2063 // libc, pthread
2064 st->print("libc:");
2065 st->print("%s ", os::Linux::glibc_version());
2066 st->print("%s ", os::Linux::libpthread_version());
2067 st->cr();
2068 }
2069
2070 void os::Linux::print_proc_sys_info(outputStream* st) {
2071 st->cr();
2072 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2073 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2074 st->cr();
2075 st->cr();
2076
2077 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2078 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2079 st->cr();
2080 st->cr();
2081
2082 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2083 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2084 st->cr();
2085 st->cr();
2086 }
2087
2088 void os::Linux::print_full_memory_info(outputStream* st) {
2089 st->print("\n/proc/meminfo:\n");
2090 _print_ascii_file("/proc/meminfo", st);
2091 st->cr();
2092 }
2093
2094 void os::Linux::print_ld_preload_file(outputStream* st) {
2095 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2096 st->cr();
2097 }
2098
2099 void os::Linux::print_container_info(outputStream* st) {
2100 if (!OSContainer::is_containerized()) {
2101 return;
2102 }
2103
2104 st->print("container (cgroup) information:\n");
2105
2106 const char *p_ct = OSContainer::container_type();
2107 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2108
2109 char *p = OSContainer::cpu_cpuset_cpus();
2110 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2111 free(p);
2112
2113 p = OSContainer::cpu_cpuset_memory_nodes();
2114 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2115 free(p);
2116
2117 int i = OSContainer::active_processor_count();
2118 if (i > 0) {
2119 st->print("active_processor_count: %d\n", i);
2120 } else {
2121 st->print("active_processor_count: failed\n");
2122 }
2123
2124 i = OSContainer::cpu_quota();
2125 st->print("cpu_quota: %d\n", i);
2126
2127 i = OSContainer::cpu_period();
2128 st->print("cpu_period: %d\n", i);
2129
2130 i = OSContainer::cpu_shares();
2131 st->print("cpu_shares: %d\n", i);
2132
2133 jlong j = OSContainer::memory_limit_in_bytes();
2134 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2135
2136 j = OSContainer::memory_and_swap_limit_in_bytes();
2137 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2138
2139 j = OSContainer::memory_soft_limit_in_bytes();
2140 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2141
2142 j = OSContainer::OSContainer::memory_usage_in_bytes();
2143 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2144
2145 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2146 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2147 st->cr();
2148 }
2149
2150 void os::print_memory_info(outputStream* st) {
2151
2152 st->print("Memory:");
2153 st->print(" %dk page", os::vm_page_size()>>10);
2154
2155 // values in struct sysinfo are "unsigned long"
2156 struct sysinfo si;
2157 sysinfo(&si);
2158
2159 st->print(", physical " UINT64_FORMAT "k",
2160 os::physical_memory() >> 10);
2161 st->print("(" UINT64_FORMAT "k free)",
2162 os::available_memory() >> 10);
2163 st->print(", swap " UINT64_FORMAT "k",
2164 ((jlong)si.totalswap * si.mem_unit) >> 10);
2165 st->print("(" UINT64_FORMAT "k free)",
2166 ((jlong)si.freeswap * si.mem_unit) >> 10);
2167 st->cr();
2168 }
2169
2170 // Print the first "model name" line and the first "flags" line
2171 // that we find and nothing more. We assume "model name" comes
2172 // before "flags" so if we find a second "model name", then the
2173 // "flags" field is considered missing.
2174 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2175 #if defined(IA32) || defined(AMD64)
2176 // Other platforms have less repetitive cpuinfo files
2177 FILE *fp = fopen("/proc/cpuinfo", "r");
2178 if (fp) {
2179 while (!feof(fp)) {
2180 if (fgets(buf, buflen, fp)) {
2181 // Assume model name comes before flags
2182 bool model_name_printed = false;
2183 if (strstr(buf, "model name") != NULL) {
2184 if (!model_name_printed) {
2185 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2186 st->print_raw(buf);
2187 model_name_printed = true;
2188 } else {
2189 // model name printed but not flags? Odd, just return
2190 fclose(fp);
2191 return true;
2192 }
2193 }
2194 // print the flags line too
2195 if (strstr(buf, "flags") != NULL) {
2196 st->print_raw(buf);
2197 fclose(fp);
2198 return true;
2199 }
2200 }
2201 }
2202 fclose(fp);
2203 }
2204 #endif // x86 platforms
2205 return false;
2206 }
2207
2208 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2209 // Only print the model name if the platform provides this as a summary
2210 if (!print_model_name_and_flags(st, buf, buflen)) {
2211 st->print("\n/proc/cpuinfo:\n");
2212 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2213 st->print_cr(" <Not Available>");
2214 }
2215 }
2216 }
2217
2218 #if defined(AMD64) || defined(IA32) || defined(X32)
2219 const char* search_string = "model name";
2220 #elif defined(M68K)
2221 const char* search_string = "CPU";
2222 #elif defined(PPC64)
2223 const char* search_string = "cpu";
2224 #elif defined(S390)
2225 const char* search_string = "processor";
2226 #elif defined(SPARC)
2227 const char* search_string = "cpu";
2228 #else
2229 const char* search_string = "Processor";
2230 #endif
2231
2232 // Parses the cpuinfo file for string representing the model name.
2233 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2234 FILE* fp = fopen("/proc/cpuinfo", "r");
2235 if (fp != NULL) {
2236 while (!feof(fp)) {
2237 char buf[256];
2238 if (fgets(buf, sizeof(buf), fp)) {
2239 char* start = strstr(buf, search_string);
2240 if (start != NULL) {
2241 char *ptr = start + strlen(search_string);
2242 char *end = buf + strlen(buf);
2243 while (ptr != end) {
2244 // skip whitespace and colon for the rest of the name.
2245 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2246 break;
2247 }
2248 ptr++;
2249 }
2250 if (ptr != end) {
2251 // reasonable string, get rid of newline and keep the rest
2252 char* nl = strchr(buf, '\n');
2253 if (nl != NULL) *nl = '\0';
2254 strncpy(cpuinfo, ptr, length);
2255 fclose(fp);
2256 return;
2257 }
2258 }
2259 }
2260 }
2261 fclose(fp);
2262 }
2263 // cpuinfo not found or parsing failed, just print generic string. The entire
2264 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2265 #if defined(AARCH64)
2266 strncpy(cpuinfo, "AArch64", length);
2267 #elif defined(AMD64)
2268 strncpy(cpuinfo, "x86_64", length);
2269 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2270 strncpy(cpuinfo, "ARM", length);
2271 #elif defined(IA32)
2272 strncpy(cpuinfo, "x86_32", length);
2273 #elif defined(IA64)
2274 strncpy(cpuinfo, "IA64", length);
2275 #elif defined(PPC)
2276 strncpy(cpuinfo, "PPC64", length);
2277 #elif defined(S390)
2278 strncpy(cpuinfo, "S390", length);
2279 #elif defined(SPARC)
2280 strncpy(cpuinfo, "sparcv9", length);
2281 #elif defined(ZERO_LIBARCH)
2282 strncpy(cpuinfo, ZERO_LIBARCH, length);
2283 #else
2284 strncpy(cpuinfo, "unknown", length);
2285 #endif
2286 }
2287
2288 static void print_signal_handler(outputStream* st, int sig,
2289 char* buf, size_t buflen);
2290
2291 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2292 st->print_cr("Signal Handlers:");
2293 print_signal_handler(st, SIGSEGV, buf, buflen);
2294 print_signal_handler(st, SIGBUS , buf, buflen);
2295 print_signal_handler(st, SIGFPE , buf, buflen);
2296 print_signal_handler(st, SIGPIPE, buf, buflen);
2297 print_signal_handler(st, SIGXFSZ, buf, buflen);
2298 print_signal_handler(st, SIGILL , buf, buflen);
2299 print_signal_handler(st, SR_signum, buf, buflen);
2300 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2301 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2302 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2303 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2304 #if defined(PPC64)
2305 print_signal_handler(st, SIGTRAP, buf, buflen);
2306 #endif
2307 }
2308
2309 static char saved_jvm_path[MAXPATHLEN] = {0};
2310
2311 // Find the full path to the current module, libjvm.so
2312 void os::jvm_path(char *buf, jint buflen) {
2313 // Error checking.
2314 if (buflen < MAXPATHLEN) {
2315 assert(false, "must use a large-enough buffer");
2316 buf[0] = '\0';
2317 return;
2318 }
2319 // Lazy resolve the path to current module.
2320 if (saved_jvm_path[0] != 0) {
2321 strcpy(buf, saved_jvm_path);
2322 return;
2323 }
2324
2325 char dli_fname[MAXPATHLEN];
2326 bool ret = dll_address_to_library_name(
2327 CAST_FROM_FN_PTR(address, os::jvm_path),
2328 dli_fname, sizeof(dli_fname), NULL);
2329 assert(ret, "cannot locate libjvm");
2330 char *rp = NULL;
2331 if (ret && dli_fname[0] != '\0') {
2332 rp = os::Posix::realpath(dli_fname, buf, buflen);
2333 }
2334 if (rp == NULL) {
2335 return;
2336 }
2337
2338 if (Arguments::sun_java_launcher_is_altjvm()) {
2339 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2340 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2341 // If "/jre/lib/" appears at the right place in the string, then
2342 // assume we are installed in a JDK and we're done. Otherwise, check
2343 // for a JAVA_HOME environment variable and fix up the path so it
2344 // looks like libjvm.so is installed there (append a fake suffix
2345 // hotspot/libjvm.so).
2346 const char *p = buf + strlen(buf) - 1;
2347 for (int count = 0; p > buf && count < 5; ++count) {
2348 for (--p; p > buf && *p != '/'; --p)
2349 /* empty */ ;
2350 }
2351
2352 if (strncmp(p, "/jre/lib/", 9) != 0) {
2353 // Look for JAVA_HOME in the environment.
2354 char* java_home_var = ::getenv("JAVA_HOME");
2355 if (java_home_var != NULL && java_home_var[0] != 0) {
2356 char* jrelib_p;
2357 int len;
2358
2359 // Check the current module name "libjvm.so".
2360 p = strrchr(buf, '/');
2361 if (p == NULL) {
2362 return;
2363 }
2364 assert(strstr(p, "/libjvm") == p, "invalid library name");
2365
2366 rp = os::Posix::realpath(java_home_var, buf, buflen);
2367 if (rp == NULL) {
2368 return;
2369 }
2370
2371 // determine if this is a legacy image or modules image
2372 // modules image doesn't have "jre" subdirectory
2373 len = strlen(buf);
2374 assert(len < buflen, "Ran out of buffer room");
2375 jrelib_p = buf + len;
2376 snprintf(jrelib_p, buflen-len, "/jre/lib");
2377 if (0 != access(buf, F_OK)) {
2378 snprintf(jrelib_p, buflen-len, "/lib");
2379 }
2380
2381 if (0 == access(buf, F_OK)) {
2382 // Use current module name "libjvm.so"
2383 len = strlen(buf);
2384 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2385 } else {
2386 // Go back to path of .so
2387 rp = os::Posix::realpath(dli_fname, buf, buflen);
2388 if (rp == NULL) {
2389 return;
2390 }
2391 }
2392 }
2393 }
2394 }
2395
2396 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2397 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2398 }
2399
2400 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2401 // no prefix required, not even "_"
2402 }
2403
2404 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2405 // no suffix required
2406 }
2407
2408 ////////////////////////////////////////////////////////////////////////////////
2409 // sun.misc.Signal support
2410
2411 static volatile jint sigint_count = 0;
2412
2413 static void UserHandler(int sig, void *siginfo, void *context) {
2414 // 4511530 - sem_post is serialized and handled by the manager thread. When
2415 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2416 // don't want to flood the manager thread with sem_post requests.
2417 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2418 return;
2419 }
2420
2421 // Ctrl-C is pressed during error reporting, likely because the error
2422 // handler fails to abort. Let VM die immediately.
2423 if (sig == SIGINT && VMError::is_error_reported()) {
2424 os::die();
2425 }
2426
2427 os::signal_notify(sig);
2428 }
2429
2430 void* os::user_handler() {
2431 return CAST_FROM_FN_PTR(void*, UserHandler);
2432 }
2433
2434 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2435 struct timespec ts;
2436 // Semaphore's are always associated with CLOCK_REALTIME
2437 os::Posix::clock_gettime(CLOCK_REALTIME, &ts);
2438 // see os_posix.cpp for discussion on overflow checking
2439 if (sec >= MAX_SECS) {
2440 ts.tv_sec += MAX_SECS;
2441 ts.tv_nsec = 0;
2442 } else {
2443 ts.tv_sec += sec;
2444 ts.tv_nsec += nsec;
2445 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2446 ts.tv_nsec -= NANOSECS_PER_SEC;
2447 ++ts.tv_sec; // note: this must be <= max_secs
2448 }
2449 }
2450
2451 return ts;
2452 }
2453
2454 extern "C" {
2455 typedef void (*sa_handler_t)(int);
2456 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2457 }
2458
2459 void* os::signal(int signal_number, void* handler) {
2460 struct sigaction sigAct, oldSigAct;
2461
2462 sigfillset(&(sigAct.sa_mask));
2463 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2464 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2465
2466 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2467 // -1 means registration failed
2468 return (void *)-1;
2469 }
2470
2471 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2472 }
2473
2474 void os::signal_raise(int signal_number) {
2475 ::raise(signal_number);
2476 }
2477
2478 // The following code is moved from os.cpp for making this
2479 // code platform specific, which it is by its very nature.
2480
2481 // Will be modified when max signal is changed to be dynamic
2482 int os::sigexitnum_pd() {
2483 return NSIG;
2484 }
2485
2486 // a counter for each possible signal value
2487 static volatile jint pending_signals[NSIG+1] = { 0 };
2488
2489 // Linux(POSIX) specific hand shaking semaphore.
2490 static Semaphore* sig_sem = NULL;
2491 static PosixSemaphore sr_semaphore;
2492
2493 static void jdk_misc_signal_init() {
2494 // Initialize signal structures
2495 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2496
2497 // Initialize signal semaphore
2498 sig_sem = new Semaphore();
2499 }
2500
2501 void os::signal_notify(int sig) {
2502 if (sig_sem != NULL) {
2503 Atomic::inc(&pending_signals[sig]);
2504 sig_sem->signal();
2505 } else {
2506 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2507 // initialization isn't called.
2508 assert(ReduceSignalUsage, "signal semaphore should be created");
2509 }
2510 }
2511
2512 static int check_pending_signals() {
2513 Atomic::store(0, &sigint_count);
2514 for (;;) {
2515 for (int i = 0; i < NSIG + 1; i++) {
2516 jint n = pending_signals[i];
2517 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2518 return i;
2519 }
2520 }
2521 JavaThread *thread = JavaThread::current();
2522 ThreadBlockInVM tbivm(thread);
2523
2524 bool threadIsSuspended;
2525 do {
2526 thread->set_suspend_equivalent();
2527 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2528 sig_sem->wait();
2529
2530 // were we externally suspended while we were waiting?
2531 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2532 if (threadIsSuspended) {
2533 // The semaphore has been incremented, but while we were waiting
2534 // another thread suspended us. We don't want to continue running
2535 // while suspended because that would surprise the thread that
2536 // suspended us.
2537 sig_sem->signal();
2538
2539 thread->java_suspend_self();
2540 }
2541 } while (threadIsSuspended);
2542 }
2543 }
2544
2545 int os::signal_wait() {
2546 return check_pending_signals();
2547 }
2548
2549 ////////////////////////////////////////////////////////////////////////////////
2550 // Virtual Memory
2551
2552 int os::vm_page_size() {
2553 // Seems redundant as all get out
2554 assert(os::Linux::page_size() != -1, "must call os::init");
2555 return os::Linux::page_size();
2556 }
2557
2558 // Solaris allocates memory by pages.
2559 int os::vm_allocation_granularity() {
2560 assert(os::Linux::page_size() != -1, "must call os::init");
2561 return os::Linux::page_size();
2562 }
2563
2564 // Rationale behind this function:
2565 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2566 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2567 // samples for JITted code. Here we create private executable mapping over the code cache
2568 // and then we can use standard (well, almost, as mapping can change) way to provide
2569 // info for the reporting script by storing timestamp and location of symbol
2570 void linux_wrap_code(char* base, size_t size) {
2571 static volatile jint cnt = 0;
2572
2573 if (!UseOprofile) {
2574 return;
2575 }
2576
2577 char buf[PATH_MAX+1];
2578 int num = Atomic::add(1, &cnt);
2579
2580 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2581 os::get_temp_directory(), os::current_process_id(), num);
2582 unlink(buf);
2583
2584 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2585
2586 if (fd != -1) {
2587 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2588 if (rv != (off_t)-1) {
2589 if (::write(fd, "", 1) == 1) {
2590 mmap(base, size,
2591 PROT_READ|PROT_WRITE|PROT_EXEC,
2592 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2593 }
2594 }
2595 ::close(fd);
2596 unlink(buf);
2597 }
2598 }
2599
2600 static bool recoverable_mmap_error(int err) {
2601 // See if the error is one we can let the caller handle. This
2602 // list of errno values comes from JBS-6843484. I can't find a
2603 // Linux man page that documents this specific set of errno
2604 // values so while this list currently matches Solaris, it may
2605 // change as we gain experience with this failure mode.
2606 switch (err) {
2607 case EBADF:
2608 case EINVAL:
2609 case ENOTSUP:
2610 // let the caller deal with these errors
2611 return true;
2612
2613 default:
2614 // Any remaining errors on this OS can cause our reserved mapping
2615 // to be lost. That can cause confusion where different data
2616 // structures think they have the same memory mapped. The worst
2617 // scenario is if both the VM and a library think they have the
2618 // same memory mapped.
2619 return false;
2620 }
2621 }
2622
2623 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2624 int err) {
2625 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2626 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2627 os::strerror(err), err);
2628 }
2629
2630 static void warn_fail_commit_memory(char* addr, size_t size,
2631 size_t alignment_hint, bool exec,
2632 int err) {
2633 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2634 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2635 alignment_hint, exec, os::strerror(err), err);
2636 }
2637
2638 // NOTE: Linux kernel does not really reserve the pages for us.
2639 // All it does is to check if there are enough free pages
2640 // left at the time of mmap(). This could be a potential
2641 // problem.
2642 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2643 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2644 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2645 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2646 if (res != (uintptr_t) MAP_FAILED) {
2647 if (UseNUMAInterleaving) {
2648 numa_make_global(addr, size);
2649 }
2650 return 0;
2651 }
2652
2653 int err = errno; // save errno from mmap() call above
2654
2655 if (!recoverable_mmap_error(err)) {
2656 warn_fail_commit_memory(addr, size, exec, err);
2657 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2658 }
2659
2660 return err;
2661 }
2662
2663 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2664 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2665 }
2666
2667 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2668 const char* mesg) {
2669 assert(mesg != NULL, "mesg must be specified");
2670 int err = os::Linux::commit_memory_impl(addr, size, exec);
2671 if (err != 0) {
2672 // the caller wants all commit errors to exit with the specified mesg:
2673 warn_fail_commit_memory(addr, size, exec, err);
2674 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2675 }
2676 }
2677
2678 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2679 #ifndef MAP_HUGETLB
2680 #define MAP_HUGETLB 0x40000
2681 #endif
2682
2683 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2684 #ifndef MADV_HUGEPAGE
2685 #define MADV_HUGEPAGE 14
2686 #endif
2687
2688 int os::Linux::commit_memory_impl(char* addr, size_t size,
2689 size_t alignment_hint, bool exec) {
2690 int err = os::Linux::commit_memory_impl(addr, size, exec);
2691 if (err == 0) {
2692 realign_memory(addr, size, alignment_hint);
2693 }
2694 return err;
2695 }
2696
2697 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2698 bool exec) {
2699 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2700 }
2701
2702 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2703 size_t alignment_hint, bool exec,
2704 const char* mesg) {
2705 assert(mesg != NULL, "mesg must be specified");
2706 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2707 if (err != 0) {
2708 // the caller wants all commit errors to exit with the specified mesg:
2709 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2710 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2711 }
2712 }
2713
2714 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2715 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2716 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2717 // be supported or the memory may already be backed by huge pages.
2718 ::madvise(addr, bytes, MADV_HUGEPAGE);
2719 }
2720 }
2721
2722 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2723 // This method works by doing an mmap over an existing mmaping and effectively discarding
2724 // the existing pages. However it won't work for SHM-based large pages that cannot be
2725 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2726 // small pages on top of the SHM segment. This method always works for small pages, so we
2727 // allow that in any case.
2728 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2729 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2730 }
2731 }
2732
2733 void os::numa_make_global(char *addr, size_t bytes) {
2734 Linux::numa_interleave_memory(addr, bytes);
2735 }
2736
2737 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2738 // bind policy to MPOL_PREFERRED for the current thread.
2739 #define USE_MPOL_PREFERRED 0
2740
2741 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2742 // To make NUMA and large pages more robust when both enabled, we need to ease
2743 // the requirements on where the memory should be allocated. MPOL_BIND is the
2744 // default policy and it will force memory to be allocated on the specified
2745 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2746 // the specified node, but will not force it. Using this policy will prevent
2747 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2748 // free large pages.
2749 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2750 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2751 }
2752
2753 bool os::numa_topology_changed() { return false; }
2754
2755 size_t os::numa_get_groups_num() {
2756 // Return just the number of nodes in which it's possible to allocate memory
2757 // (in numa terminology, configured nodes).
2758 return Linux::numa_num_configured_nodes();
2759 }
2760
2761 int os::numa_get_group_id() {
2762 int cpu_id = Linux::sched_getcpu();
2763 if (cpu_id != -1) {
2764 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2765 if (lgrp_id != -1) {
2766 return lgrp_id;
2767 }
2768 }
2769 return 0;
2770 }
2771
2772 int os::Linux::get_existing_num_nodes() {
2773 int node;
2774 int highest_node_number = Linux::numa_max_node();
2775 int num_nodes = 0;
2776
2777 // Get the total number of nodes in the system including nodes without memory.
2778 for (node = 0; node <= highest_node_number; node++) {
2779 if (isnode_in_existing_nodes(node)) {
2780 num_nodes++;
2781 }
2782 }
2783 return num_nodes;
2784 }
2785
2786 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2787 int highest_node_number = Linux::numa_max_node();
2788 size_t i = 0;
2789
2790 // Map all node ids in which it is possible to allocate memory. Also nodes are
2791 // not always consecutively available, i.e. available from 0 to the highest
2792 // node number. If the nodes have been bound explicitly using numactl membind,
2793 // then allocate memory from those nodes only.
2794 for (int node = 0; node <= highest_node_number; node++) {
2795 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2796 ids[i++] = node;
2797 }
2798 }
2799 return i;
2800 }
2801
2802 bool os::get_page_info(char *start, page_info* info) {
2803 return false;
2804 }
2805
2806 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2807 page_info* page_found) {
2808 return end;
2809 }
2810
2811
2812 int os::Linux::sched_getcpu_syscall(void) {
2813 unsigned int cpu = 0;
2814 int retval = -1;
2815
2816 #if defined(IA32)
2817 #ifndef SYS_getcpu
2818 #define SYS_getcpu 318
2819 #endif
2820 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2821 #elif defined(AMD64)
2822 // Unfortunately we have to bring all these macros here from vsyscall.h
2823 // to be able to compile on old linuxes.
2824 #define __NR_vgetcpu 2
2825 #define VSYSCALL_START (-10UL << 20)
2826 #define VSYSCALL_SIZE 1024
2827 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2828 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2829 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2830 retval = vgetcpu(&cpu, NULL, NULL);
2831 #endif
2832
2833 return (retval == -1) ? retval : cpu;
2834 }
2835
2836 void os::Linux::sched_getcpu_init() {
2837 // sched_getcpu() should be in libc.
2838 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2839 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2840
2841 // If it's not, try a direct syscall.
2842 if (sched_getcpu() == -1) {
2843 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2844 (void*)&sched_getcpu_syscall));
2845 }
2846
2847 if (sched_getcpu() == -1) {
2848 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2849 }
2850 }
2851
2852 // Something to do with the numa-aware allocator needs these symbols
2853 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2854 extern "C" JNIEXPORT void numa_error(char *where) { }
2855
2856 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2857 // load symbol from base version instead.
2858 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2859 void *f = dlvsym(handle, name, "libnuma_1.1");
2860 if (f == NULL) {
2861 f = dlsym(handle, name);
2862 }
2863 return f;
2864 }
2865
2866 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2867 // Return NULL if the symbol is not defined in this particular version.
2868 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2869 return dlvsym(handle, name, "libnuma_1.2");
2870 }
2871
2872 bool os::Linux::libnuma_init() {
2873 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2874 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2875 if (handle != NULL) {
2876 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2877 libnuma_dlsym(handle, "numa_node_to_cpus")));
2878 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2879 libnuma_dlsym(handle, "numa_max_node")));
2880 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2881 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2882 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2883 libnuma_dlsym(handle, "numa_available")));
2884 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2885 libnuma_dlsym(handle, "numa_tonode_memory")));
2886 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2887 libnuma_dlsym(handle, "numa_interleave_memory")));
2888 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2889 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2890 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2891 libnuma_dlsym(handle, "numa_set_bind_policy")));
2892 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2893 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2894 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2895 libnuma_dlsym(handle, "numa_distance")));
2896 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2897 libnuma_v2_dlsym(handle, "numa_get_membind")));
2898
2899 if (numa_available() != -1) {
2900 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2901 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2902 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2903 // Create an index -> node mapping, since nodes are not always consecutive
2904 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2905 rebuild_nindex_to_node_map();
2906 // Create a cpu -> node mapping
2907 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2908 rebuild_cpu_to_node_map();
2909 return true;
2910 }
2911 }
2912 }
2913 return false;
2914 }
2915
2916 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2917 // Creating guard page is very expensive. Java thread has HotSpot
2918 // guard pages, only enable glibc guard page for non-Java threads.
2919 // (Remember: compiler thread is a Java thread, too!)
2920 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2921 }
2922
2923 void os::Linux::rebuild_nindex_to_node_map() {
2924 int highest_node_number = Linux::numa_max_node();
2925
2926 nindex_to_node()->clear();
2927 for (int node = 0; node <= highest_node_number; node++) {
2928 if (Linux::isnode_in_existing_nodes(node)) {
2929 nindex_to_node()->append(node);
2930 }
2931 }
2932 }
2933
2934 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2935 // The table is later used in get_node_by_cpu().
2936 void os::Linux::rebuild_cpu_to_node_map() {
2937 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2938 // in libnuma (possible values are starting from 16,
2939 // and continuing up with every other power of 2, but less
2940 // than the maximum number of CPUs supported by kernel), and
2941 // is a subject to change (in libnuma version 2 the requirements
2942 // are more reasonable) we'll just hardcode the number they use
2943 // in the library.
2944 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2945
2946 size_t cpu_num = processor_count();
2947 size_t cpu_map_size = NCPUS / BitsPerCLong;
2948 size_t cpu_map_valid_size =
2949 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2950
2951 cpu_to_node()->clear();
2952 cpu_to_node()->at_grow(cpu_num - 1);
2953
2954 size_t node_num = get_existing_num_nodes();
2955
2956 int distance = 0;
2957 int closest_distance = INT_MAX;
2958 int closest_node = 0;
2959 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2960 for (size_t i = 0; i < node_num; i++) {
2961 // Check if node is configured (not a memory-less node). If it is not, find
2962 // the closest configured node. Check also if node is bound, i.e. it's allowed
2963 // to allocate memory from the node. If it's not allowed, map cpus in that node
2964 // to the closest node from which memory allocation is allowed.
2965 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
2966 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
2967 closest_distance = INT_MAX;
2968 // Check distance from all remaining nodes in the system. Ignore distance
2969 // from itself, from another non-configured node, and from another non-bound
2970 // node.
2971 for (size_t m = 0; m < node_num; m++) {
2972 if (m != i &&
2973 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
2974 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
2975 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2976 // If a closest node is found, update. There is always at least one
2977 // configured and bound node in the system so there is always at least
2978 // one node close.
2979 if (distance != 0 && distance < closest_distance) {
2980 closest_distance = distance;
2981 closest_node = nindex_to_node()->at(m);
2982 }
2983 }
2984 }
2985 } else {
2986 // Current node is already a configured node.
2987 closest_node = nindex_to_node()->at(i);
2988 }
2989
2990 // Get cpus from the original node and map them to the closest node. If node
2991 // is a configured node (not a memory-less node), then original node and
2992 // closest node are the same.
2993 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2994 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2995 if (cpu_map[j] != 0) {
2996 for (size_t k = 0; k < BitsPerCLong; k++) {
2997 if (cpu_map[j] & (1UL << k)) {
2998 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2999 }
3000 }
3001 }
3002 }
3003 }
3004 }
3005 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3006 }
3007
3008 int os::Linux::get_node_by_cpu(int cpu_id) {
3009 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3010 return cpu_to_node()->at(cpu_id);
3011 }
3012 return -1;
3013 }
3014
3015 GrowableArray<int>* os::Linux::_cpu_to_node;
3016 GrowableArray<int>* os::Linux::_nindex_to_node;
3017 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3018 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3019 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3020 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3021 os::Linux::numa_available_func_t os::Linux::_numa_available;
3022 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3023 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3024 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3025 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3026 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3027 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3028 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3029 unsigned long* os::Linux::_numa_all_nodes;
3030 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3031 struct bitmask* os::Linux::_numa_nodes_ptr;
3032
3033 bool os::pd_uncommit_memory(char* addr, size_t size) {
3034 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3035 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3036 return res != (uintptr_t) MAP_FAILED;
3037 }
3038
3039 static address get_stack_commited_bottom(address bottom, size_t size) {
3040 address nbot = bottom;
3041 address ntop = bottom + size;
3042
3043 size_t page_sz = os::vm_page_size();
3044 unsigned pages = size / page_sz;
3045
3046 unsigned char vec[1];
3047 unsigned imin = 1, imax = pages + 1, imid;
3048 int mincore_return_value = 0;
3049
3050 assert(imin <= imax, "Unexpected page size");
3051
3052 while (imin < imax) {
3053 imid = (imax + imin) / 2;
3054 nbot = ntop - (imid * page_sz);
3055
3056 // Use a trick with mincore to check whether the page is mapped or not.
3057 // mincore sets vec to 1 if page resides in memory and to 0 if page
3058 // is swapped output but if page we are asking for is unmapped
3059 // it returns -1,ENOMEM
3060 mincore_return_value = mincore(nbot, page_sz, vec);
3061
3062 if (mincore_return_value == -1) {
3063 // Page is not mapped go up
3064 // to find first mapped page
3065 if (errno != EAGAIN) {
3066 assert(errno == ENOMEM, "Unexpected mincore errno");
3067 imax = imid;
3068 }
3069 } else {
3070 // Page is mapped go down
3071 // to find first not mapped page
3072 imin = imid + 1;
3073 }
3074 }
3075
3076 nbot = nbot + page_sz;
3077
3078 // Adjust stack bottom one page up if last checked page is not mapped
3079 if (mincore_return_value == -1) {
3080 nbot = nbot + page_sz;
3081 }
3082
3083 return nbot;
3084 }
3085
3086 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3087 int mincore_return_value;
3088 const size_t stripe = 1024; // query this many pages each time
3089 unsigned char vec[stripe + 1];
3090 // set a guard
3091 vec[stripe] = 'X';
3092
3093 const size_t page_sz = os::vm_page_size();
3094 size_t pages = size / page_sz;
3095
3096 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3097 assert(is_aligned(size, page_sz), "Size must be page aligned");
3098
3099 committed_start = NULL;
3100
3101 int loops = (pages + stripe - 1) / stripe;
3102 int committed_pages = 0;
3103 address loop_base = start;
3104 bool found_range = false;
3105
3106 for (int index = 0; index < loops && !found_range; index ++) {
3107 assert(pages > 0, "Nothing to do");
3108 int pages_to_query = (pages >= stripe) ? stripe : pages;
3109 pages -= pages_to_query;
3110
3111 // Get stable read
3112 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3113
3114 // During shutdown, some memory goes away without properly notifying NMT,
3115 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3116 // Bailout and return as not committed for now.
3117 if (mincore_return_value == -1 && errno == ENOMEM) {
3118 return false;
3119 }
3120
3121 assert(vec[stripe] == 'X', "overflow guard");
3122 assert(mincore_return_value == 0, "Range must be valid");
3123 // Process this stripe
3124 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3125 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3126 // End of current contiguous region
3127 if (committed_start != NULL) {
3128 found_range = true;
3129 break;
3130 }
3131 } else { // committed
3132 // Start of region
3133 if (committed_start == NULL) {
3134 committed_start = loop_base + page_sz * vecIdx;
3135 }
3136 committed_pages ++;
3137 }
3138 }
3139
3140 loop_base += pages_to_query * page_sz;
3141 }
3142
3143 if (committed_start != NULL) {
3144 assert(committed_pages > 0, "Must have committed region");
3145 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3146 assert(committed_start >= start && committed_start < start + size, "Out of range");
3147 committed_size = page_sz * committed_pages;
3148 return true;
3149 } else {
3150 assert(committed_pages == 0, "Should not have committed region");
3151 return false;
3152 }
3153 }
3154
3155
3156 // Linux uses a growable mapping for the stack, and if the mapping for
3157 // the stack guard pages is not removed when we detach a thread the
3158 // stack cannot grow beyond the pages where the stack guard was
3159 // mapped. If at some point later in the process the stack expands to
3160 // that point, the Linux kernel cannot expand the stack any further
3161 // because the guard pages are in the way, and a segfault occurs.
3162 //
3163 // However, it's essential not to split the stack region by unmapping
3164 // a region (leaving a hole) that's already part of the stack mapping,
3165 // so if the stack mapping has already grown beyond the guard pages at
3166 // the time we create them, we have to truncate the stack mapping.
3167 // So, we need to know the extent of the stack mapping when
3168 // create_stack_guard_pages() is called.
3169
3170 // We only need this for stacks that are growable: at the time of
3171 // writing thread stacks don't use growable mappings (i.e. those
3172 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3173 // only applies to the main thread.
3174
3175 // If the (growable) stack mapping already extends beyond the point
3176 // where we're going to put our guard pages, truncate the mapping at
3177 // that point by munmap()ping it. This ensures that when we later
3178 // munmap() the guard pages we don't leave a hole in the stack
3179 // mapping. This only affects the main/primordial thread
3180
3181 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3182 if (os::is_primordial_thread()) {
3183 // As we manually grow stack up to bottom inside create_attached_thread(),
3184 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3185 // we don't need to do anything special.
3186 // Check it first, before calling heavy function.
3187 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3188 unsigned char vec[1];
3189
3190 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3191 // Fallback to slow path on all errors, including EAGAIN
3192 stack_extent = (uintptr_t) get_stack_commited_bottom(
3193 os::Linux::initial_thread_stack_bottom(),
3194 (size_t)addr - stack_extent);
3195 }
3196
3197 if (stack_extent < (uintptr_t)addr) {
3198 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3199 }
3200 }
3201
3202 return os::commit_memory(addr, size, !ExecMem);
3203 }
3204
3205 // If this is a growable mapping, remove the guard pages entirely by
3206 // munmap()ping them. If not, just call uncommit_memory(). This only
3207 // affects the main/primordial thread, but guard against future OS changes.
3208 // It's safe to always unmap guard pages for primordial thread because we
3209 // always place it right after end of the mapped region.
3210
3211 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3212 uintptr_t stack_extent, stack_base;
3213
3214 if (os::is_primordial_thread()) {
3215 return ::munmap(addr, size) == 0;
3216 }
3217
3218 return os::uncommit_memory(addr, size);
3219 }
3220
3221 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3222 // at 'requested_addr'. If there are existing memory mappings at the same
3223 // location, however, they will be overwritten. If 'fixed' is false,
3224 // 'requested_addr' is only treated as a hint, the return value may or
3225 // may not start from the requested address. Unlike Linux mmap(), this
3226 // function returns NULL to indicate failure.
3227 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3228 char * addr;
3229 int flags;
3230
3231 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3232 if (fixed) {
3233 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3234 flags |= MAP_FIXED;
3235 }
3236
3237 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3238 // touch an uncommitted page. Otherwise, the read/write might
3239 // succeed if we have enough swap space to back the physical page.
3240 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3241 flags, -1, 0);
3242
3243 return addr == MAP_FAILED ? NULL : addr;
3244 }
3245
3246 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3247 // (req_addr != NULL) or with a given alignment.
3248 // - bytes shall be a multiple of alignment.
3249 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3250 // - alignment sets the alignment at which memory shall be allocated.
3251 // It must be a multiple of allocation granularity.
3252 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3253 // req_addr or NULL.
3254 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3255
3256 size_t extra_size = bytes;
3257 if (req_addr == NULL && alignment > 0) {
3258 extra_size += alignment;
3259 }
3260
3261 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3262 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3263 -1, 0);
3264 if (start == MAP_FAILED) {
3265 start = NULL;
3266 } else {
3267 if (req_addr != NULL) {
3268 if (start != req_addr) {
3269 ::munmap(start, extra_size);
3270 start = NULL;
3271 }
3272 } else {
3273 char* const start_aligned = align_up(start, alignment);
3274 char* const end_aligned = start_aligned + bytes;
3275 char* const end = start + extra_size;
3276 if (start_aligned > start) {
3277 ::munmap(start, start_aligned - start);
3278 }
3279 if (end_aligned < end) {
3280 ::munmap(end_aligned, end - end_aligned);
3281 }
3282 start = start_aligned;
3283 }
3284 }
3285 return start;
3286 }
3287
3288 static int anon_munmap(char * addr, size_t size) {
3289 return ::munmap(addr, size) == 0;
3290 }
3291
3292 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3293 size_t alignment_hint) {
3294 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3295 }
3296
3297 bool os::pd_release_memory(char* addr, size_t size) {
3298 return anon_munmap(addr, size);
3299 }
3300
3301 static bool linux_mprotect(char* addr, size_t size, int prot) {
3302 // Linux wants the mprotect address argument to be page aligned.
3303 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3304
3305 // According to SUSv3, mprotect() should only be used with mappings
3306 // established by mmap(), and mmap() always maps whole pages. Unaligned
3307 // 'addr' likely indicates problem in the VM (e.g. trying to change
3308 // protection of malloc'ed or statically allocated memory). Check the
3309 // caller if you hit this assert.
3310 assert(addr == bottom, "sanity check");
3311
3312 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3313 return ::mprotect(bottom, size, prot) == 0;
3314 }
3315
3316 // Set protections specified
3317 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3318 bool is_committed) {
3319 unsigned int p = 0;
3320 switch (prot) {
3321 case MEM_PROT_NONE: p = PROT_NONE; break;
3322 case MEM_PROT_READ: p = PROT_READ; break;
3323 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3324 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3325 default:
3326 ShouldNotReachHere();
3327 }
3328 // is_committed is unused.
3329 return linux_mprotect(addr, bytes, p);
3330 }
3331
3332 bool os::guard_memory(char* addr, size_t size) {
3333 return linux_mprotect(addr, size, PROT_NONE);
3334 }
3335
3336 bool os::unguard_memory(char* addr, size_t size) {
3337 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3338 }
3339
3340 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3341 size_t page_size) {
3342 bool result = false;
3343 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3344 MAP_ANONYMOUS|MAP_PRIVATE,
3345 -1, 0);
3346 if (p != MAP_FAILED) {
3347 void *aligned_p = align_up(p, page_size);
3348
3349 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3350
3351 munmap(p, page_size * 2);
3352 }
3353
3354 if (warn && !result) {
3355 warning("TransparentHugePages is not supported by the operating system.");
3356 }
3357
3358 return result;
3359 }
3360
3361 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3362 bool result = false;
3363 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3364 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3365 -1, 0);
3366
3367 if (p != MAP_FAILED) {
3368 // We don't know if this really is a huge page or not.
3369 FILE *fp = fopen("/proc/self/maps", "r");
3370 if (fp) {
3371 while (!feof(fp)) {
3372 char chars[257];
3373 long x = 0;
3374 if (fgets(chars, sizeof(chars), fp)) {
3375 if (sscanf(chars, "%lx-%*x", &x) == 1
3376 && x == (long)p) {
3377 if (strstr (chars, "hugepage")) {
3378 result = true;
3379 break;
3380 }
3381 }
3382 }
3383 }
3384 fclose(fp);
3385 }
3386 munmap(p, page_size);
3387 }
3388
3389 if (warn && !result) {
3390 warning("HugeTLBFS is not supported by the operating system.");
3391 }
3392
3393 return result;
3394 }
3395
3396 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3397 //
3398 // From the coredump_filter documentation:
3399 //
3400 // - (bit 0) anonymous private memory
3401 // - (bit 1) anonymous shared memory
3402 // - (bit 2) file-backed private memory
3403 // - (bit 3) file-backed shared memory
3404 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3405 // effective only if the bit 2 is cleared)
3406 // - (bit 5) hugetlb private memory
3407 // - (bit 6) hugetlb shared memory
3408 // - (bit 7) dax private memory
3409 // - (bit 8) dax shared memory
3410 //
3411 static void set_coredump_filter(CoredumpFilterBit bit) {
3412 FILE *f;
3413 long cdm;
3414
3415 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3416 return;
3417 }
3418
3419 if (fscanf(f, "%lx", &cdm) != 1) {
3420 fclose(f);
3421 return;
3422 }
3423
3424 long saved_cdm = cdm;
3425 rewind(f);
3426 cdm |= bit;
3427
3428 if (cdm != saved_cdm) {
3429 fprintf(f, "%#lx", cdm);
3430 }
3431
3432 fclose(f);
3433 }
3434
3435 // Large page support
3436
3437 static size_t _large_page_size = 0;
3438
3439 size_t os::Linux::find_large_page_size() {
3440 size_t large_page_size = 0;
3441
3442 // large_page_size on Linux is used to round up heap size. x86 uses either
3443 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3444 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3445 // page as large as 256M.
3446 //
3447 // Here we try to figure out page size by parsing /proc/meminfo and looking
3448 // for a line with the following format:
3449 // Hugepagesize: 2048 kB
3450 //
3451 // If we can't determine the value (e.g. /proc is not mounted, or the text
3452 // format has been changed), we'll use the largest page size supported by
3453 // the processor.
3454
3455 #ifndef ZERO
3456 large_page_size =
3457 AARCH64_ONLY(2 * M)
3458 AMD64_ONLY(2 * M)
3459 ARM32_ONLY(2 * M)
3460 IA32_ONLY(4 * M)
3461 IA64_ONLY(256 * M)
3462 PPC_ONLY(4 * M)
3463 S390_ONLY(1 * M)
3464 SPARC_ONLY(4 * M);
3465 #endif // ZERO
3466
3467 FILE *fp = fopen("/proc/meminfo", "r");
3468 if (fp) {
3469 while (!feof(fp)) {
3470 int x = 0;
3471 char buf[16];
3472 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3473 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3474 large_page_size = x * K;
3475 break;
3476 }
3477 } else {
3478 // skip to next line
3479 for (;;) {
3480 int ch = fgetc(fp);
3481 if (ch == EOF || ch == (int)'\n') break;
3482 }
3483 }
3484 }
3485 fclose(fp);
3486 }
3487
3488 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3489 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3490 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3491 proper_unit_for_byte_size(large_page_size));
3492 }
3493
3494 return large_page_size;
3495 }
3496
3497 size_t os::Linux::setup_large_page_size() {
3498 _large_page_size = Linux::find_large_page_size();
3499 const size_t default_page_size = (size_t)Linux::page_size();
3500 if (_large_page_size > default_page_size) {
3501 _page_sizes[0] = _large_page_size;
3502 _page_sizes[1] = default_page_size;
3503 _page_sizes[2] = 0;
3504 }
3505
3506 return _large_page_size;
3507 }
3508
3509 bool os::Linux::setup_large_page_type(size_t page_size) {
3510 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3511 FLAG_IS_DEFAULT(UseSHM) &&
3512 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3513
3514 // The type of large pages has not been specified by the user.
3515
3516 // Try UseHugeTLBFS and then UseSHM.
3517 UseHugeTLBFS = UseSHM = true;
3518
3519 // Don't try UseTransparentHugePages since there are known
3520 // performance issues with it turned on. This might change in the future.
3521 UseTransparentHugePages = false;
3522 }
3523
3524 if (UseTransparentHugePages) {
3525 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3526 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3527 UseHugeTLBFS = false;
3528 UseSHM = false;
3529 return true;
3530 }
3531 UseTransparentHugePages = false;
3532 }
3533
3534 if (UseHugeTLBFS) {
3535 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3536 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3537 UseSHM = false;
3538 return true;
3539 }
3540 UseHugeTLBFS = false;
3541 }
3542
3543 return UseSHM;
3544 }
3545
3546 void os::large_page_init() {
3547 if (!UseLargePages &&
3548 !UseTransparentHugePages &&
3549 !UseHugeTLBFS &&
3550 !UseSHM) {
3551 // Not using large pages.
3552 return;
3553 }
3554
3555 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3556 // The user explicitly turned off large pages.
3557 // Ignore the rest of the large pages flags.
3558 UseTransparentHugePages = false;
3559 UseHugeTLBFS = false;
3560 UseSHM = false;
3561 return;
3562 }
3563
3564 size_t large_page_size = Linux::setup_large_page_size();
3565 UseLargePages = Linux::setup_large_page_type(large_page_size);
3566
3567 set_coredump_filter(LARGEPAGES_BIT);
3568 }
3569
3570 #ifndef SHM_HUGETLB
3571 #define SHM_HUGETLB 04000
3572 #endif
3573
3574 #define shm_warning_format(format, ...) \
3575 do { \
3576 if (UseLargePages && \
3577 (!FLAG_IS_DEFAULT(UseLargePages) || \
3578 !FLAG_IS_DEFAULT(UseSHM) || \
3579 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3580 warning(format, __VA_ARGS__); \
3581 } \
3582 } while (0)
3583
3584 #define shm_warning(str) shm_warning_format("%s", str)
3585
3586 #define shm_warning_with_errno(str) \
3587 do { \
3588 int err = errno; \
3589 shm_warning_format(str " (error = %d)", err); \
3590 } while (0)
3591
3592 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3593 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3594
3595 if (!is_aligned(alignment, SHMLBA)) {
3596 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3597 return NULL;
3598 }
3599
3600 // To ensure that we get 'alignment' aligned memory from shmat,
3601 // we pre-reserve aligned virtual memory and then attach to that.
3602
3603 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3604 if (pre_reserved_addr == NULL) {
3605 // Couldn't pre-reserve aligned memory.
3606 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3607 return NULL;
3608 }
3609
3610 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3611 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3612
3613 if ((intptr_t)addr == -1) {
3614 int err = errno;
3615 shm_warning_with_errno("Failed to attach shared memory.");
3616
3617 assert(err != EACCES, "Unexpected error");
3618 assert(err != EIDRM, "Unexpected error");
3619 assert(err != EINVAL, "Unexpected error");
3620
3621 // Since we don't know if the kernel unmapped the pre-reserved memory area
3622 // we can't unmap it, since that would potentially unmap memory that was
3623 // mapped from other threads.
3624 return NULL;
3625 }
3626
3627 return addr;
3628 }
3629
3630 static char* shmat_at_address(int shmid, char* req_addr) {
3631 if (!is_aligned(req_addr, SHMLBA)) {
3632 assert(false, "Requested address needs to be SHMLBA aligned");
3633 return NULL;
3634 }
3635
3636 char* addr = (char*)shmat(shmid, req_addr, 0);
3637
3638 if ((intptr_t)addr == -1) {
3639 shm_warning_with_errno("Failed to attach shared memory.");
3640 return NULL;
3641 }
3642
3643 return addr;
3644 }
3645
3646 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3647 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3648 if (req_addr != NULL) {
3649 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3650 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3651 return shmat_at_address(shmid, req_addr);
3652 }
3653
3654 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3655 // return large page size aligned memory addresses when req_addr == NULL.
3656 // However, if the alignment is larger than the large page size, we have
3657 // to manually ensure that the memory returned is 'alignment' aligned.
3658 if (alignment > os::large_page_size()) {
3659 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3660 return shmat_with_alignment(shmid, bytes, alignment);
3661 } else {
3662 return shmat_at_address(shmid, NULL);
3663 }
3664 }
3665
3666 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3667 char* req_addr, bool exec) {
3668 // "exec" is passed in but not used. Creating the shared image for
3669 // the code cache doesn't have an SHM_X executable permission to check.
3670 assert(UseLargePages && UseSHM, "only for SHM large pages");
3671 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3672 assert(is_aligned(req_addr, alignment), "Unaligned address");
3673
3674 if (!is_aligned(bytes, os::large_page_size())) {
3675 return NULL; // Fallback to small pages.
3676 }
3677
3678 // Create a large shared memory region to attach to based on size.
3679 // Currently, size is the total size of the heap.
3680 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3681 if (shmid == -1) {
3682 // Possible reasons for shmget failure:
3683 // 1. shmmax is too small for Java heap.
3684 // > check shmmax value: cat /proc/sys/kernel/shmmax
3685 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3686 // 2. not enough large page memory.
3687 // > check available large pages: cat /proc/meminfo
3688 // > increase amount of large pages:
3689 // echo new_value > /proc/sys/vm/nr_hugepages
3690 // Note 1: different Linux may use different name for this property,
3691 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3692 // Note 2: it's possible there's enough physical memory available but
3693 // they are so fragmented after a long run that they can't
3694 // coalesce into large pages. Try to reserve large pages when
3695 // the system is still "fresh".
3696 shm_warning_with_errno("Failed to reserve shared memory.");
3697 return NULL;
3698 }
3699
3700 // Attach to the region.
3701 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3702
3703 // Remove shmid. If shmat() is successful, the actual shared memory segment
3704 // will be deleted when it's detached by shmdt() or when the process
3705 // terminates. If shmat() is not successful this will remove the shared
3706 // segment immediately.
3707 shmctl(shmid, IPC_RMID, NULL);
3708
3709 return addr;
3710 }
3711
3712 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3713 int error) {
3714 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3715
3716 bool warn_on_failure = UseLargePages &&
3717 (!FLAG_IS_DEFAULT(UseLargePages) ||
3718 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3719 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3720
3721 if (warn_on_failure) {
3722 char msg[128];
3723 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3724 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3725 warning("%s", msg);
3726 }
3727 }
3728
3729 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3730 char* req_addr,
3731 bool exec) {
3732 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3733 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3734 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3735
3736 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3737 char* addr = (char*)::mmap(req_addr, bytes, prot,
3738 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3739 -1, 0);
3740
3741 if (addr == MAP_FAILED) {
3742 warn_on_large_pages_failure(req_addr, bytes, errno);
3743 return NULL;
3744 }
3745
3746 assert(is_aligned(addr, os::large_page_size()), "Must be");
3747
3748 return addr;
3749 }
3750
3751 // Reserve memory using mmap(MAP_HUGETLB).
3752 // - bytes shall be a multiple of alignment.
3753 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3754 // - alignment sets the alignment at which memory shall be allocated.
3755 // It must be a multiple of allocation granularity.
3756 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3757 // req_addr or NULL.
3758 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3759 size_t alignment,
3760 char* req_addr,
3761 bool exec) {
3762 size_t large_page_size = os::large_page_size();
3763 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3764
3765 assert(is_aligned(req_addr, alignment), "Must be");
3766 assert(is_aligned(bytes, alignment), "Must be");
3767
3768 // First reserve - but not commit - the address range in small pages.
3769 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3770
3771 if (start == NULL) {
3772 return NULL;
3773 }
3774
3775 assert(is_aligned(start, alignment), "Must be");
3776
3777 char* end = start + bytes;
3778
3779 // Find the regions of the allocated chunk that can be promoted to large pages.
3780 char* lp_start = align_up(start, large_page_size);
3781 char* lp_end = align_down(end, large_page_size);
3782
3783 size_t lp_bytes = lp_end - lp_start;
3784
3785 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3786
3787 if (lp_bytes == 0) {
3788 // The mapped region doesn't even span the start and the end of a large page.
3789 // Fall back to allocate a non-special area.
3790 ::munmap(start, end - start);
3791 return NULL;
3792 }
3793
3794 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3795
3796 void* result;
3797
3798 // Commit small-paged leading area.
3799 if (start != lp_start) {
3800 result = ::mmap(start, lp_start - start, prot,
3801 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3802 -1, 0);
3803 if (result == MAP_FAILED) {
3804 ::munmap(lp_start, end - lp_start);
3805 return NULL;
3806 }
3807 }
3808
3809 // Commit large-paged area.
3810 result = ::mmap(lp_start, lp_bytes, prot,
3811 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3812 -1, 0);
3813 if (result == MAP_FAILED) {
3814 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3815 // If the mmap above fails, the large pages region will be unmapped and we
3816 // have regions before and after with small pages. Release these regions.
3817 //
3818 // | mapped | unmapped | mapped |
3819 // ^ ^ ^ ^
3820 // start lp_start lp_end end
3821 //
3822 ::munmap(start, lp_start - start);
3823 ::munmap(lp_end, end - lp_end);
3824 return NULL;
3825 }
3826
3827 // Commit small-paged trailing area.
3828 if (lp_end != end) {
3829 result = ::mmap(lp_end, end - lp_end, prot,
3830 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3831 -1, 0);
3832 if (result == MAP_FAILED) {
3833 ::munmap(start, lp_end - start);
3834 return NULL;
3835 }
3836 }
3837
3838 return start;
3839 }
3840
3841 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3842 size_t alignment,
3843 char* req_addr,
3844 bool exec) {
3845 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3846 assert(is_aligned(req_addr, alignment), "Must be");
3847 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3848 assert(is_power_of_2(os::large_page_size()), "Must be");
3849 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3850
3851 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3852 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3853 } else {
3854 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3855 }
3856 }
3857
3858 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3859 char* req_addr, bool exec) {
3860 assert(UseLargePages, "only for large pages");
3861
3862 char* addr;
3863 if (UseSHM) {
3864 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3865 } else {
3866 assert(UseHugeTLBFS, "must be");
3867 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3868 }
3869
3870 if (addr != NULL) {
3871 if (UseNUMAInterleaving) {
3872 numa_make_global(addr, bytes);
3873 }
3874
3875 // The memory is committed
3876 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3877 }
3878
3879 return addr;
3880 }
3881
3882 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3883 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3884 return shmdt(base) == 0;
3885 }
3886
3887 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3888 return pd_release_memory(base, bytes);
3889 }
3890
3891 bool os::release_memory_special(char* base, size_t bytes) {
3892 bool res;
3893 if (MemTracker::tracking_level() > NMT_minimal) {
3894 Tracker tkr(Tracker::release);
3895 res = os::Linux::release_memory_special_impl(base, bytes);
3896 if (res) {
3897 tkr.record((address)base, bytes);
3898 }
3899
3900 } else {
3901 res = os::Linux::release_memory_special_impl(base, bytes);
3902 }
3903 return res;
3904 }
3905
3906 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3907 assert(UseLargePages, "only for large pages");
3908 bool res;
3909
3910 if (UseSHM) {
3911 res = os::Linux::release_memory_special_shm(base, bytes);
3912 } else {
3913 assert(UseHugeTLBFS, "must be");
3914 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3915 }
3916 return res;
3917 }
3918
3919 size_t os::large_page_size() {
3920 return _large_page_size;
3921 }
3922
3923 // With SysV SHM the entire memory region must be allocated as shared
3924 // memory.
3925 // HugeTLBFS allows application to commit large page memory on demand.
3926 // However, when committing memory with HugeTLBFS fails, the region
3927 // that was supposed to be committed will lose the old reservation
3928 // and allow other threads to steal that memory region. Because of this
3929 // behavior we can't commit HugeTLBFS memory.
3930 bool os::can_commit_large_page_memory() {
3931 return UseTransparentHugePages;
3932 }
3933
3934 bool os::can_execute_large_page_memory() {
3935 return UseTransparentHugePages || UseHugeTLBFS;
3936 }
3937
3938 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3939 assert(file_desc >= 0, "file_desc is not valid");
3940 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3941 if (result != NULL) {
3942 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3943 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3944 }
3945 }
3946 return result;
3947 }
3948
3949 // Reserve memory at an arbitrary address, only if that area is
3950 // available (and not reserved for something else).
3951
3952 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3953 const int max_tries = 10;
3954 char* base[max_tries];
3955 size_t size[max_tries];
3956 const size_t gap = 0x000000;
3957
3958 // Assert only that the size is a multiple of the page size, since
3959 // that's all that mmap requires, and since that's all we really know
3960 // about at this low abstraction level. If we need higher alignment,
3961 // we can either pass an alignment to this method or verify alignment
3962 // in one of the methods further up the call chain. See bug 5044738.
3963 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3964
3965 // Repeatedly allocate blocks until the block is allocated at the
3966 // right spot.
3967
3968 // Linux mmap allows caller to pass an address as hint; give it a try first,
3969 // if kernel honors the hint then we can return immediately.
3970 char * addr = anon_mmap(requested_addr, bytes, false);
3971 if (addr == requested_addr) {
3972 return requested_addr;
3973 }
3974
3975 if (addr != NULL) {
3976 // mmap() is successful but it fails to reserve at the requested address
3977 anon_munmap(addr, bytes);
3978 }
3979
3980 int i;
3981 for (i = 0; i < max_tries; ++i) {
3982 base[i] = reserve_memory(bytes);
3983
3984 if (base[i] != NULL) {
3985 // Is this the block we wanted?
3986 if (base[i] == requested_addr) {
3987 size[i] = bytes;
3988 break;
3989 }
3990
3991 // Does this overlap the block we wanted? Give back the overlapped
3992 // parts and try again.
3993
3994 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3995 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3996 unmap_memory(base[i], top_overlap);
3997 base[i] += top_overlap;
3998 size[i] = bytes - top_overlap;
3999 } else {
4000 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4001 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4002 unmap_memory(requested_addr, bottom_overlap);
4003 size[i] = bytes - bottom_overlap;
4004 } else {
4005 size[i] = bytes;
4006 }
4007 }
4008 }
4009 }
4010
4011 // Give back the unused reserved pieces.
4012
4013 for (int j = 0; j < i; ++j) {
4014 if (base[j] != NULL) {
4015 unmap_memory(base[j], size[j]);
4016 }
4017 }
4018
4019 if (i < max_tries) {
4020 return requested_addr;
4021 } else {
4022 return NULL;
4023 }
4024 }
4025
4026 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4027 return ::read(fd, buf, nBytes);
4028 }
4029
4030 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4031 return ::pread(fd, buf, nBytes, offset);
4032 }
4033
4034 // Short sleep, direct OS call.
4035 //
4036 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4037 // sched_yield(2) will actually give up the CPU:
4038 //
4039 // * Alone on this pariticular CPU, keeps running.
4040 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4041 // (pre 2.6.39).
4042 //
4043 // So calling this with 0 is an alternative.
4044 //
4045 void os::naked_short_sleep(jlong ms) {
4046 struct timespec req;
4047
4048 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4049 req.tv_sec = 0;
4050 if (ms > 0) {
4051 req.tv_nsec = (ms % 1000) * 1000000;
4052 } else {
4053 req.tv_nsec = 1;
4054 }
4055
4056 nanosleep(&req, NULL);
4057
4058 return;
4059 }
4060
4061 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4062 void os::infinite_sleep() {
4063 while (true) { // sleep forever ...
4064 ::sleep(100); // ... 100 seconds at a time
4065 }
4066 }
4067
4068 // Used to convert frequent JVM_Yield() to nops
4069 bool os::dont_yield() {
4070 return DontYieldALot;
4071 }
4072
4073 void os::naked_yield() {
4074 sched_yield();
4075 }
4076
4077 ////////////////////////////////////////////////////////////////////////////////
4078 // thread priority support
4079
4080 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4081 // only supports dynamic priority, static priority must be zero. For real-time
4082 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4083 // However, for large multi-threaded applications, SCHED_RR is not only slower
4084 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4085 // of 5 runs - Sep 2005).
4086 //
4087 // The following code actually changes the niceness of kernel-thread/LWP. It
4088 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4089 // not the entire user process, and user level threads are 1:1 mapped to kernel
4090 // threads. It has always been the case, but could change in the future. For
4091 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4092 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4093
4094 int os::java_to_os_priority[CriticalPriority + 1] = {
4095 19, // 0 Entry should never be used
4096
4097 4, // 1 MinPriority
4098 3, // 2
4099 2, // 3
4100
4101 1, // 4
4102 0, // 5 NormPriority
4103 -1, // 6
4104
4105 -2, // 7
4106 -3, // 8
4107 -4, // 9 NearMaxPriority
4108
4109 -5, // 10 MaxPriority
4110
4111 -5 // 11 CriticalPriority
4112 };
4113
4114 static int prio_init() {
4115 if (ThreadPriorityPolicy == 1) {
4116 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4117 // if effective uid is not root. Perhaps, a more elegant way of doing
4118 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4119 if (geteuid() != 0) {
4120 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4121 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4122 }
4123 ThreadPriorityPolicy = 0;
4124 }
4125 }
4126 if (UseCriticalJavaThreadPriority) {
4127 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4128 }
4129 return 0;
4130 }
4131
4132 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4133 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4134
4135 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4136 return (ret == 0) ? OS_OK : OS_ERR;
4137 }
4138
4139 OSReturn os::get_native_priority(const Thread* const thread,
4140 int *priority_ptr) {
4141 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4142 *priority_ptr = java_to_os_priority[NormPriority];
4143 return OS_OK;
4144 }
4145
4146 errno = 0;
4147 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4148 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4149 }
4150
4151 ////////////////////////////////////////////////////////////////////////////////
4152 // suspend/resume support
4153
4154 // The low-level signal-based suspend/resume support is a remnant from the
4155 // old VM-suspension that used to be for java-suspension, safepoints etc,
4156 // within hotspot. Currently used by JFR's OSThreadSampler
4157 //
4158 // The remaining code is greatly simplified from the more general suspension
4159 // code that used to be used.
4160 //
4161 // The protocol is quite simple:
4162 // - suspend:
4163 // - sends a signal to the target thread
4164 // - polls the suspend state of the osthread using a yield loop
4165 // - target thread signal handler (SR_handler) sets suspend state
4166 // and blocks in sigsuspend until continued
4167 // - resume:
4168 // - sets target osthread state to continue
4169 // - sends signal to end the sigsuspend loop in the SR_handler
4170 //
4171 // Note that the SR_lock plays no role in this suspend/resume protocol,
4172 // but is checked for NULL in SR_handler as a thread termination indicator.
4173 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4174 //
4175 // Note that resume_clear_context() and suspend_save_context() are needed
4176 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4177 // which in part is used by:
4178 // - Forte Analyzer: AsyncGetCallTrace()
4179 // - StackBanging: get_frame_at_stack_banging_point()
4180
4181 static void resume_clear_context(OSThread *osthread) {
4182 osthread->set_ucontext(NULL);
4183 osthread->set_siginfo(NULL);
4184 }
4185
4186 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4187 ucontext_t* context) {
4188 osthread->set_ucontext(context);
4189 osthread->set_siginfo(siginfo);
4190 }
4191
4192 // Handler function invoked when a thread's execution is suspended or
4193 // resumed. We have to be careful that only async-safe functions are
4194 // called here (Note: most pthread functions are not async safe and
4195 // should be avoided.)
4196 //
4197 // Note: sigwait() is a more natural fit than sigsuspend() from an
4198 // interface point of view, but sigwait() prevents the signal hander
4199 // from being run. libpthread would get very confused by not having
4200 // its signal handlers run and prevents sigwait()'s use with the
4201 // mutex granting granting signal.
4202 //
4203 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4204 //
4205 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4206 // Save and restore errno to avoid confusing native code with EINTR
4207 // after sigsuspend.
4208 int old_errno = errno;
4209
4210 Thread* thread = Thread::current_or_null_safe();
4211 assert(thread != NULL, "Missing current thread in SR_handler");
4212
4213 // On some systems we have seen signal delivery get "stuck" until the signal
4214 // mask is changed as part of thread termination. Check that the current thread
4215 // has not already terminated (via SR_lock()) - else the following assertion
4216 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4217 // destructor has completed.
4218
4219 if (thread->SR_lock() == NULL) {
4220 return;
4221 }
4222
4223 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4224
4225 OSThread* osthread = thread->osthread();
4226
4227 os::SuspendResume::State current = osthread->sr.state();
4228 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4229 suspend_save_context(osthread, siginfo, context);
4230
4231 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4232 os::SuspendResume::State state = osthread->sr.suspended();
4233 if (state == os::SuspendResume::SR_SUSPENDED) {
4234 sigset_t suspend_set; // signals for sigsuspend()
4235 sigemptyset(&suspend_set);
4236 // get current set of blocked signals and unblock resume signal
4237 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4238 sigdelset(&suspend_set, SR_signum);
4239
4240 sr_semaphore.signal();
4241 // wait here until we are resumed
4242 while (1) {
4243 sigsuspend(&suspend_set);
4244
4245 os::SuspendResume::State result = osthread->sr.running();
4246 if (result == os::SuspendResume::SR_RUNNING) {
4247 sr_semaphore.signal();
4248 break;
4249 }
4250 }
4251
4252 } else if (state == os::SuspendResume::SR_RUNNING) {
4253 // request was cancelled, continue
4254 } else {
4255 ShouldNotReachHere();
4256 }
4257
4258 resume_clear_context(osthread);
4259 } else if (current == os::SuspendResume::SR_RUNNING) {
4260 // request was cancelled, continue
4261 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4262 // ignore
4263 } else {
4264 // ignore
4265 }
4266
4267 errno = old_errno;
4268 }
4269
4270 static int SR_initialize() {
4271 struct sigaction act;
4272 char *s;
4273
4274 // Get signal number to use for suspend/resume
4275 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4276 int sig = ::strtol(s, 0, 10);
4277 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4278 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4279 SR_signum = sig;
4280 } else {
4281 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4282 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4283 }
4284 }
4285
4286 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4287 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4288
4289 sigemptyset(&SR_sigset);
4290 sigaddset(&SR_sigset, SR_signum);
4291
4292 // Set up signal handler for suspend/resume
4293 act.sa_flags = SA_RESTART|SA_SIGINFO;
4294 act.sa_handler = (void (*)(int)) SR_handler;
4295
4296 // SR_signum is blocked by default.
4297 // 4528190 - We also need to block pthread restart signal (32 on all
4298 // supported Linux platforms). Note that LinuxThreads need to block
4299 // this signal for all threads to work properly. So we don't have
4300 // to use hard-coded signal number when setting up the mask.
4301 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4302
4303 if (sigaction(SR_signum, &act, 0) == -1) {
4304 return -1;
4305 }
4306
4307 // Save signal flag
4308 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4309 return 0;
4310 }
4311
4312 static int sr_notify(OSThread* osthread) {
4313 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4314 assert_status(status == 0, status, "pthread_kill");
4315 return status;
4316 }
4317
4318 // "Randomly" selected value for how long we want to spin
4319 // before bailing out on suspending a thread, also how often
4320 // we send a signal to a thread we want to resume
4321 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4322 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4323
4324 // returns true on success and false on error - really an error is fatal
4325 // but this seems the normal response to library errors
4326 static bool do_suspend(OSThread* osthread) {
4327 assert(osthread->sr.is_running(), "thread should be running");
4328 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4329
4330 // mark as suspended and send signal
4331 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4332 // failed to switch, state wasn't running?
4333 ShouldNotReachHere();
4334 return false;
4335 }
4336
4337 if (sr_notify(osthread) != 0) {
4338 ShouldNotReachHere();
4339 }
4340
4341 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4342 while (true) {
4343 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4344 break;
4345 } else {
4346 // timeout
4347 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4348 if (cancelled == os::SuspendResume::SR_RUNNING) {
4349 return false;
4350 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4351 // make sure that we consume the signal on the semaphore as well
4352 sr_semaphore.wait();
4353 break;
4354 } else {
4355 ShouldNotReachHere();
4356 return false;
4357 }
4358 }
4359 }
4360
4361 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4362 return true;
4363 }
4364
4365 static void do_resume(OSThread* osthread) {
4366 assert(osthread->sr.is_suspended(), "thread should be suspended");
4367 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4368
4369 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4370 // failed to switch to WAKEUP_REQUEST
4371 ShouldNotReachHere();
4372 return;
4373 }
4374
4375 while (true) {
4376 if (sr_notify(osthread) == 0) {
4377 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4378 if (osthread->sr.is_running()) {
4379 return;
4380 }
4381 }
4382 } else {
4383 ShouldNotReachHere();
4384 }
4385 }
4386
4387 guarantee(osthread->sr.is_running(), "Must be running!");
4388 }
4389
4390 ///////////////////////////////////////////////////////////////////////////////////
4391 // signal handling (except suspend/resume)
4392
4393 // This routine may be used by user applications as a "hook" to catch signals.
4394 // The user-defined signal handler must pass unrecognized signals to this
4395 // routine, and if it returns true (non-zero), then the signal handler must
4396 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4397 // routine will never retun false (zero), but instead will execute a VM panic
4398 // routine kill the process.
4399 //
4400 // If this routine returns false, it is OK to call it again. This allows
4401 // the user-defined signal handler to perform checks either before or after
4402 // the VM performs its own checks. Naturally, the user code would be making
4403 // a serious error if it tried to handle an exception (such as a null check
4404 // or breakpoint) that the VM was generating for its own correct operation.
4405 //
4406 // This routine may recognize any of the following kinds of signals:
4407 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4408 // It should be consulted by handlers for any of those signals.
4409 //
4410 // The caller of this routine must pass in the three arguments supplied
4411 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4412 // field of the structure passed to sigaction(). This routine assumes that
4413 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4414 //
4415 // Note that the VM will print warnings if it detects conflicting signal
4416 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4417 //
4418 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4419 siginfo_t* siginfo,
4420 void* ucontext,
4421 int abort_if_unrecognized);
4422
4423 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4424 assert(info != NULL && uc != NULL, "it must be old kernel");
4425 int orig_errno = errno; // Preserve errno value over signal handler.
4426 JVM_handle_linux_signal(sig, info, uc, true);
4427 errno = orig_errno;
4428 }
4429
4430
4431 // This boolean allows users to forward their own non-matching signals
4432 // to JVM_handle_linux_signal, harmlessly.
4433 bool os::Linux::signal_handlers_are_installed = false;
4434
4435 // For signal-chaining
4436 struct sigaction sigact[NSIG];
4437 uint64_t sigs = 0;
4438 #if (64 < NSIG-1)
4439 #error "Not all signals can be encoded in sigs. Adapt its type!"
4440 #endif
4441 bool os::Linux::libjsig_is_loaded = false;
4442 typedef struct sigaction *(*get_signal_t)(int);
4443 get_signal_t os::Linux::get_signal_action = NULL;
4444
4445 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4446 struct sigaction *actp = NULL;
4447
4448 if (libjsig_is_loaded) {
4449 // Retrieve the old signal handler from libjsig
4450 actp = (*get_signal_action)(sig);
4451 }
4452 if (actp == NULL) {
4453 // Retrieve the preinstalled signal handler from jvm
4454 actp = get_preinstalled_handler(sig);
4455 }
4456
4457 return actp;
4458 }
4459
4460 static bool call_chained_handler(struct sigaction *actp, int sig,
4461 siginfo_t *siginfo, void *context) {
4462 // Call the old signal handler
4463 if (actp->sa_handler == SIG_DFL) {
4464 // It's more reasonable to let jvm treat it as an unexpected exception
4465 // instead of taking the default action.
4466 return false;
4467 } else if (actp->sa_handler != SIG_IGN) {
4468 if ((actp->sa_flags & SA_NODEFER) == 0) {
4469 // automaticlly block the signal
4470 sigaddset(&(actp->sa_mask), sig);
4471 }
4472
4473 sa_handler_t hand = NULL;
4474 sa_sigaction_t sa = NULL;
4475 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4476 // retrieve the chained handler
4477 if (siginfo_flag_set) {
4478 sa = actp->sa_sigaction;
4479 } else {
4480 hand = actp->sa_handler;
4481 }
4482
4483 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4484 actp->sa_handler = SIG_DFL;
4485 }
4486
4487 // try to honor the signal mask
4488 sigset_t oset;
4489 sigemptyset(&oset);
4490 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4491
4492 // call into the chained handler
4493 if (siginfo_flag_set) {
4494 (*sa)(sig, siginfo, context);
4495 } else {
4496 (*hand)(sig);
4497 }
4498
4499 // restore the signal mask
4500 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4501 }
4502 // Tell jvm's signal handler the signal is taken care of.
4503 return true;
4504 }
4505
4506 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4507 bool chained = false;
4508 // signal-chaining
4509 if (UseSignalChaining) {
4510 struct sigaction *actp = get_chained_signal_action(sig);
4511 if (actp != NULL) {
4512 chained = call_chained_handler(actp, sig, siginfo, context);
4513 }
4514 }
4515 return chained;
4516 }
4517
4518 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4519 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4520 return &sigact[sig];
4521 }
4522 return NULL;
4523 }
4524
4525 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4526 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4527 sigact[sig] = oldAct;
4528 sigs |= (uint64_t)1 << (sig-1);
4529 }
4530
4531 // for diagnostic
4532 int sigflags[NSIG];
4533
4534 int os::Linux::get_our_sigflags(int sig) {
4535 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4536 return sigflags[sig];
4537 }
4538
4539 void os::Linux::set_our_sigflags(int sig, int flags) {
4540 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4541 if (sig > 0 && sig < NSIG) {
4542 sigflags[sig] = flags;
4543 }
4544 }
4545
4546 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4547 // Check for overwrite.
4548 struct sigaction oldAct;
4549 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4550
4551 void* oldhand = oldAct.sa_sigaction
4552 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4553 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4554 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4555 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4556 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4557 if (AllowUserSignalHandlers || !set_installed) {
4558 // Do not overwrite; user takes responsibility to forward to us.
4559 return;
4560 } else if (UseSignalChaining) {
4561 // save the old handler in jvm
4562 save_preinstalled_handler(sig, oldAct);
4563 // libjsig also interposes the sigaction() call below and saves the
4564 // old sigaction on it own.
4565 } else {
4566 fatal("Encountered unexpected pre-existing sigaction handler "
4567 "%#lx for signal %d.", (long)oldhand, sig);
4568 }
4569 }
4570
4571 struct sigaction sigAct;
4572 sigfillset(&(sigAct.sa_mask));
4573 sigAct.sa_handler = SIG_DFL;
4574 if (!set_installed) {
4575 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4576 } else {
4577 sigAct.sa_sigaction = signalHandler;
4578 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4579 }
4580 // Save flags, which are set by ours
4581 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4582 sigflags[sig] = sigAct.sa_flags;
4583
4584 int ret = sigaction(sig, &sigAct, &oldAct);
4585 assert(ret == 0, "check");
4586
4587 void* oldhand2 = oldAct.sa_sigaction
4588 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4589 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4590 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4591 }
4592
4593 // install signal handlers for signals that HotSpot needs to
4594 // handle in order to support Java-level exception handling.
4595
4596 void os::Linux::install_signal_handlers() {
4597 if (!signal_handlers_are_installed) {
4598 signal_handlers_are_installed = true;
4599
4600 // signal-chaining
4601 typedef void (*signal_setting_t)();
4602 signal_setting_t begin_signal_setting = NULL;
4603 signal_setting_t end_signal_setting = NULL;
4604 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4605 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4606 if (begin_signal_setting != NULL) {
4607 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4608 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4609 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4610 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4611 libjsig_is_loaded = true;
4612 assert(UseSignalChaining, "should enable signal-chaining");
4613 }
4614 if (libjsig_is_loaded) {
4615 // Tell libjsig jvm is setting signal handlers
4616 (*begin_signal_setting)();
4617 }
4618
4619 set_signal_handler(SIGSEGV, true);
4620 set_signal_handler(SIGPIPE, true);
4621 set_signal_handler(SIGBUS, true);
4622 set_signal_handler(SIGILL, true);
4623 set_signal_handler(SIGFPE, true);
4624 #if defined(PPC64)
4625 set_signal_handler(SIGTRAP, true);
4626 #endif
4627 set_signal_handler(SIGXFSZ, true);
4628
4629 if (libjsig_is_loaded) {
4630 // Tell libjsig jvm finishes setting signal handlers
4631 (*end_signal_setting)();
4632 }
4633
4634 // We don't activate signal checker if libjsig is in place, we trust ourselves
4635 // and if UserSignalHandler is installed all bets are off.
4636 // Log that signal checking is off only if -verbose:jni is specified.
4637 if (CheckJNICalls) {
4638 if (libjsig_is_loaded) {
4639 if (PrintJNIResolving) {
4640 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4641 }
4642 check_signals = false;
4643 }
4644 if (AllowUserSignalHandlers) {
4645 if (PrintJNIResolving) {
4646 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4647 }
4648 check_signals = false;
4649 }
4650 }
4651 }
4652 }
4653
4654 // This is the fastest way to get thread cpu time on Linux.
4655 // Returns cpu time (user+sys) for any thread, not only for current.
4656 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4657 // It might work on 2.6.10+ with a special kernel/glibc patch.
4658 // For reference, please, see IEEE Std 1003.1-2004:
4659 // http://www.unix.org/single_unix_specification
4660
4661 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4662 struct timespec tp;
4663 int rc = os::Posix::clock_gettime(clockid, &tp);
4664 assert(rc == 0, "clock_gettime is expected to return 0 code");
4665
4666 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4667 }
4668
4669 void os::Linux::initialize_os_info() {
4670 assert(_os_version == 0, "OS info already initialized");
4671
4672 struct utsname _uname;
4673
4674 uint32_t major;
4675 uint32_t minor;
4676 uint32_t fix;
4677
4678 int rc;
4679
4680 // Kernel version is unknown if
4681 // verification below fails.
4682 _os_version = 0x01000000;
4683
4684 rc = uname(&_uname);
4685 if (rc != -1) {
4686
4687 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4688 if (rc == 3) {
4689
4690 if (major < 256 && minor < 256 && fix < 256) {
4691 // Kernel version format is as expected,
4692 // set it overriding unknown state.
4693 _os_version = (major << 16) |
4694 (minor << 8 ) |
4695 (fix << 0 ) ;
4696 }
4697 }
4698 }
4699 }
4700
4701 uint32_t os::Linux::os_version() {
4702 assert(_os_version != 0, "not initialized");
4703 return _os_version & 0x00FFFFFF;
4704 }
4705
4706 bool os::Linux::os_version_is_known() {
4707 assert(_os_version != 0, "not initialized");
4708 return _os_version & 0x01000000 ? false : true;
4709 }
4710
4711 /////
4712 // glibc on Linux platform uses non-documented flag
4713 // to indicate, that some special sort of signal
4714 // trampoline is used.
4715 // We will never set this flag, and we should
4716 // ignore this flag in our diagnostic
4717 #ifdef SIGNIFICANT_SIGNAL_MASK
4718 #undef SIGNIFICANT_SIGNAL_MASK
4719 #endif
4720 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4721
4722 static const char* get_signal_handler_name(address handler,
4723 char* buf, int buflen) {
4724 int offset = 0;
4725 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4726 if (found) {
4727 // skip directory names
4728 const char *p1, *p2;
4729 p1 = buf;
4730 size_t len = strlen(os::file_separator());
4731 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4732 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4733 } else {
4734 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4735 }
4736 return buf;
4737 }
4738
4739 static void print_signal_handler(outputStream* st, int sig,
4740 char* buf, size_t buflen) {
4741 struct sigaction sa;
4742
4743 sigaction(sig, NULL, &sa);
4744
4745 // See comment for SIGNIFICANT_SIGNAL_MASK define
4746 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4747
4748 st->print("%s: ", os::exception_name(sig, buf, buflen));
4749
4750 address handler = (sa.sa_flags & SA_SIGINFO)
4751 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4752 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4753
4754 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4755 st->print("SIG_DFL");
4756 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4757 st->print("SIG_IGN");
4758 } else {
4759 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4760 }
4761
4762 st->print(", sa_mask[0]=");
4763 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4764
4765 address rh = VMError::get_resetted_sighandler(sig);
4766 // May be, handler was resetted by VMError?
4767 if (rh != NULL) {
4768 handler = rh;
4769 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4770 }
4771
4772 st->print(", sa_flags=");
4773 os::Posix::print_sa_flags(st, sa.sa_flags);
4774
4775 // Check: is it our handler?
4776 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4777 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4778 // It is our signal handler
4779 // check for flags, reset system-used one!
4780 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4781 st->print(
4782 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4783 os::Linux::get_our_sigflags(sig));
4784 }
4785 }
4786 st->cr();
4787 }
4788
4789
4790 #define DO_SIGNAL_CHECK(sig) \
4791 do { \
4792 if (!sigismember(&check_signal_done, sig)) { \
4793 os::Linux::check_signal_handler(sig); \
4794 } \
4795 } while (0)
4796
4797 // This method is a periodic task to check for misbehaving JNI applications
4798 // under CheckJNI, we can add any periodic checks here
4799
4800 void os::run_periodic_checks() {
4801 if (check_signals == false) return;
4802
4803 // SEGV and BUS if overridden could potentially prevent
4804 // generation of hs*.log in the event of a crash, debugging
4805 // such a case can be very challenging, so we absolutely
4806 // check the following for a good measure:
4807 DO_SIGNAL_CHECK(SIGSEGV);
4808 DO_SIGNAL_CHECK(SIGILL);
4809 DO_SIGNAL_CHECK(SIGFPE);
4810 DO_SIGNAL_CHECK(SIGBUS);
4811 DO_SIGNAL_CHECK(SIGPIPE);
4812 DO_SIGNAL_CHECK(SIGXFSZ);
4813 #if defined(PPC64)
4814 DO_SIGNAL_CHECK(SIGTRAP);
4815 #endif
4816
4817 // ReduceSignalUsage allows the user to override these handlers
4818 // see comments at the very top and jvm_md.h
4819 if (!ReduceSignalUsage) {
4820 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4821 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4822 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4823 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4824 }
4825
4826 DO_SIGNAL_CHECK(SR_signum);
4827 }
4828
4829 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4830
4831 static os_sigaction_t os_sigaction = NULL;
4832
4833 void os::Linux::check_signal_handler(int sig) {
4834 char buf[O_BUFLEN];
4835 address jvmHandler = NULL;
4836
4837
4838 struct sigaction act;
4839 if (os_sigaction == NULL) {
4840 // only trust the default sigaction, in case it has been interposed
4841 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4842 if (os_sigaction == NULL) return;
4843 }
4844
4845 os_sigaction(sig, (struct sigaction*)NULL, &act);
4846
4847
4848 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4849
4850 address thisHandler = (act.sa_flags & SA_SIGINFO)
4851 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4852 : CAST_FROM_FN_PTR(address, act.sa_handler);
4853
4854
4855 switch (sig) {
4856 case SIGSEGV:
4857 case SIGBUS:
4858 case SIGFPE:
4859 case SIGPIPE:
4860 case SIGILL:
4861 case SIGXFSZ:
4862 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4863 break;
4864
4865 case SHUTDOWN1_SIGNAL:
4866 case SHUTDOWN2_SIGNAL:
4867 case SHUTDOWN3_SIGNAL:
4868 case BREAK_SIGNAL:
4869 jvmHandler = (address)user_handler();
4870 break;
4871
4872 default:
4873 if (sig == SR_signum) {
4874 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4875 } else {
4876 return;
4877 }
4878 break;
4879 }
4880
4881 if (thisHandler != jvmHandler) {
4882 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4883 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4884 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4885 // No need to check this sig any longer
4886 sigaddset(&check_signal_done, sig);
4887 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4888 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4889 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4890 exception_name(sig, buf, O_BUFLEN));
4891 }
4892 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4893 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4894 tty->print("expected:");
4895 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4896 tty->cr();
4897 tty->print(" found:");
4898 os::Posix::print_sa_flags(tty, act.sa_flags);
4899 tty->cr();
4900 // No need to check this sig any longer
4901 sigaddset(&check_signal_done, sig);
4902 }
4903
4904 // Dump all the signal
4905 if (sigismember(&check_signal_done, sig)) {
4906 print_signal_handlers(tty, buf, O_BUFLEN);
4907 }
4908 }
4909
4910 extern void report_error(char* file_name, int line_no, char* title,
4911 char* format, ...);
4912
4913 // this is called _before_ most of the global arguments have been parsed
4914 void os::init(void) {
4915 char dummy; // used to get a guess on initial stack address
4916
4917 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4918
4919 init_random(1234567);
4920
4921 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4922 if (Linux::page_size() == -1) {
4923 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4924 os::strerror(errno));
4925 }
4926 init_page_sizes((size_t) Linux::page_size());
4927
4928 Linux::initialize_system_info();
4929
4930 Linux::initialize_os_info();
4931
4932 // _main_thread points to the thread that created/loaded the JVM.
4933 Linux::_main_thread = pthread_self();
4934
4935 // retrieve entry point for pthread_setname_np
4936 Linux::_pthread_setname_np =
4937 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4938
4939 os::Posix::init();
4940
4941 initial_time_count = javaTimeNanos();
4942
4943 // Always warn if no monotonic clock available
4944 if (!os::Posix::supports_monotonic_clock()) {
4945 warning("No monotonic clock was available - timed services may " \
4946 "be adversely affected if the time-of-day clock changes");
4947 }
4948 }
4949
4950 // To install functions for atexit system call
4951 extern "C" {
4952 static void perfMemory_exit_helper() {
4953 perfMemory_exit();
4954 }
4955 }
4956
4957 void os::pd_init_container_support() {
4958 OSContainer::init();
4959 }
4960
4961 // this is called _after_ the global arguments have been parsed
4962 jint os::init_2(void) {
4963
4964 // This could be set after os::Posix::init() but all platforms
4965 // have to set it the same so we have to mirror Solaris.
4966 DEBUG_ONLY(os::set_mutex_init_done();)
4967
4968 os::Posix::init_2();
4969
4970 Linux::fast_thread_clock_init();
4971
4972 // initialize suspend/resume support - must do this before signal_sets_init()
4973 if (SR_initialize() != 0) {
4974 perror("SR_initialize failed");
4975 return JNI_ERR;
4976 }
4977
4978 Linux::signal_sets_init();
4979 Linux::install_signal_handlers();
4980 // Initialize data for jdk.internal.misc.Signal
4981 if (!ReduceSignalUsage) {
4982 jdk_misc_signal_init();
4983 }
4984
4985 // Check and sets minimum stack sizes against command line options
4986 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4987 return JNI_ERR;
4988 }
4989
4990 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
4991 if (!suppress_primordial_thread_resolution) {
4992 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4993 }
4994
4995 #if defined(IA32)
4996 workaround_expand_exec_shield_cs_limit();
4997 #endif
4998
4999 Linux::libpthread_init();
5000 Linux::sched_getcpu_init();
5001 log_info(os)("HotSpot is running with %s, %s",
5002 Linux::glibc_version(), Linux::libpthread_version());
5003
5004 if (UseNUMA) {
5005 if (!Linux::libnuma_init()) {
5006 UseNUMA = false;
5007 } else {
5008 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5009 // If there's only one node (they start from 0) or if the process
5010 // is bound explicitly to a single node using membind, disable NUMA.
5011 UseNUMA = false;
5012 }
5013 }
5014
5015 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5016 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5017 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5018 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5019 // and disable adaptive resizing.
5020 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5021 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5022 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5023 UseAdaptiveSizePolicy = false;
5024 UseAdaptiveNUMAChunkSizing = false;
5025 }
5026 }
5027
5028 if (!UseNUMA && ForceNUMA) {
5029 UseNUMA = true;
5030 }
5031 }
5032
5033 if (MaxFDLimit) {
5034 // set the number of file descriptors to max. print out error
5035 // if getrlimit/setrlimit fails but continue regardless.
5036 struct rlimit nbr_files;
5037 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5038 if (status != 0) {
5039 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5040 } else {
5041 nbr_files.rlim_cur = nbr_files.rlim_max;
5042 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5043 if (status != 0) {
5044 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5045 }
5046 }
5047 }
5048
5049 // Initialize lock used to serialize thread creation (see os::create_thread)
5050 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5051
5052 // at-exit methods are called in the reverse order of their registration.
5053 // atexit functions are called on return from main or as a result of a
5054 // call to exit(3C). There can be only 32 of these functions registered
5055 // and atexit() does not set errno.
5056
5057 if (PerfAllowAtExitRegistration) {
5058 // only register atexit functions if PerfAllowAtExitRegistration is set.
5059 // atexit functions can be delayed until process exit time, which
5060 // can be problematic for embedded VM situations. Embedded VMs should
5061 // call DestroyJavaVM() to assure that VM resources are released.
5062
5063 // note: perfMemory_exit_helper atexit function may be removed in
5064 // the future if the appropriate cleanup code can be added to the
5065 // VM_Exit VMOperation's doit method.
5066 if (atexit(perfMemory_exit_helper) != 0) {
5067 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5068 }
5069 }
5070
5071 // initialize thread priority policy
5072 prio_init();
5073
5074 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5075 set_coredump_filter(DAX_SHARED_BIT);
5076 }
5077
5078 if (UseSharedSpaces && DumpPrivateMappingsInCore) {
5079 set_coredump_filter(FILE_BACKED_PVT_BIT);
5080 }
5081
5082 return JNI_OK;
5083 }
5084
5085 // Mark the polling page as unreadable
5086 void os::make_polling_page_unreadable(void) {
5087 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5088 fatal("Could not disable polling page");
5089 }
5090 }
5091
5092 // Mark the polling page as readable
5093 void os::make_polling_page_readable(void) {
5094 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5095 fatal("Could not enable polling page");
5096 }
5097 }
5098
5099 // older glibc versions don't have this macro (which expands to
5100 // an optimized bit-counting function) so we have to roll our own
5101 #ifndef CPU_COUNT
5102
5103 static int _cpu_count(const cpu_set_t* cpus) {
5104 int count = 0;
5105 // only look up to the number of configured processors
5106 for (int i = 0; i < os::processor_count(); i++) {
5107 if (CPU_ISSET(i, cpus)) {
5108 count++;
5109 }
5110 }
5111 return count;
5112 }
5113
5114 #define CPU_COUNT(cpus) _cpu_count(cpus)
5115
5116 #endif // CPU_COUNT
5117
5118 // Get the current number of available processors for this process.
5119 // This value can change at any time during a process's lifetime.
5120 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5121 // If it appears there may be more than 1024 processors then we do a
5122 // dynamic check - see 6515172 for details.
5123 // If anything goes wrong we fallback to returning the number of online
5124 // processors - which can be greater than the number available to the process.
5125 int os::Linux::active_processor_count() {
5126 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5127 cpu_set_t* cpus_p = &cpus;
5128 int cpus_size = sizeof(cpu_set_t);
5129
5130 int configured_cpus = os::processor_count(); // upper bound on available cpus
5131 int cpu_count = 0;
5132
5133 // old build platforms may not support dynamic cpu sets
5134 #ifdef CPU_ALLOC
5135
5136 // To enable easy testing of the dynamic path on different platforms we
5137 // introduce a diagnostic flag: UseCpuAllocPath
5138 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5139 // kernel may use a mask bigger than cpu_set_t
5140 log_trace(os)("active_processor_count: using dynamic path %s"
5141 "- configured processors: %d",
5142 UseCpuAllocPath ? "(forced) " : "",
5143 configured_cpus);
5144 cpus_p = CPU_ALLOC(configured_cpus);
5145 if (cpus_p != NULL) {
5146 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5147 // zero it just to be safe
5148 CPU_ZERO_S(cpus_size, cpus_p);
5149 }
5150 else {
5151 // failed to allocate so fallback to online cpus
5152 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5153 log_trace(os)("active_processor_count: "
5154 "CPU_ALLOC failed (%s) - using "
5155 "online processor count: %d",
5156 os::strerror(errno), online_cpus);
5157 return online_cpus;
5158 }
5159 }
5160 else {
5161 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5162 configured_cpus);
5163 }
5164 #else // CPU_ALLOC
5165 // these stubs won't be executed
5166 #define CPU_COUNT_S(size, cpus) -1
5167 #define CPU_FREE(cpus)
5168
5169 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5170 configured_cpus);
5171 #endif // CPU_ALLOC
5172
5173 // pid 0 means the current thread - which we have to assume represents the process
5174 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5175 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5176 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5177 }
5178 else {
5179 cpu_count = CPU_COUNT(cpus_p);
5180 }
5181 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5182 }
5183 else {
5184 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5185 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5186 "which may exceed available processors", os::strerror(errno), cpu_count);
5187 }
5188
5189 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5190 CPU_FREE(cpus_p);
5191 }
5192
5193 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5194 return cpu_count;
5195 }
5196
5197 // Determine the active processor count from one of
5198 // three different sources:
5199 //
5200 // 1. User option -XX:ActiveProcessorCount
5201 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5202 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5203 //
5204 // Option 1, if specified, will always override.
5205 // If the cgroup subsystem is active and configured, we
5206 // will return the min of the cgroup and option 2 results.
5207 // This is required since tools, such as numactl, that
5208 // alter cpu affinity do not update cgroup subsystem
5209 // cpuset configuration files.
5210 int os::active_processor_count() {
5211 // User has overridden the number of active processors
5212 if (ActiveProcessorCount > 0) {
5213 log_trace(os)("active_processor_count: "
5214 "active processor count set by user : %d",
5215 ActiveProcessorCount);
5216 return ActiveProcessorCount;
5217 }
5218
5219 int active_cpus;
5220 if (OSContainer::is_containerized()) {
5221 active_cpus = OSContainer::active_processor_count();
5222 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5223 active_cpus);
5224 } else {
5225 active_cpus = os::Linux::active_processor_count();
5226 }
5227
5228 return active_cpus;
5229 }
5230
5231 uint os::processor_id() {
5232 const int id = Linux::sched_getcpu();
5233 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5234 return (uint)id;
5235 }
5236
5237 void os::set_native_thread_name(const char *name) {
5238 if (Linux::_pthread_setname_np) {
5239 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5240 snprintf(buf, sizeof(buf), "%s", name);
5241 buf[sizeof(buf) - 1] = '\0';
5242 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5243 // ERANGE should not happen; all other errors should just be ignored.
5244 assert(rc != ERANGE, "pthread_setname_np failed");
5245 }
5246 }
5247
5248 bool os::distribute_processes(uint length, uint* distribution) {
5249 // Not yet implemented.
5250 return false;
5251 }
5252
5253 bool os::bind_to_processor(uint processor_id) {
5254 // Not yet implemented.
5255 return false;
5256 }
5257
5258 ///
5259
5260 void os::SuspendedThreadTask::internal_do_task() {
5261 if (do_suspend(_thread->osthread())) {
5262 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5263 do_task(context);
5264 do_resume(_thread->osthread());
5265 }
5266 }
5267
5268 ////////////////////////////////////////////////////////////////////////////////
5269 // debug support
5270
5271 bool os::find(address addr, outputStream* st) {
5272 Dl_info dlinfo;
5273 memset(&dlinfo, 0, sizeof(dlinfo));
5274 if (dladdr(addr, &dlinfo) != 0) {
5275 st->print(PTR_FORMAT ": ", p2i(addr));
5276 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5277 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5278 p2i(addr) - p2i(dlinfo.dli_saddr));
5279 } else if (dlinfo.dli_fbase != NULL) {
5280 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5281 } else {
5282 st->print("<absolute address>");
5283 }
5284 if (dlinfo.dli_fname != NULL) {
5285 st->print(" in %s", dlinfo.dli_fname);
5286 }
5287 if (dlinfo.dli_fbase != NULL) {
5288 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5289 }
5290 st->cr();
5291
5292 if (Verbose) {
5293 // decode some bytes around the PC
5294 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5295 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5296 address lowest = (address) dlinfo.dli_sname;
5297 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5298 if (begin < lowest) begin = lowest;
5299 Dl_info dlinfo2;
5300 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5301 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5302 end = (address) dlinfo2.dli_saddr;
5303 }
5304 Disassembler::decode(begin, end, st);
5305 }
5306 return true;
5307 }
5308 return false;
5309 }
5310
5311 ////////////////////////////////////////////////////////////////////////////////
5312 // misc
5313
5314 // This does not do anything on Linux. This is basically a hook for being
5315 // able to use structured exception handling (thread-local exception filters)
5316 // on, e.g., Win32.
5317 void
5318 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5319 JavaCallArguments* args, Thread* thread) {
5320 f(value, method, args, thread);
5321 }
5322
5323 void os::print_statistics() {
5324 }
5325
5326 bool os::message_box(const char* title, const char* message) {
5327 int i;
5328 fdStream err(defaultStream::error_fd());
5329 for (i = 0; i < 78; i++) err.print_raw("=");
5330 err.cr();
5331 err.print_raw_cr(title);
5332 for (i = 0; i < 78; i++) err.print_raw("-");
5333 err.cr();
5334 err.print_raw_cr(message);
5335 for (i = 0; i < 78; i++) err.print_raw("=");
5336 err.cr();
5337
5338 char buf[16];
5339 // Prevent process from exiting upon "read error" without consuming all CPU
5340 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5341
5342 return buf[0] == 'y' || buf[0] == 'Y';
5343 }
5344
5345 // Is a (classpath) directory empty?
5346 bool os::dir_is_empty(const char* path) {
5347 DIR *dir = NULL;
5348 struct dirent *ptr;
5349
5350 dir = opendir(path);
5351 if (dir == NULL) return true;
5352
5353 // Scan the directory
5354 bool result = true;
5355 while (result && (ptr = readdir(dir)) != NULL) {
5356 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5357 result = false;
5358 }
5359 }
5360 closedir(dir);
5361 return result;
5362 }
5363
5364 // This code originates from JDK's sysOpen and open64_w
5365 // from src/solaris/hpi/src/system_md.c
5366
5367 int os::open(const char *path, int oflag, int mode) {
5368 if (strlen(path) > MAX_PATH - 1) {
5369 errno = ENAMETOOLONG;
5370 return -1;
5371 }
5372
5373 // All file descriptors that are opened in the Java process and not
5374 // specifically destined for a subprocess should have the close-on-exec
5375 // flag set. If we don't set it, then careless 3rd party native code
5376 // might fork and exec without closing all appropriate file descriptors
5377 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5378 // turn might:
5379 //
5380 // - cause end-of-file to fail to be detected on some file
5381 // descriptors, resulting in mysterious hangs, or
5382 //
5383 // - might cause an fopen in the subprocess to fail on a system
5384 // suffering from bug 1085341.
5385 //
5386 // (Yes, the default setting of the close-on-exec flag is a Unix
5387 // design flaw)
5388 //
5389 // See:
5390 // 1085341: 32-bit stdio routines should support file descriptors >255
5391 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5392 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5393 //
5394 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5395 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5396 // because it saves a system call and removes a small window where the flag
5397 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5398 // and we fall back to using FD_CLOEXEC (see below).
5399 #ifdef O_CLOEXEC
5400 oflag |= O_CLOEXEC;
5401 #endif
5402
5403 int fd = ::open64(path, oflag, mode);
5404 if (fd == -1) return -1;
5405
5406 //If the open succeeded, the file might still be a directory
5407 {
5408 struct stat64 buf64;
5409 int ret = ::fstat64(fd, &buf64);
5410 int st_mode = buf64.st_mode;
5411
5412 if (ret != -1) {
5413 if ((st_mode & S_IFMT) == S_IFDIR) {
5414 errno = EISDIR;
5415 ::close(fd);
5416 return -1;
5417 }
5418 } else {
5419 ::close(fd);
5420 return -1;
5421 }
5422 }
5423
5424 #ifdef FD_CLOEXEC
5425 // Validate that the use of the O_CLOEXEC flag on open above worked.
5426 // With recent kernels, we will perform this check exactly once.
5427 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5428 if (!O_CLOEXEC_is_known_to_work) {
5429 int flags = ::fcntl(fd, F_GETFD);
5430 if (flags != -1) {
5431 if ((flags & FD_CLOEXEC) != 0)
5432 O_CLOEXEC_is_known_to_work = 1;
5433 else
5434 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5435 }
5436 }
5437 #endif
5438
5439 return fd;
5440 }
5441
5442
5443 // create binary file, rewriting existing file if required
5444 int os::create_binary_file(const char* path, bool rewrite_existing) {
5445 int oflags = O_WRONLY | O_CREAT;
5446 if (!rewrite_existing) {
5447 oflags |= O_EXCL;
5448 }
5449 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5450 }
5451
5452 // return current position of file pointer
5453 jlong os::current_file_offset(int fd) {
5454 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5455 }
5456
5457 // move file pointer to the specified offset
5458 jlong os::seek_to_file_offset(int fd, jlong offset) {
5459 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5460 }
5461
5462 // This code originates from JDK's sysAvailable
5463 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5464
5465 int os::available(int fd, jlong *bytes) {
5466 jlong cur, end;
5467 int mode;
5468 struct stat64 buf64;
5469
5470 if (::fstat64(fd, &buf64) >= 0) {
5471 mode = buf64.st_mode;
5472 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5473 int n;
5474 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5475 *bytes = n;
5476 return 1;
5477 }
5478 }
5479 }
5480 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5481 return 0;
5482 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5483 return 0;
5484 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5485 return 0;
5486 }
5487 *bytes = end - cur;
5488 return 1;
5489 }
5490
5491 // Map a block of memory.
5492 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5493 char *addr, size_t bytes, bool read_only,
5494 bool allow_exec) {
5495 int prot;
5496 int flags = MAP_PRIVATE;
5497
5498 if (read_only) {
5499 prot = PROT_READ;
5500 } else {
5501 prot = PROT_READ | PROT_WRITE;
5502 }
5503
5504 if (allow_exec) {
5505 prot |= PROT_EXEC;
5506 }
5507
5508 if (addr != NULL) {
5509 flags |= MAP_FIXED;
5510 }
5511
5512 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5513 fd, file_offset);
5514 if (mapped_address == MAP_FAILED) {
5515 return NULL;
5516 }
5517 return mapped_address;
5518 }
5519
5520
5521 // Remap a block of memory.
5522 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5523 char *addr, size_t bytes, bool read_only,
5524 bool allow_exec) {
5525 // same as map_memory() on this OS
5526 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5527 allow_exec);
5528 }
5529
5530
5531 // Unmap a block of memory.
5532 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5533 return munmap(addr, bytes) == 0;
5534 }
5535
5536 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5537
5538 static jlong fast_cpu_time(Thread *thread) {
5539 clockid_t clockid;
5540 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5541 &clockid);
5542 if (rc == 0) {
5543 return os::Linux::fast_thread_cpu_time(clockid);
5544 } else {
5545 // It's possible to encounter a terminated native thread that failed
5546 // to detach itself from the VM - which should result in ESRCH.
5547 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5548 return -1;
5549 }
5550 }
5551
5552 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5553 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5554 // of a thread.
5555 //
5556 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5557 // the fast estimate available on the platform.
5558
5559 jlong os::current_thread_cpu_time() {
5560 if (os::Linux::supports_fast_thread_cpu_time()) {
5561 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5562 } else {
5563 // return user + sys since the cost is the same
5564 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5565 }
5566 }
5567
5568 jlong os::thread_cpu_time(Thread* thread) {
5569 // consistent with what current_thread_cpu_time() returns
5570 if (os::Linux::supports_fast_thread_cpu_time()) {
5571 return fast_cpu_time(thread);
5572 } else {
5573 return slow_thread_cpu_time(thread, true /* user + sys */);
5574 }
5575 }
5576
5577 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5578 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5579 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5580 } else {
5581 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5582 }
5583 }
5584
5585 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5586 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5587 return fast_cpu_time(thread);
5588 } else {
5589 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5590 }
5591 }
5592
5593 // -1 on error.
5594 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5595 pid_t tid = thread->osthread()->thread_id();
5596 char *s;
5597 char stat[2048];
5598 int statlen;
5599 char proc_name[64];
5600 int count;
5601 long sys_time, user_time;
5602 char cdummy;
5603 int idummy;
5604 long ldummy;
5605 FILE *fp;
5606
5607 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5608 fp = fopen(proc_name, "r");
5609 if (fp == NULL) return -1;
5610 statlen = fread(stat, 1, 2047, fp);
5611 stat[statlen] = '\0';
5612 fclose(fp);
5613
5614 // Skip pid and the command string. Note that we could be dealing with
5615 // weird command names, e.g. user could decide to rename java launcher
5616 // to "java 1.4.2 :)", then the stat file would look like
5617 // 1234 (java 1.4.2 :)) R ... ...
5618 // We don't really need to know the command string, just find the last
5619 // occurrence of ")" and then start parsing from there. See bug 4726580.
5620 s = strrchr(stat, ')');
5621 if (s == NULL) return -1;
5622
5623 // Skip blank chars
5624 do { s++; } while (s && isspace(*s));
5625
5626 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5627 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5628 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5629 &user_time, &sys_time);
5630 if (count != 13) return -1;
5631 if (user_sys_cpu_time) {
5632 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5633 } else {
5634 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5635 }
5636 }
5637
5638 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5639 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5640 info_ptr->may_skip_backward = false; // elapsed time not wall time
5641 info_ptr->may_skip_forward = false; // elapsed time not wall time
5642 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5643 }
5644
5645 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5646 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5647 info_ptr->may_skip_backward = false; // elapsed time not wall time
5648 info_ptr->may_skip_forward = false; // elapsed time not wall time
5649 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5650 }
5651
5652 bool os::is_thread_cpu_time_supported() {
5653 return true;
5654 }
5655
5656 // System loadavg support. Returns -1 if load average cannot be obtained.
5657 // Linux doesn't yet have a (official) notion of processor sets,
5658 // so just return the system wide load average.
5659 int os::loadavg(double loadavg[], int nelem) {
5660 return ::getloadavg(loadavg, nelem);
5661 }
5662
5663 void os::pause() {
5664 char filename[MAX_PATH];
5665 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5666 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5667 } else {
5668 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5669 }
5670
5671 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5672 if (fd != -1) {
5673 struct stat buf;
5674 ::close(fd);
5675 while (::stat(filename, &buf) == 0) {
5676 (void)::poll(NULL, 0, 100);
5677 }
5678 } else {
5679 jio_fprintf(stderr,
5680 "Could not open pause file '%s', continuing immediately.\n", filename);
5681 }
5682 }
5683
5684 extern char** environ;
5685
5686 // Run the specified command in a separate process. Return its exit value,
5687 // or -1 on failure (e.g. can't fork a new process).
5688 // Unlike system(), this function can be called from signal handler. It
5689 // doesn't block SIGINT et al.
5690 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5691 const char * argv[4] = {"sh", "-c", cmd, NULL};
5692
5693 pid_t pid ;
5694
5695 if (use_vfork_if_available) {
5696 pid = vfork();
5697 } else {
5698 pid = fork();
5699 }
5700
5701 if (pid < 0) {
5702 // fork failed
5703 return -1;
5704
5705 } else if (pid == 0) {
5706 // child process
5707
5708 execve("/bin/sh", (char* const*)argv, environ);
5709
5710 // execve failed
5711 _exit(-1);
5712
5713 } else {
5714 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5715 // care about the actual exit code, for now.
5716
5717 int status;
5718
5719 // Wait for the child process to exit. This returns immediately if
5720 // the child has already exited. */
5721 while (waitpid(pid, &status, 0) < 0) {
5722 switch (errno) {
5723 case ECHILD: return 0;
5724 case EINTR: break;
5725 default: return -1;
5726 }
5727 }
5728
5729 if (WIFEXITED(status)) {
5730 // The child exited normally; get its exit code.
5731 return WEXITSTATUS(status);
5732 } else if (WIFSIGNALED(status)) {
5733 // The child exited because of a signal
5734 // The best value to return is 0x80 + signal number,
5735 // because that is what all Unix shells do, and because
5736 // it allows callers to distinguish between process exit and
5737 // process death by signal.
5738 return 0x80 + WTERMSIG(status);
5739 } else {
5740 // Unknown exit code; pass it through
5741 return status;
5742 }
5743 }
5744 }
5745
5746 // Get the default path to the core file
5747 // Returns the length of the string
5748 int os::get_core_path(char* buffer, size_t bufferSize) {
5749 /*
5750 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5751 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5752 */
5753 const int core_pattern_len = 129;
5754 char core_pattern[core_pattern_len] = {0};
5755
5756 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5757 if (core_pattern_file == -1) {
5758 return -1;
5759 }
5760
5761 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5762 ::close(core_pattern_file);
5763 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5764 return -1;
5765 }
5766 if (core_pattern[ret-1] == '\n') {
5767 core_pattern[ret-1] = '\0';
5768 } else {
5769 core_pattern[ret] = '\0';
5770 }
5771
5772 // Replace the %p in the core pattern with the process id. NOTE: we do this
5773 // only if the pattern doesn't start with "|", and we support only one %p in
5774 // the pattern.
5775 char *pid_pos = strstr(core_pattern, "%p");
5776 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5777 int written;
5778
5779 if (core_pattern[0] == '/') {
5780 if (pid_pos != NULL) {
5781 *pid_pos = '\0';
5782 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5783 current_process_id(), tail);
5784 } else {
5785 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5786 }
5787 } else {
5788 char cwd[PATH_MAX];
5789
5790 const char* p = get_current_directory(cwd, PATH_MAX);
5791 if (p == NULL) {
5792 return -1;
5793 }
5794
5795 if (core_pattern[0] == '|') {
5796 written = jio_snprintf(buffer, bufferSize,
5797 "\"%s\" (or dumping to %s/core.%d)",
5798 &core_pattern[1], p, current_process_id());
5799 } else if (pid_pos != NULL) {
5800 *pid_pos = '\0';
5801 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5802 current_process_id(), tail);
5803 } else {
5804 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5805 }
5806 }
5807
5808 if (written < 0) {
5809 return -1;
5810 }
5811
5812 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5813 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5814
5815 if (core_uses_pid_file != -1) {
5816 char core_uses_pid = 0;
5817 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5818 ::close(core_uses_pid_file);
5819
5820 if (core_uses_pid == '1') {
5821 jio_snprintf(buffer + written, bufferSize - written,
5822 ".%d", current_process_id());
5823 }
5824 }
5825 }
5826
5827 return strlen(buffer);
5828 }
5829
5830 bool os::start_debugging(char *buf, int buflen) {
5831 int len = (int)strlen(buf);
5832 char *p = &buf[len];
5833
5834 jio_snprintf(p, buflen-len,
5835 "\n\n"
5836 "Do you want to debug the problem?\n\n"
5837 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5838 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5839 "Otherwise, press RETURN to abort...",
5840 os::current_process_id(), os::current_process_id(),
5841 os::current_thread_id(), os::current_thread_id());
5842
5843 bool yes = os::message_box("Unexpected Error", buf);
5844
5845 if (yes) {
5846 // yes, user asked VM to launch debugger
5847 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5848 os::current_process_id(), os::current_process_id());
5849
5850 os::fork_and_exec(buf);
5851 yes = false;
5852 }
5853 return yes;
5854 }
5855
5856
5857 // Java/Compiler thread:
5858 //
5859 // Low memory addresses
5860 // P0 +------------------------+
5861 // | |\ Java thread created by VM does not have glibc
5862 // | glibc guard page | - guard page, attached Java thread usually has
5863 // | |/ 1 glibc guard page.
5864 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5865 // | |\
5866 // | HotSpot Guard Pages | - red, yellow and reserved pages
5867 // | |/
5868 // +------------------------+ JavaThread::stack_reserved_zone_base()
5869 // | |\
5870 // | Normal Stack | -
5871 // | |/
5872 // P2 +------------------------+ Thread::stack_base()
5873 //
5874 // Non-Java thread:
5875 //
5876 // Low memory addresses
5877 // P0 +------------------------+
5878 // | |\
5879 // | glibc guard page | - usually 1 page
5880 // | |/
5881 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5882 // | |\
5883 // | Normal Stack | -
5884 // | |/
5885 // P2 +------------------------+ Thread::stack_base()
5886 //
5887 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5888 // returned from pthread_attr_getstack().
5889 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5890 // of the stack size given in pthread_attr. We work around this for
5891 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5892 //
5893 #ifndef ZERO
5894 static void current_stack_region(address * bottom, size_t * size) {
5895 if (os::is_primordial_thread()) {
5896 // primordial thread needs special handling because pthread_getattr_np()
5897 // may return bogus value.
5898 *bottom = os::Linux::initial_thread_stack_bottom();
5899 *size = os::Linux::initial_thread_stack_size();
5900 } else {
5901 pthread_attr_t attr;
5902
5903 int rslt = pthread_getattr_np(pthread_self(), &attr);
5904
5905 // JVM needs to know exact stack location, abort if it fails
5906 if (rslt != 0) {
5907 if (rslt == ENOMEM) {
5908 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5909 } else {
5910 fatal("pthread_getattr_np failed with error = %d", rslt);
5911 }
5912 }
5913
5914 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5915 fatal("Cannot locate current stack attributes!");
5916 }
5917
5918 // Work around NPTL stack guard error.
5919 size_t guard_size = 0;
5920 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5921 if (rslt != 0) {
5922 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5923 }
5924 *bottom += guard_size;
5925 *size -= guard_size;
5926
5927 pthread_attr_destroy(&attr);
5928
5929 }
5930 assert(os::current_stack_pointer() >= *bottom &&
5931 os::current_stack_pointer() < *bottom + *size, "just checking");
5932 }
5933
5934 address os::current_stack_base() {
5935 address bottom;
5936 size_t size;
5937 current_stack_region(&bottom, &size);
5938 return (bottom + size);
5939 }
5940
5941 size_t os::current_stack_size() {
5942 // This stack size includes the usable stack and HotSpot guard pages
5943 // (for the threads that have Hotspot guard pages).
5944 address bottom;
5945 size_t size;
5946 current_stack_region(&bottom, &size);
5947 return size;
5948 }
5949 #endif
5950
5951 static inline struct timespec get_mtime(const char* filename) {
5952 struct stat st;
5953 int ret = os::stat(filename, &st);
5954 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
5955 return st.st_mtim;
5956 }
5957
5958 int os::compare_file_modified_times(const char* file1, const char* file2) {
5959 struct timespec filetime1 = get_mtime(file1);
5960 struct timespec filetime2 = get_mtime(file2);
5961 int diff = filetime1.tv_sec - filetime2.tv_sec;
5962 if (diff == 0) {
5963 return filetime1.tv_nsec - filetime2.tv_nsec;
5964 }
5965 return diff;
5966 }
5967
5968 /////////////// Unit tests ///////////////
5969
5970 #ifndef PRODUCT
5971
5972 class TestReserveMemorySpecial : AllStatic {
5973 public:
5974 static void small_page_write(void* addr, size_t size) {
5975 size_t page_size = os::vm_page_size();
5976
5977 char* end = (char*)addr + size;
5978 for (char* p = (char*)addr; p < end; p += page_size) {
5979 *p = 1;
5980 }
5981 }
5982
5983 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5984 if (!UseHugeTLBFS) {
5985 return;
5986 }
5987
5988 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5989
5990 if (addr != NULL) {
5991 small_page_write(addr, size);
5992
5993 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5994 }
5995 }
5996
5997 static void test_reserve_memory_special_huge_tlbfs_only() {
5998 if (!UseHugeTLBFS) {
5999 return;
6000 }
6001
6002 size_t lp = os::large_page_size();
6003
6004 for (size_t size = lp; size <= lp * 10; size += lp) {
6005 test_reserve_memory_special_huge_tlbfs_only(size);
6006 }
6007 }
6008
6009 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6010 size_t lp = os::large_page_size();
6011 size_t ag = os::vm_allocation_granularity();
6012
6013 // sizes to test
6014 const size_t sizes[] = {
6015 lp, lp + ag, lp + lp / 2, lp * 2,
6016 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6017 lp * 10, lp * 10 + lp / 2
6018 };
6019 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6020
6021 // For each size/alignment combination, we test three scenarios:
6022 // 1) with req_addr == NULL
6023 // 2) with a non-null req_addr at which we expect to successfully allocate
6024 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6025 // expect the allocation to either fail or to ignore req_addr
6026
6027 // Pre-allocate two areas; they shall be as large as the largest allocation
6028 // and aligned to the largest alignment we will be testing.
6029 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6030 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6031 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6032 -1, 0);
6033 assert(mapping1 != MAP_FAILED, "should work");
6034
6035 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6036 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6037 -1, 0);
6038 assert(mapping2 != MAP_FAILED, "should work");
6039
6040 // Unmap the first mapping, but leave the second mapping intact: the first
6041 // mapping will serve as a value for a "good" req_addr (case 2). The second
6042 // mapping, still intact, as "bad" req_addr (case 3).
6043 ::munmap(mapping1, mapping_size);
6044
6045 // Case 1
6046 for (int i = 0; i < num_sizes; i++) {
6047 const size_t size = sizes[i];
6048 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6049 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
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 for (int i = 0; i < num_sizes; i++) {
6060 const size_t size = sizes[i];
6061 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6062 char* const req_addr = align_up(mapping1, alignment);
6063 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6064 if (p != NULL) {
6065 assert(p == req_addr, "must be");
6066 small_page_write(p, size);
6067 os::Linux::release_memory_special_huge_tlbfs(p, size);
6068 }
6069 }
6070 }
6071
6072 // Case 3
6073 for (int i = 0; i < num_sizes; i++) {
6074 const size_t size = sizes[i];
6075 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6076 char* const req_addr = align_up(mapping2, alignment);
6077 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6078 // as the area around req_addr contains already existing mappings, the API should always
6079 // return NULL (as per contract, it cannot return another address)
6080 assert(p == NULL, "must be");
6081 }
6082 }
6083
6084 ::munmap(mapping2, mapping_size);
6085
6086 }
6087
6088 static void test_reserve_memory_special_huge_tlbfs() {
6089 if (!UseHugeTLBFS) {
6090 return;
6091 }
6092
6093 test_reserve_memory_special_huge_tlbfs_only();
6094 test_reserve_memory_special_huge_tlbfs_mixed();
6095 }
6096
6097 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6098 if (!UseSHM) {
6099 return;
6100 }
6101
6102 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6103
6104 if (addr != NULL) {
6105 assert(is_aligned(addr, alignment), "Check");
6106 assert(is_aligned(addr, os::large_page_size()), "Check");
6107
6108 small_page_write(addr, size);
6109
6110 os::Linux::release_memory_special_shm(addr, size);
6111 }
6112 }
6113
6114 static void test_reserve_memory_special_shm() {
6115 size_t lp = os::large_page_size();
6116 size_t ag = os::vm_allocation_granularity();
6117
6118 for (size_t size = ag; size < lp * 3; size += ag) {
6119 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6120 test_reserve_memory_special_shm(size, alignment);
6121 }
6122 }
6123 }
6124
6125 static void test() {
6126 test_reserve_memory_special_huge_tlbfs();
6127 test_reserve_memory_special_shm();
6128 }
6129 };
6130
6131 void TestReserveMemorySpecial_test() {
6132 TestReserveMemorySpecial::test();
6133 }
6134
6135 #endif