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 #define LARGEPAGES_BIT (1 << 6)
132 #define DAX_SHARED_BIT (1 << 8)
133 ////////////////////////////////////////////////////////////////////////////////
134 // global variables
135 julong os::Linux::_physical_memory = 0;
136
137 address os::Linux::_initial_thread_stack_bottom = NULL;
138 uintptr_t os::Linux::_initial_thread_stack_size = 0;
139
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 uint32_t os::Linux::_os_version = 0;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // If the VM might have been created on the primordial thread, we need to resolve the
155 // primordial thread stack bounds and check if the current thread might be the
156 // primordial thread in places. If we know that the primordial thread is never used,
157 // such as when the VM was created by one of the standard java launchers, we can
158 // avoid this
159 static bool suppress_primordial_thread_resolution = false;
160
161 // For diagnostics to print a message once. see run_periodic_checks
162 static sigset_t check_signal_done;
163 static bool check_signals = true;
164
165 // Signal number used to suspend/resume a thread
166
167 // do not use any signal number less than SIGSEGV, see 4355769
168 static int SR_signum = SIGUSR2;
169 sigset_t SR_sigset;
170
171 // utility functions
172
173 static int SR_initialize();
174
175 julong os::available_memory() {
176 return Linux::available_memory();
177 }
178
179 julong os::Linux::available_memory() {
180 // values in struct sysinfo are "unsigned long"
181 struct sysinfo si;
182 julong avail_mem;
183
184 if (OSContainer::is_containerized()) {
185 jlong mem_limit, mem_usage;
186 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
187 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
188 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
189 }
190 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
191 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
192 }
193 if (mem_limit > 0 && mem_usage > 0 ) {
194 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
195 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
196 return avail_mem;
197 }
198 }
199
200 sysinfo(&si);
201 avail_mem = (julong)si.freeram * si.mem_unit;
202 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
203 return avail_mem;
204 }
205
206 julong os::physical_memory() {
207 jlong phys_mem = 0;
208 if (OSContainer::is_containerized()) {
209 jlong mem_limit;
210 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
211 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
212 return mem_limit;
213 }
214 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
215 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
216 }
217
218 phys_mem = Linux::physical_memory();
219 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
220 return phys_mem;
221 }
222
223 // Return true if user is running as root.
224
225 bool os::have_special_privileges() {
226 static bool init = false;
227 static bool privileges = false;
228 if (!init) {
229 privileges = (getuid() != geteuid()) || (getgid() != getegid());
230 init = true;
231 }
232 return privileges;
233 }
234
235
236 #ifndef SYS_gettid
237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
238 #ifdef __ia64__
239 #define SYS_gettid 1105
240 #else
241 #ifdef __i386__
242 #define SYS_gettid 224
243 #else
244 #ifdef __amd64__
245 #define SYS_gettid 186
246 #else
247 #ifdef __sparc__
248 #define SYS_gettid 143
249 #else
250 #error define gettid for the arch
251 #endif
252 #endif
253 #endif
254 #endif
255 #endif
256
257
258 // pid_t gettid()
259 //
260 // Returns the kernel thread id of the currently running thread. Kernel
261 // thread id is used to access /proc.
262 pid_t os::Linux::gettid() {
263 int rslt = syscall(SYS_gettid);
264 assert(rslt != -1, "must be."); // old linuxthreads implementation?
265 return (pid_t)rslt;
266 }
267
268 // Most versions of linux have a bug where the number of processors are
269 // determined by looking at the /proc file system. In a chroot environment,
270 // the system call returns 1.
271 static bool unsafe_chroot_detected = false;
272 static const char *unstable_chroot_error = "/proc file system not found.\n"
273 "Java may be unstable running multithreaded in a chroot "
274 "environment on Linux when /proc filesystem is not mounted.";
275
276 void os::Linux::initialize_system_info() {
277 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
278 if (processor_count() == 1) {
279 pid_t pid = os::Linux::gettid();
280 char fname[32];
281 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
282 FILE *fp = fopen(fname, "r");
283 if (fp == NULL) {
284 unsafe_chroot_detected = true;
285 } else {
286 fclose(fp);
287 }
288 }
289 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
290 assert(processor_count() > 0, "linux error");
291 }
292
293 void os::init_system_properties_values() {
294 // The next steps are taken in the product version:
295 //
296 // Obtain the JAVA_HOME value from the location of libjvm.so.
297 // This library should be located at:
298 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
299 //
300 // If "/jre/lib/" appears at the right place in the path, then we
301 // assume libjvm.so is installed in a JDK and we use this path.
302 //
303 // Otherwise exit with message: "Could not create the Java virtual machine."
304 //
305 // The following extra steps are taken in the debugging version:
306 //
307 // If "/jre/lib/" does NOT appear at the right place in the path
308 // instead of exit check for $JAVA_HOME environment variable.
309 //
310 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
311 // then we append a fake suffix "hotspot/libjvm.so" to this path so
312 // it looks like libjvm.so is installed there
313 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
314 //
315 // Otherwise exit.
316 //
317 // Important note: if the location of libjvm.so changes this
318 // code needs to be changed accordingly.
319
320 // See ld(1):
321 // The linker uses the following search paths to locate required
322 // shared libraries:
323 // 1: ...
324 // ...
325 // 7: The default directories, normally /lib and /usr/lib.
326 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
327 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
328 #else
329 #define DEFAULT_LIBPATH "/lib:/usr/lib"
330 #endif
331
332 // Base path of extensions installed on the system.
333 #define SYS_EXT_DIR "/usr/java/packages"
334 #define EXTENSIONS_DIR "/lib/ext"
335
336 // Buffer that fits several sprintfs.
337 // Note that the space for the colon and the trailing null are provided
338 // by the nulls included by the sizeof operator.
339 const size_t bufsize =
340 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
341 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
342 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
343
344 // sysclasspath, java_home, dll_dir
345 {
346 char *pslash;
347 os::jvm_path(buf, bufsize);
348
349 // Found the full path to libjvm.so.
350 // Now cut the path to <java_home>/jre if we can.
351 pslash = strrchr(buf, '/');
352 if (pslash != NULL) {
353 *pslash = '\0'; // Get rid of /libjvm.so.
354 }
355 pslash = strrchr(buf, '/');
356 if (pslash != NULL) {
357 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
358 }
359 Arguments::set_dll_dir(buf);
360
361 if (pslash != NULL) {
362 pslash = strrchr(buf, '/');
363 if (pslash != NULL) {
364 *pslash = '\0'; // Get rid of /lib.
365 }
366 }
367 Arguments::set_java_home(buf);
368 if (!set_boot_path('/', ':')) {
369 vm_exit_during_initialization("Failed setting boot class path.", NULL);
370 }
371 }
372
373 // Where to look for native libraries.
374 //
375 // Note: Due to a legacy implementation, most of the library path
376 // is set in the launcher. This was to accomodate linking restrictions
377 // on legacy Linux implementations (which are no longer supported).
378 // Eventually, all the library path setting will be done here.
379 //
380 // However, to prevent the proliferation of improperly built native
381 // libraries, the new path component /usr/java/packages is added here.
382 // Eventually, all the library path setting will be done here.
383 {
384 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
385 // should always exist (until the legacy problem cited above is
386 // addressed).
387 const char *v = ::getenv("LD_LIBRARY_PATH");
388 const char *v_colon = ":";
389 if (v == NULL) { v = ""; v_colon = ""; }
390 // That's +1 for the colon and +1 for the trailing '\0'.
391 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
392 strlen(v) + 1 +
393 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
394 mtInternal);
395 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
396 Arguments::set_library_path(ld_library_path);
397 FREE_C_HEAP_ARRAY(char, ld_library_path);
398 }
399
400 // Extensions directories.
401 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
402 Arguments::set_ext_dirs(buf);
403
404 FREE_C_HEAP_ARRAY(char, buf);
405
406 #undef DEFAULT_LIBPATH
407 #undef SYS_EXT_DIR
408 #undef EXTENSIONS_DIR
409 }
410
411 ////////////////////////////////////////////////////////////////////////////////
412 // breakpoint support
413
414 void os::breakpoint() {
415 BREAKPOINT;
416 }
417
418 extern "C" void breakpoint() {
419 // use debugger to set breakpoint here
420 }
421
422 ////////////////////////////////////////////////////////////////////////////////
423 // signal support
424
425 debug_only(static bool signal_sets_initialized = false);
426 static sigset_t unblocked_sigs, vm_sigs;
427
428 void os::Linux::signal_sets_init() {
429 // Should also have an assertion stating we are still single-threaded.
430 assert(!signal_sets_initialized, "Already initialized");
431 // Fill in signals that are necessarily unblocked for all threads in
432 // the VM. Currently, we unblock the following signals:
433 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
434 // by -Xrs (=ReduceSignalUsage));
435 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
436 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
437 // the dispositions or masks wrt these signals.
438 // Programs embedding the VM that want to use the above signals for their
439 // own purposes must, at this time, use the "-Xrs" option to prevent
440 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
441 // (See bug 4345157, and other related bugs).
442 // In reality, though, unblocking these signals is really a nop, since
443 // these signals are not blocked by default.
444 sigemptyset(&unblocked_sigs);
445 sigaddset(&unblocked_sigs, SIGILL);
446 sigaddset(&unblocked_sigs, SIGSEGV);
447 sigaddset(&unblocked_sigs, SIGBUS);
448 sigaddset(&unblocked_sigs, SIGFPE);
449 #if defined(PPC64)
450 sigaddset(&unblocked_sigs, SIGTRAP);
451 #endif
452 sigaddset(&unblocked_sigs, SR_signum);
453
454 if (!ReduceSignalUsage) {
455 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
456 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
457 }
458 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
459 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
460 }
461 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
462 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
463 }
464 }
465 // Fill in signals that are blocked by all but the VM thread.
466 sigemptyset(&vm_sigs);
467 if (!ReduceSignalUsage) {
468 sigaddset(&vm_sigs, BREAK_SIGNAL);
469 }
470 debug_only(signal_sets_initialized = true);
471
472 }
473
474 // These are signals that are unblocked while a thread is running Java.
475 // (For some reason, they get blocked by default.)
476 sigset_t* os::Linux::unblocked_signals() {
477 assert(signal_sets_initialized, "Not initialized");
478 return &unblocked_sigs;
479 }
480
481 // These are the signals that are blocked while a (non-VM) thread is
482 // running Java. Only the VM thread handles these signals.
483 sigset_t* os::Linux::vm_signals() {
484 assert(signal_sets_initialized, "Not initialized");
485 return &vm_sigs;
486 }
487
488 void os::Linux::hotspot_sigmask(Thread* thread) {
489
490 //Save caller's signal mask before setting VM signal mask
491 sigset_t caller_sigmask;
492 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
493
494 OSThread* osthread = thread->osthread();
495 osthread->set_caller_sigmask(caller_sigmask);
496
497 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
498
499 if (!ReduceSignalUsage) {
500 if (thread->is_VM_thread()) {
501 // Only the VM thread handles BREAK_SIGNAL ...
502 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
503 } else {
504 // ... all other threads block BREAK_SIGNAL
505 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
506 }
507 }
508 }
509
510 //////////////////////////////////////////////////////////////////////////////
511 // detecting pthread library
512
513 void os::Linux::libpthread_init() {
514 // Save glibc and pthread version strings.
515 #if !defined(_CS_GNU_LIBC_VERSION) || \
516 !defined(_CS_GNU_LIBPTHREAD_VERSION)
517 #error "glibc too old (< 2.3.2)"
518 #endif
519
520 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
521 assert(n > 0, "cannot retrieve glibc version");
522 char *str = (char *)malloc(n, mtInternal);
523 confstr(_CS_GNU_LIBC_VERSION, str, n);
524 os::Linux::set_glibc_version(str);
525
526 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
527 assert(n > 0, "cannot retrieve pthread version");
528 str = (char *)malloc(n, mtInternal);
529 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
530 os::Linux::set_libpthread_version(str);
531 }
532
533 /////////////////////////////////////////////////////////////////////////////
534 // thread stack expansion
535
536 // os::Linux::manually_expand_stack() takes care of expanding the thread
537 // stack. Note that this is normally not needed: pthread stacks allocate
538 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
539 // committed. Therefore it is not necessary to expand the stack manually.
540 //
541 // Manually expanding the stack was historically needed on LinuxThreads
542 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
543 // it is kept to deal with very rare corner cases:
544 //
545 // For one, user may run the VM on an own implementation of threads
546 // whose stacks are - like the old LinuxThreads - implemented using
547 // mmap(MAP_GROWSDOWN).
548 //
549 // Also, this coding may be needed if the VM is running on the primordial
550 // thread. Normally we avoid running on the primordial thread; however,
551 // user may still invoke the VM on the primordial thread.
552 //
553 // The following historical comment describes the details about running
554 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
555
556
557 // Force Linux kernel to expand current thread stack. If "bottom" is close
558 // to the stack guard, caller should block all signals.
559 //
560 // MAP_GROWSDOWN:
561 // A special mmap() flag that is used to implement thread stacks. It tells
562 // kernel that the memory region should extend downwards when needed. This
563 // allows early versions of LinuxThreads to only mmap the first few pages
564 // when creating a new thread. Linux kernel will automatically expand thread
565 // stack as needed (on page faults).
566 //
567 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
568 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
569 // region, it's hard to tell if the fault is due to a legitimate stack
570 // access or because of reading/writing non-exist memory (e.g. buffer
571 // overrun). As a rule, if the fault happens below current stack pointer,
572 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
573 // application (see Linux kernel fault.c).
574 //
575 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
576 // stack overflow detection.
577 //
578 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
579 // not use MAP_GROWSDOWN.
580 //
581 // To get around the problem and allow stack banging on Linux, we need to
582 // manually expand thread stack after receiving the SIGSEGV.
583 //
584 // There are two ways to expand thread stack to address "bottom", we used
585 // both of them in JVM before 1.5:
586 // 1. adjust stack pointer first so that it is below "bottom", and then
587 // touch "bottom"
588 // 2. mmap() the page in question
589 //
590 // Now alternate signal stack is gone, it's harder to use 2. For instance,
591 // if current sp is already near the lower end of page 101, and we need to
592 // call mmap() to map page 100, it is possible that part of the mmap() frame
593 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
594 // That will destroy the mmap() frame and cause VM to crash.
595 //
596 // The following code works by adjusting sp first, then accessing the "bottom"
597 // page to force a page fault. Linux kernel will then automatically expand the
598 // stack mapping.
599 //
600 // _expand_stack_to() assumes its frame size is less than page size, which
601 // should always be true if the function is not inlined.
602
603 static void NOINLINE _expand_stack_to(address bottom) {
604 address sp;
605 size_t size;
606 volatile char *p;
607
608 // Adjust bottom to point to the largest address within the same page, it
609 // gives us a one-page buffer if alloca() allocates slightly more memory.
610 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
611 bottom += os::Linux::page_size() - 1;
612
613 // sp might be slightly above current stack pointer; if that's the case, we
614 // will alloca() a little more space than necessary, which is OK. Don't use
615 // os::current_stack_pointer(), as its result can be slightly below current
616 // stack pointer, causing us to not alloca enough to reach "bottom".
617 sp = (address)&sp;
618
619 if (sp > bottom) {
620 size = sp - bottom;
621 p = (volatile char *)alloca(size);
622 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
623 p[0] = '\0';
624 }
625 }
626
627 void os::Linux::expand_stack_to(address bottom) {
628 _expand_stack_to(bottom);
629 }
630
631 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
632 assert(t!=NULL, "just checking");
633 assert(t->osthread()->expanding_stack(), "expand should be set");
634 assert(t->stack_base() != NULL, "stack_base was not initialized");
635
636 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
637 sigset_t mask_all, old_sigset;
638 sigfillset(&mask_all);
639 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
640 _expand_stack_to(addr);
641 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
642 return true;
643 }
644 return false;
645 }
646
647 //////////////////////////////////////////////////////////////////////////////
648 // create new thread
649
650 // Thread start routine for all newly created threads
651 static void *thread_native_entry(Thread *thread) {
652
653 thread->record_stack_base_and_size();
654
655 // Try to randomize the cache line index of hot stack frames.
656 // This helps when threads of the same stack traces evict each other's
657 // cache lines. The threads can be either from the same JVM instance, or
658 // from different JVM instances. The benefit is especially true for
659 // processors with hyperthreading technology.
660 static int counter = 0;
661 int pid = os::current_process_id();
662 alloca(((pid ^ counter++) & 7) * 128);
663
664 thread->initialize_thread_current();
665
666 OSThread* osthread = thread->osthread();
667 Monitor* sync = osthread->startThread_lock();
668
669 osthread->set_thread_id(os::current_thread_id());
670
671 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
672 os::current_thread_id(), (uintx) pthread_self());
673
674 if (UseNUMA) {
675 int lgrp_id = os::numa_get_group_id();
676 if (lgrp_id != -1) {
677 thread->set_lgrp_id(lgrp_id);
678 }
679 }
680 // initialize signal mask for this thread
681 os::Linux::hotspot_sigmask(thread);
682
683 // initialize floating point control register
684 os::Linux::init_thread_fpu_state();
685
686 // handshaking with parent thread
687 {
688 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
689
690 // notify parent thread
691 osthread->set_state(INITIALIZED);
692 sync->notify_all();
693
694 // wait until os::start_thread()
695 while (osthread->get_state() == INITIALIZED) {
696 sync->wait(Mutex::_no_safepoint_check_flag);
697 }
698 }
699
700 // call one more level start routine
701 thread->call_run();
702
703 // Note: at this point the thread object may already have deleted itself.
704 // Prevent dereferencing it from here on out.
705 thread = NULL;
706
707 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
708 os::current_thread_id(), (uintx) pthread_self());
709
710 return 0;
711 }
712
713 bool os::create_thread(Thread* thread, ThreadType thr_type,
714 size_t req_stack_size) {
715 assert(thread->osthread() == NULL, "caller responsible");
716
717 // Allocate the OSThread object
718 OSThread* osthread = new OSThread(NULL, NULL);
719 if (osthread == NULL) {
720 return false;
721 }
722
723 // set the correct thread state
724 osthread->set_thread_type(thr_type);
725
726 // Initial state is ALLOCATED but not INITIALIZED
727 osthread->set_state(ALLOCATED);
728
729 thread->set_osthread(osthread);
730
731 // init thread attributes
732 pthread_attr_t attr;
733 pthread_attr_init(&attr);
734 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
735
736 // Calculate stack size if it's not specified by caller.
737 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
738 // In the Linux NPTL pthread implementation the guard size mechanism
739 // is not implemented properly. The posix standard requires adding
740 // the size of the guard pages to the stack size, instead Linux
741 // takes the space out of 'stacksize'. Thus we adapt the requested
742 // stack_size by the size of the guard pages to mimick proper
743 // behaviour. However, be careful not to end up with a size
744 // of zero due to overflow. Don't add the guard page in that case.
745 size_t guard_size = os::Linux::default_guard_size(thr_type);
746 if (stack_size <= SIZE_MAX - guard_size) {
747 stack_size += guard_size;
748 }
749 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
750
751 int status = pthread_attr_setstacksize(&attr, stack_size);
752 assert_status(status == 0, status, "pthread_attr_setstacksize");
753
754 // Configure glibc guard page.
755 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
756
757 ThreadState state;
758
759 {
760 pthread_t tid;
761 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
762
763 char buf[64];
764 if (ret == 0) {
765 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
766 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
767 } else {
768 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
769 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
770 }
771
772 pthread_attr_destroy(&attr);
773
774 if (ret != 0) {
775 // Need to clean up stuff we've allocated so far
776 thread->set_osthread(NULL);
777 delete osthread;
778 return false;
779 }
780
781 // Store pthread info into the OSThread
782 osthread->set_pthread_id(tid);
783
784 // Wait until child thread is either initialized or aborted
785 {
786 Monitor* sync_with_child = osthread->startThread_lock();
787 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
788 while ((state = osthread->get_state()) == ALLOCATED) {
789 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
790 }
791 }
792 }
793
794 // Aborted due to thread limit being reached
795 if (state == ZOMBIE) {
796 thread->set_osthread(NULL);
797 delete osthread;
798 return false;
799 }
800
801 // The thread is returned suspended (in state INITIALIZED),
802 // and is started higher up in the call chain
803 assert(state == INITIALIZED, "race condition");
804 return true;
805 }
806
807 /////////////////////////////////////////////////////////////////////////////
808 // attach existing thread
809
810 // bootstrap the main thread
811 bool os::create_main_thread(JavaThread* thread) {
812 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
813 return create_attached_thread(thread);
814 }
815
816 bool os::create_attached_thread(JavaThread* thread) {
817 #ifdef ASSERT
818 thread->verify_not_published();
819 #endif
820
821 // Allocate the OSThread object
822 OSThread* osthread = new OSThread(NULL, NULL);
823
824 if (osthread == NULL) {
825 return false;
826 }
827
828 // Store pthread info into the OSThread
829 osthread->set_thread_id(os::Linux::gettid());
830 osthread->set_pthread_id(::pthread_self());
831
832 // initialize floating point control register
833 os::Linux::init_thread_fpu_state();
834
835 // Initial thread state is RUNNABLE
836 osthread->set_state(RUNNABLE);
837
838 thread->set_osthread(osthread);
839
840 if (UseNUMA) {
841 int lgrp_id = os::numa_get_group_id();
842 if (lgrp_id != -1) {
843 thread->set_lgrp_id(lgrp_id);
844 }
845 }
846
847 if (os::is_primordial_thread()) {
848 // If current thread is primordial thread, its stack is mapped on demand,
849 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
850 // the entire stack region to avoid SEGV in stack banging.
851 // It is also useful to get around the heap-stack-gap problem on SuSE
852 // kernel (see 4821821 for details). We first expand stack to the top
853 // of yellow zone, then enable stack yellow zone (order is significant,
854 // enabling yellow zone first will crash JVM on SuSE Linux), so there
855 // is no gap between the last two virtual memory regions.
856
857 JavaThread *jt = (JavaThread *)thread;
858 address addr = jt->stack_reserved_zone_base();
859 assert(addr != NULL, "initialization problem?");
860 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
861
862 osthread->set_expanding_stack();
863 os::Linux::manually_expand_stack(jt, addr);
864 osthread->clear_expanding_stack();
865 }
866
867 // initialize signal mask for this thread
868 // and save the caller's signal mask
869 os::Linux::hotspot_sigmask(thread);
870
871 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
872 os::current_thread_id(), (uintx) pthread_self());
873
874 return true;
875 }
876
877 void os::pd_start_thread(Thread* thread) {
878 OSThread * osthread = thread->osthread();
879 assert(osthread->get_state() != INITIALIZED, "just checking");
880 Monitor* sync_with_child = osthread->startThread_lock();
881 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
882 sync_with_child->notify();
883 }
884
885 // Free Linux resources related to the OSThread
886 void os::free_thread(OSThread* osthread) {
887 assert(osthread != NULL, "osthread not set");
888
889 // We are told to free resources of the argument thread,
890 // but we can only really operate on the current thread.
891 assert(Thread::current()->osthread() == osthread,
892 "os::free_thread but not current thread");
893
894 #ifdef ASSERT
895 sigset_t current;
896 sigemptyset(¤t);
897 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
898 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
899 #endif
900
901 // Restore caller's signal mask
902 sigset_t sigmask = osthread->caller_sigmask();
903 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
904
905 delete osthread;
906 }
907
908 //////////////////////////////////////////////////////////////////////////////
909 // primordial thread
910
911 // Check if current thread is the primordial thread, similar to Solaris thr_main.
912 bool os::is_primordial_thread(void) {
913 if (suppress_primordial_thread_resolution) {
914 return false;
915 }
916 char dummy;
917 // If called before init complete, thread stack bottom will be null.
918 // Can be called if fatal error occurs before initialization.
919 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
920 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
921 os::Linux::initial_thread_stack_size() != 0,
922 "os::init did not locate primordial thread's stack region");
923 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
924 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
925 os::Linux::initial_thread_stack_size()) {
926 return true;
927 } else {
928 return false;
929 }
930 }
931
932 // Find the virtual memory area that contains addr
933 static bool find_vma(address addr, address* vma_low, address* vma_high) {
934 FILE *fp = fopen("/proc/self/maps", "r");
935 if (fp) {
936 address low, high;
937 while (!feof(fp)) {
938 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
939 if (low <= addr && addr < high) {
940 if (vma_low) *vma_low = low;
941 if (vma_high) *vma_high = high;
942 fclose(fp);
943 return true;
944 }
945 }
946 for (;;) {
947 int ch = fgetc(fp);
948 if (ch == EOF || ch == (int)'\n') break;
949 }
950 }
951 fclose(fp);
952 }
953 return false;
954 }
955
956 // Locate primordial thread stack. This special handling of primordial thread stack
957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
958 // bogus value for the primordial process thread. While the launcher has created
959 // the VM in a new thread since JDK 6, we still have to allow for the use of the
960 // JNI invocation API from a primordial thread.
961 void os::Linux::capture_initial_stack(size_t max_size) {
962
963 // max_size is either 0 (which means accept OS default for thread stacks) or
964 // a user-specified value known to be at least the minimum needed. If we
965 // are actually on the primordial thread we can make it appear that we have a
966 // smaller max_size stack by inserting the guard pages at that location. But we
967 // cannot do anything to emulate a larger stack than what has been provided by
968 // the OS or threading library. In fact if we try to use a stack greater than
969 // what is set by rlimit then we will crash the hosting process.
970
971 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
972 // If this is "unlimited" then it will be a huge value.
973 struct rlimit rlim;
974 getrlimit(RLIMIT_STACK, &rlim);
975 size_t stack_size = rlim.rlim_cur;
976
977 // 6308388: a bug in ld.so will relocate its own .data section to the
978 // lower end of primordial stack; reduce ulimit -s value a little bit
979 // so we won't install guard page on ld.so's data section.
980 // But ensure we don't underflow the stack size - allow 1 page spare
981 if (stack_size >= (size_t)(3 * page_size())) {
982 stack_size -= 2 * page_size();
983 }
984
985 // Try to figure out where the stack base (top) is. This is harder.
986 //
987 // When an application is started, glibc saves the initial stack pointer in
988 // a global variable "__libc_stack_end", which is then used by system
989 // libraries. __libc_stack_end should be pretty close to stack top. The
990 // variable is available since the very early days. However, because it is
991 // a private interface, it could disappear in the future.
992 //
993 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
994 // to __libc_stack_end, it is very close to stack top, but isn't the real
995 // stack top. Note that /proc may not exist if VM is running as a chroot
996 // program, so reading /proc/<pid>/stat could fail. Also the contents of
997 // /proc/<pid>/stat could change in the future (though unlikely).
998 //
999 // We try __libc_stack_end first. If that doesn't work, look for
1000 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1001 // as a hint, which should work well in most cases.
1002
1003 uintptr_t stack_start;
1004
1005 // try __libc_stack_end first
1006 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1007 if (p && *p) {
1008 stack_start = *p;
1009 } else {
1010 // see if we can get the start_stack field from /proc/self/stat
1011 FILE *fp;
1012 int pid;
1013 char state;
1014 int ppid;
1015 int pgrp;
1016 int session;
1017 int nr;
1018 int tpgrp;
1019 unsigned long flags;
1020 unsigned long minflt;
1021 unsigned long cminflt;
1022 unsigned long majflt;
1023 unsigned long cmajflt;
1024 unsigned long utime;
1025 unsigned long stime;
1026 long cutime;
1027 long cstime;
1028 long prio;
1029 long nice;
1030 long junk;
1031 long it_real;
1032 uintptr_t start;
1033 uintptr_t vsize;
1034 intptr_t rss;
1035 uintptr_t rsslim;
1036 uintptr_t scodes;
1037 uintptr_t ecode;
1038 int i;
1039
1040 // Figure what the primordial thread stack base is. Code is inspired
1041 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1042 // followed by command name surrounded by parentheses, state, etc.
1043 char stat[2048];
1044 int statlen;
1045
1046 fp = fopen("/proc/self/stat", "r");
1047 if (fp) {
1048 statlen = fread(stat, 1, 2047, fp);
1049 stat[statlen] = '\0';
1050 fclose(fp);
1051
1052 // Skip pid and the command string. Note that we could be dealing with
1053 // weird command names, e.g. user could decide to rename java launcher
1054 // to "java 1.4.2 :)", then the stat file would look like
1055 // 1234 (java 1.4.2 :)) R ... ...
1056 // We don't really need to know the command string, just find the last
1057 // occurrence of ")" and then start parsing from there. See bug 4726580.
1058 char * s = strrchr(stat, ')');
1059
1060 i = 0;
1061 if (s) {
1062 // Skip blank chars
1063 do { s++; } while (s && isspace(*s));
1064
1065 #define _UFM UINTX_FORMAT
1066 #define _DFM INTX_FORMAT
1067
1068 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1069 // 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
1070 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1071 &state, // 3 %c
1072 &ppid, // 4 %d
1073 &pgrp, // 5 %d
1074 &session, // 6 %d
1075 &nr, // 7 %d
1076 &tpgrp, // 8 %d
1077 &flags, // 9 %lu
1078 &minflt, // 10 %lu
1079 &cminflt, // 11 %lu
1080 &majflt, // 12 %lu
1081 &cmajflt, // 13 %lu
1082 &utime, // 14 %lu
1083 &stime, // 15 %lu
1084 &cutime, // 16 %ld
1085 &cstime, // 17 %ld
1086 &prio, // 18 %ld
1087 &nice, // 19 %ld
1088 &junk, // 20 %ld
1089 &it_real, // 21 %ld
1090 &start, // 22 UINTX_FORMAT
1091 &vsize, // 23 UINTX_FORMAT
1092 &rss, // 24 INTX_FORMAT
1093 &rsslim, // 25 UINTX_FORMAT
1094 &scodes, // 26 UINTX_FORMAT
1095 &ecode, // 27 UINTX_FORMAT
1096 &stack_start); // 28 UINTX_FORMAT
1097 }
1098
1099 #undef _UFM
1100 #undef _DFM
1101
1102 if (i != 28 - 2) {
1103 assert(false, "Bad conversion from /proc/self/stat");
1104 // product mode - assume we are the primordial thread, good luck in the
1105 // embedded case.
1106 warning("Can't detect primordial thread stack location - bad conversion");
1107 stack_start = (uintptr_t) &rlim;
1108 }
1109 } else {
1110 // For some reason we can't open /proc/self/stat (for example, running on
1111 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1112 // most cases, so don't abort:
1113 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1114 stack_start = (uintptr_t) &rlim;
1115 }
1116 }
1117
1118 // Now we have a pointer (stack_start) very close to the stack top, the
1119 // next thing to do is to figure out the exact location of stack top. We
1120 // can find out the virtual memory area that contains stack_start by
1121 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1122 // and its upper limit is the real stack top. (again, this would fail if
1123 // running inside chroot, because /proc may not exist.)
1124
1125 uintptr_t stack_top;
1126 address low, high;
1127 if (find_vma((address)stack_start, &low, &high)) {
1128 // success, "high" is the true stack top. (ignore "low", because initial
1129 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1130 stack_top = (uintptr_t)high;
1131 } else {
1132 // failed, likely because /proc/self/maps does not exist
1133 warning("Can't detect primordial thread stack location - find_vma failed");
1134 // best effort: stack_start is normally within a few pages below the real
1135 // stack top, use it as stack top, and reduce stack size so we won't put
1136 // guard page outside stack.
1137 stack_top = stack_start;
1138 stack_size -= 16 * page_size();
1139 }
1140
1141 // stack_top could be partially down the page so align it
1142 stack_top = align_up(stack_top, page_size());
1143
1144 // Allowed stack value is minimum of max_size and what we derived from rlimit
1145 if (max_size > 0) {
1146 _initial_thread_stack_size = MIN2(max_size, stack_size);
1147 } else {
1148 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1149 // clamp it at 8MB as we do on Solaris
1150 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1151 }
1152 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1153 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1154
1155 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1156
1157 if (log_is_enabled(Info, os, thread)) {
1158 // See if we seem to be on primordial process thread
1159 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1160 uintptr_t(&rlim) < stack_top;
1161
1162 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1163 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1164 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1165 stack_top, intptr_t(_initial_thread_stack_bottom));
1166 }
1167 }
1168
1169 ////////////////////////////////////////////////////////////////////////////////
1170 // time support
1171
1172 #ifndef SUPPORTS_CLOCK_MONOTONIC
1173 #error "Build platform doesn't support clock_gettime and related functionality"
1174 #endif
1175
1176 // Time since start-up in seconds to a fine granularity.
1177 // Used by VMSelfDestructTimer and the MemProfiler.
1178 double os::elapsedTime() {
1179
1180 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1181 }
1182
1183 jlong os::elapsed_counter() {
1184 return javaTimeNanos() - initial_time_count;
1185 }
1186
1187 jlong os::elapsed_frequency() {
1188 return NANOSECS_PER_SEC; // nanosecond resolution
1189 }
1190
1191 bool os::supports_vtime() { return true; }
1192 bool os::enable_vtime() { return false; }
1193 bool os::vtime_enabled() { return false; }
1194
1195 double os::elapsedVTime() {
1196 struct rusage usage;
1197 int retval = getrusage(RUSAGE_THREAD, &usage);
1198 if (retval == 0) {
1199 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);
1200 } else {
1201 // better than nothing, but not much
1202 return elapsedTime();
1203 }
1204 }
1205
1206 jlong os::javaTimeMillis() {
1207 timeval time;
1208 int status = gettimeofday(&time, NULL);
1209 assert(status != -1, "linux error");
1210 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1211 }
1212
1213 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1214 timeval time;
1215 int status = gettimeofday(&time, NULL);
1216 assert(status != -1, "linux error");
1217 seconds = jlong(time.tv_sec);
1218 nanos = jlong(time.tv_usec) * 1000;
1219 }
1220
1221 void os::Linux::fast_thread_clock_init() {
1222 if (!UseLinuxPosixThreadCPUClocks) {
1223 return;
1224 }
1225 clockid_t clockid;
1226 struct timespec tp;
1227 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1228 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1229
1230 // Switch to using fast clocks for thread cpu time if
1231 // the clock_getres() returns 0 error code.
1232 // Note, that some kernels may support the current thread
1233 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1234 // returned by the pthread_getcpuclockid().
1235 // If the fast Posix clocks are supported then the clock_getres()
1236 // must return at least tp.tv_sec == 0 which means a resolution
1237 // better than 1 sec. This is extra check for reliability.
1238
1239 if (pthread_getcpuclockid_func &&
1240 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1241 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1242 _supports_fast_thread_cpu_time = true;
1243 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1244 }
1245 }
1246
1247 jlong os::javaTimeNanos() {
1248 if (os::supports_monotonic_clock()) {
1249 struct timespec tp;
1250 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1251 assert(status == 0, "gettime error");
1252 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1253 return result;
1254 } else {
1255 timeval time;
1256 int status = gettimeofday(&time, NULL);
1257 assert(status != -1, "linux error");
1258 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1259 return 1000 * usecs;
1260 }
1261 }
1262
1263 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1264 if (os::supports_monotonic_clock()) {
1265 info_ptr->max_value = ALL_64_BITS;
1266
1267 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1268 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1269 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1270 } else {
1271 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1272 info_ptr->max_value = ALL_64_BITS;
1273
1274 // gettimeofday is a real time clock so it skips
1275 info_ptr->may_skip_backward = true;
1276 info_ptr->may_skip_forward = true;
1277 }
1278
1279 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1280 }
1281
1282 // Return the real, user, and system times in seconds from an
1283 // arbitrary fixed point in the past.
1284 bool os::getTimesSecs(double* process_real_time,
1285 double* process_user_time,
1286 double* process_system_time) {
1287 struct tms ticks;
1288 clock_t real_ticks = times(&ticks);
1289
1290 if (real_ticks == (clock_t) (-1)) {
1291 return false;
1292 } else {
1293 double ticks_per_second = (double) clock_tics_per_sec;
1294 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1295 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1296 *process_real_time = ((double) real_ticks) / ticks_per_second;
1297
1298 return true;
1299 }
1300 }
1301
1302
1303 char * os::local_time_string(char *buf, size_t buflen) {
1304 struct tm t;
1305 time_t long_time;
1306 time(&long_time);
1307 localtime_r(&long_time, &t);
1308 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1309 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1310 t.tm_hour, t.tm_min, t.tm_sec);
1311 return buf;
1312 }
1313
1314 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1315 return localtime_r(clock, res);
1316 }
1317
1318 ////////////////////////////////////////////////////////////////////////////////
1319 // runtime exit support
1320
1321 // Note: os::shutdown() might be called very early during initialization, or
1322 // called from signal handler. Before adding something to os::shutdown(), make
1323 // sure it is async-safe and can handle partially initialized VM.
1324 void os::shutdown() {
1325
1326 // allow PerfMemory to attempt cleanup of any persistent resources
1327 perfMemory_exit();
1328
1329 // needs to remove object in file system
1330 AttachListener::abort();
1331
1332 // flush buffered output, finish log files
1333 ostream_abort();
1334
1335 // Check for abort hook
1336 abort_hook_t abort_hook = Arguments::abort_hook();
1337 if (abort_hook != NULL) {
1338 abort_hook();
1339 }
1340
1341 }
1342
1343 // Note: os::abort() might be called very early during initialization, or
1344 // called from signal handler. Before adding something to os::abort(), make
1345 // sure it is async-safe and can handle partially initialized VM.
1346 void os::abort(bool dump_core, void* siginfo, const void* context) {
1347 os::shutdown();
1348 if (dump_core) {
1349 #ifndef PRODUCT
1350 fdStream out(defaultStream::output_fd());
1351 out.print_raw("Current thread is ");
1352 char buf[16];
1353 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1354 out.print_raw_cr(buf);
1355 out.print_raw_cr("Dumping core ...");
1356 #endif
1357 ::abort(); // dump core
1358 }
1359
1360 ::exit(1);
1361 }
1362
1363 // Die immediately, no exit hook, no abort hook, no cleanup.
1364 void os::die() {
1365 ::abort();
1366 }
1367
1368 // thread_id is kernel thread id (similar to Solaris LWP id)
1369 intx os::current_thread_id() { return os::Linux::gettid(); }
1370 int os::current_process_id() {
1371 return ::getpid();
1372 }
1373
1374 // DLL functions
1375
1376 const char* os::dll_file_extension() { return ".so"; }
1377
1378 // This must be hard coded because it's the system's temporary
1379 // directory not the java application's temp directory, ala java.io.tmpdir.
1380 const char* os::get_temp_directory() { return "/tmp"; }
1381
1382 static bool file_exists(const char* filename) {
1383 struct stat statbuf;
1384 if (filename == NULL || strlen(filename) == 0) {
1385 return false;
1386 }
1387 return os::stat(filename, &statbuf) == 0;
1388 }
1389
1390 // check if addr is inside libjvm.so
1391 bool os::address_is_in_vm(address addr) {
1392 static address libjvm_base_addr;
1393 Dl_info dlinfo;
1394
1395 if (libjvm_base_addr == NULL) {
1396 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1397 libjvm_base_addr = (address)dlinfo.dli_fbase;
1398 }
1399 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1400 }
1401
1402 if (dladdr((void *)addr, &dlinfo) != 0) {
1403 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1404 }
1405
1406 return false;
1407 }
1408
1409 bool os::dll_address_to_function_name(address addr, char *buf,
1410 int buflen, int *offset,
1411 bool demangle) {
1412 // buf is not optional, but offset is optional
1413 assert(buf != NULL, "sanity check");
1414
1415 Dl_info dlinfo;
1416
1417 if (dladdr((void*)addr, &dlinfo) != 0) {
1418 // see if we have a matching symbol
1419 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1420 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1421 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1422 }
1423 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1424 return true;
1425 }
1426 // no matching symbol so try for just file info
1427 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1428 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1429 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1430 return true;
1431 }
1432 }
1433 }
1434
1435 buf[0] = '\0';
1436 if (offset != NULL) *offset = -1;
1437 return false;
1438 }
1439
1440 struct _address_to_library_name {
1441 address addr; // input : memory address
1442 size_t buflen; // size of fname
1443 char* fname; // output: library name
1444 address base; // library base addr
1445 };
1446
1447 static int address_to_library_name_callback(struct dl_phdr_info *info,
1448 size_t size, void *data) {
1449 int i;
1450 bool found = false;
1451 address libbase = NULL;
1452 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1453
1454 // iterate through all loadable segments
1455 for (i = 0; i < info->dlpi_phnum; i++) {
1456 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1457 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1458 // base address of a library is the lowest address of its loaded
1459 // segments.
1460 if (libbase == NULL || libbase > segbase) {
1461 libbase = segbase;
1462 }
1463 // see if 'addr' is within current segment
1464 if (segbase <= d->addr &&
1465 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1466 found = true;
1467 }
1468 }
1469 }
1470
1471 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1472 // so dll_address_to_library_name() can fall through to use dladdr() which
1473 // can figure out executable name from argv[0].
1474 if (found && info->dlpi_name && info->dlpi_name[0]) {
1475 d->base = libbase;
1476 if (d->fname) {
1477 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1478 }
1479 return 1;
1480 }
1481 return 0;
1482 }
1483
1484 bool os::dll_address_to_library_name(address addr, char* buf,
1485 int buflen, int* offset) {
1486 // buf is not optional, but offset is optional
1487 assert(buf != NULL, "sanity check");
1488
1489 Dl_info dlinfo;
1490 struct _address_to_library_name data;
1491
1492 // There is a bug in old glibc dladdr() implementation that it could resolve
1493 // to wrong library name if the .so file has a base address != NULL. Here
1494 // we iterate through the program headers of all loaded libraries to find
1495 // out which library 'addr' really belongs to. This workaround can be
1496 // removed once the minimum requirement for glibc is moved to 2.3.x.
1497 data.addr = addr;
1498 data.fname = buf;
1499 data.buflen = buflen;
1500 data.base = NULL;
1501 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1502
1503 if (rslt) {
1504 // buf already contains library name
1505 if (offset) *offset = addr - data.base;
1506 return true;
1507 }
1508 if (dladdr((void*)addr, &dlinfo) != 0) {
1509 if (dlinfo.dli_fname != NULL) {
1510 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1511 }
1512 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1513 *offset = addr - (address)dlinfo.dli_fbase;
1514 }
1515 return true;
1516 }
1517
1518 buf[0] = '\0';
1519 if (offset) *offset = -1;
1520 return false;
1521 }
1522
1523 // Loads .dll/.so and
1524 // in case of error it checks if .dll/.so was built for the
1525 // same architecture as Hotspot is running on
1526
1527
1528 // Remember the stack's state. The Linux dynamic linker will change
1529 // the stack to 'executable' at most once, so we must safepoint only once.
1530 bool os::Linux::_stack_is_executable = false;
1531
1532 // VM operation that loads a library. This is necessary if stack protection
1533 // of the Java stacks can be lost during loading the library. If we
1534 // do not stop the Java threads, they can stack overflow before the stacks
1535 // are protected again.
1536 class VM_LinuxDllLoad: public VM_Operation {
1537 private:
1538 const char *_filename;
1539 char *_ebuf;
1540 int _ebuflen;
1541 void *_lib;
1542 public:
1543 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1544 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1545 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1546 void doit() {
1547 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1548 os::Linux::_stack_is_executable = true;
1549 }
1550 void* loaded_library() { return _lib; }
1551 };
1552
1553 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1554 void * result = NULL;
1555 bool load_attempted = false;
1556
1557 // Check whether the library to load might change execution rights
1558 // of the stack. If they are changed, the protection of the stack
1559 // guard pages will be lost. We need a safepoint to fix this.
1560 //
1561 // See Linux man page execstack(8) for more info.
1562 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1563 if (!ElfFile::specifies_noexecstack(filename)) {
1564 if (!is_init_completed()) {
1565 os::Linux::_stack_is_executable = true;
1566 // This is OK - No Java threads have been created yet, and hence no
1567 // stack guard pages to fix.
1568 //
1569 // Dynamic loader will make all stacks executable after
1570 // this function returns, and will not do that again.
1571 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1572 } else {
1573 warning("You have loaded library %s which might have disabled stack guard. "
1574 "The VM will try to fix the stack guard now.\n"
1575 "It's highly recommended that you fix the library with "
1576 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1577 filename);
1578
1579 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1580 JavaThread *jt = JavaThread::current();
1581 if (jt->thread_state() != _thread_in_native) {
1582 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1583 // that requires ExecStack. Cannot enter safe point. Let's give up.
1584 warning("Unable to fix stack guard. Giving up.");
1585 } else {
1586 if (!LoadExecStackDllInVMThread) {
1587 // This is for the case where the DLL has an static
1588 // constructor function that executes JNI code. We cannot
1589 // load such DLLs in the VMThread.
1590 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1591 }
1592
1593 ThreadInVMfromNative tiv(jt);
1594 debug_only(VMNativeEntryWrapper vew;)
1595
1596 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1597 VMThread::execute(&op);
1598 if (LoadExecStackDllInVMThread) {
1599 result = op.loaded_library();
1600 }
1601 load_attempted = true;
1602 }
1603 }
1604 }
1605 }
1606
1607 if (!load_attempted) {
1608 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1609 }
1610
1611 if (result != NULL) {
1612 // Successful loading
1613 return result;
1614 }
1615
1616 Elf32_Ehdr elf_head;
1617 int diag_msg_max_length=ebuflen-strlen(ebuf);
1618 char* diag_msg_buf=ebuf+strlen(ebuf);
1619
1620 if (diag_msg_max_length==0) {
1621 // No more space in ebuf for additional diagnostics message
1622 return NULL;
1623 }
1624
1625
1626 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1627
1628 if (file_descriptor < 0) {
1629 // Can't open library, report dlerror() message
1630 return NULL;
1631 }
1632
1633 bool failed_to_read_elf_head=
1634 (sizeof(elf_head)!=
1635 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1636
1637 ::close(file_descriptor);
1638 if (failed_to_read_elf_head) {
1639 // file i/o error - report dlerror() msg
1640 return NULL;
1641 }
1642
1643 typedef struct {
1644 Elf32_Half code; // Actual value as defined in elf.h
1645 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1646 unsigned char elf_class; // 32 or 64 bit
1647 unsigned char endianess; // MSB or LSB
1648 char* name; // String representation
1649 } arch_t;
1650
1651 #ifndef EM_486
1652 #define EM_486 6 /* Intel 80486 */
1653 #endif
1654 #ifndef EM_AARCH64
1655 #define EM_AARCH64 183 /* ARM AARCH64 */
1656 #endif
1657
1658 static const arch_t arch_array[]={
1659 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1660 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1661 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1662 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1663 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1664 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1665 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1666 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1667 #if defined(VM_LITTLE_ENDIAN)
1668 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1669 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1670 #else
1671 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1672 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1673 #endif
1674 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1675 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1676 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1677 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1678 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1679 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1680 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1681 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1682 };
1683
1684 #if (defined IA32)
1685 static Elf32_Half running_arch_code=EM_386;
1686 #elif (defined AMD64) || (defined X32)
1687 static Elf32_Half running_arch_code=EM_X86_64;
1688 #elif (defined IA64)
1689 static Elf32_Half running_arch_code=EM_IA_64;
1690 #elif (defined __sparc) && (defined _LP64)
1691 static Elf32_Half running_arch_code=EM_SPARCV9;
1692 #elif (defined __sparc) && (!defined _LP64)
1693 static Elf32_Half running_arch_code=EM_SPARC;
1694 #elif (defined __powerpc64__)
1695 static Elf32_Half running_arch_code=EM_PPC64;
1696 #elif (defined __powerpc__)
1697 static Elf32_Half running_arch_code=EM_PPC;
1698 #elif (defined AARCH64)
1699 static Elf32_Half running_arch_code=EM_AARCH64;
1700 #elif (defined ARM)
1701 static Elf32_Half running_arch_code=EM_ARM;
1702 #elif (defined S390)
1703 static Elf32_Half running_arch_code=EM_S390;
1704 #elif (defined ALPHA)
1705 static Elf32_Half running_arch_code=EM_ALPHA;
1706 #elif (defined MIPSEL)
1707 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1708 #elif (defined PARISC)
1709 static Elf32_Half running_arch_code=EM_PARISC;
1710 #elif (defined MIPS)
1711 static Elf32_Half running_arch_code=EM_MIPS;
1712 #elif (defined M68K)
1713 static Elf32_Half running_arch_code=EM_68K;
1714 #elif (defined SH)
1715 static Elf32_Half running_arch_code=EM_SH;
1716 #else
1717 #error Method os::dll_load requires that one of following is defined:\
1718 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1719 #endif
1720
1721 // Identify compatability class for VM's architecture and library's architecture
1722 // Obtain string descriptions for architectures
1723
1724 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1725 int running_arch_index=-1;
1726
1727 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1728 if (running_arch_code == arch_array[i].code) {
1729 running_arch_index = i;
1730 }
1731 if (lib_arch.code == arch_array[i].code) {
1732 lib_arch.compat_class = arch_array[i].compat_class;
1733 lib_arch.name = arch_array[i].name;
1734 }
1735 }
1736
1737 assert(running_arch_index != -1,
1738 "Didn't find running architecture code (running_arch_code) in arch_array");
1739 if (running_arch_index == -1) {
1740 // Even though running architecture detection failed
1741 // we may still continue with reporting dlerror() message
1742 return NULL;
1743 }
1744
1745 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1746 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1747 return NULL;
1748 }
1749
1750 #ifndef S390
1751 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1752 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1753 return NULL;
1754 }
1755 #endif // !S390
1756
1757 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1758 if (lib_arch.name!=NULL) {
1759 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1760 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1761 lib_arch.name, arch_array[running_arch_index].name);
1762 } else {
1763 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1764 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1765 lib_arch.code,
1766 arch_array[running_arch_index].name);
1767 }
1768 }
1769
1770 return NULL;
1771 }
1772
1773 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1774 int ebuflen) {
1775 void * result = ::dlopen(filename, RTLD_LAZY);
1776 if (result == NULL) {
1777 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1778 ebuf[ebuflen-1] = '\0';
1779 }
1780 return result;
1781 }
1782
1783 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1784 int ebuflen) {
1785 void * result = NULL;
1786 if (LoadExecStackDllInVMThread) {
1787 result = dlopen_helper(filename, ebuf, ebuflen);
1788 }
1789
1790 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1791 // library that requires an executable stack, or which does not have this
1792 // stack attribute set, dlopen changes the stack attribute to executable. The
1793 // read protection of the guard pages gets lost.
1794 //
1795 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1796 // may have been queued at the same time.
1797
1798 if (!_stack_is_executable) {
1799 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1800 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1801 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1802 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1803 warning("Attempt to reguard stack yellow zone failed.");
1804 }
1805 }
1806 }
1807 }
1808
1809 return result;
1810 }
1811
1812 void* os::dll_lookup(void* handle, const char* name) {
1813 void* res = dlsym(handle, name);
1814 return res;
1815 }
1816
1817 void* os::get_default_process_handle() {
1818 return (void*)::dlopen(NULL, RTLD_LAZY);
1819 }
1820
1821 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1822 int fd = ::open(filename, O_RDONLY);
1823 if (fd == -1) {
1824 return false;
1825 }
1826
1827 if (hdr != NULL) {
1828 st->print_cr("%s", hdr);
1829 }
1830
1831 char buf[33];
1832 int bytes;
1833 buf[32] = '\0';
1834 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1835 st->print_raw(buf, bytes);
1836 }
1837
1838 ::close(fd);
1839
1840 return true;
1841 }
1842
1843 void os::print_dll_info(outputStream *st) {
1844 st->print_cr("Dynamic libraries:");
1845
1846 char fname[32];
1847 pid_t pid = os::Linux::gettid();
1848
1849 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1850
1851 if (!_print_ascii_file(fname, st)) {
1852 st->print("Can not get library information for pid = %d\n", pid);
1853 }
1854 }
1855
1856 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1857 FILE *procmapsFile = NULL;
1858
1859 // Open the procfs maps file for the current process
1860 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1861 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1862 char line[PATH_MAX + 100];
1863
1864 // Read line by line from 'file'
1865 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1866 u8 base, top, offset, inode;
1867 char permissions[5];
1868 char device[6];
1869 char name[PATH_MAX + 1];
1870
1871 // Parse fields from line
1872 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1873 &base, &top, permissions, &offset, device, &inode, name);
1874
1875 // Filter by device id '00:00' so that we only get file system mapped files.
1876 if (strcmp(device, "00:00") != 0) {
1877
1878 // Call callback with the fields of interest
1879 if(callback(name, (address)base, (address)top, param)) {
1880 // Oops abort, callback aborted
1881 fclose(procmapsFile);
1882 return 1;
1883 }
1884 }
1885 }
1886 fclose(procmapsFile);
1887 }
1888 return 0;
1889 }
1890
1891 void os::print_os_info_brief(outputStream* st) {
1892 os::Linux::print_distro_info(st);
1893
1894 os::Posix::print_uname_info(st);
1895
1896 os::Linux::print_libversion_info(st);
1897
1898 }
1899
1900 void os::print_os_info(outputStream* st) {
1901 st->print("OS:");
1902
1903 os::Linux::print_distro_info(st);
1904
1905 os::Posix::print_uname_info(st);
1906
1907 // Print warning if unsafe chroot environment detected
1908 if (unsafe_chroot_detected) {
1909 st->print("WARNING!! ");
1910 st->print_cr("%s", unstable_chroot_error);
1911 }
1912
1913 os::Linux::print_libversion_info(st);
1914
1915 os::Posix::print_rlimit_info(st);
1916
1917 os::Posix::print_load_average(st);
1918
1919 os::Linux::print_full_memory_info(st);
1920
1921 os::Linux::print_proc_sys_info(st);
1922
1923 os::Linux::print_ld_preload_file(st);
1924
1925 os::Linux::print_container_info(st);
1926 }
1927
1928 // Try to identify popular distros.
1929 // Most Linux distributions have a /etc/XXX-release file, which contains
1930 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1931 // file that also contains the OS version string. Some have more than one
1932 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1933 // /etc/redhat-release.), so the order is important.
1934 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1935 // their own specific XXX-release file as well as a redhat-release file.
1936 // Because of this the XXX-release file needs to be searched for before the
1937 // redhat-release file.
1938 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1939 // search for redhat-release / SuSE-release needs to be before lsb-release.
1940 // Since the lsb-release file is the new standard it needs to be searched
1941 // before the older style release files.
1942 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1943 // next to last resort. The os-release file is a new standard that contains
1944 // distribution information and the system-release file seems to be an old
1945 // standard that has been replaced by the lsb-release and os-release files.
1946 // Searching for the debian_version file is the last resort. It contains
1947 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1948 // "Debian " is printed before the contents of the debian_version file.
1949
1950 const char* distro_files[] = {
1951 "/etc/oracle-release",
1952 "/etc/mandriva-release",
1953 "/etc/mandrake-release",
1954 "/etc/sun-release",
1955 "/etc/redhat-release",
1956 "/etc/SuSE-release",
1957 "/etc/lsb-release",
1958 "/etc/turbolinux-release",
1959 "/etc/gentoo-release",
1960 "/etc/ltib-release",
1961 "/etc/angstrom-version",
1962 "/etc/system-release",
1963 "/etc/os-release",
1964 NULL };
1965
1966 void os::Linux::print_distro_info(outputStream* st) {
1967 for (int i = 0;; i++) {
1968 const char* file = distro_files[i];
1969 if (file == NULL) {
1970 break; // done
1971 }
1972 // If file prints, we found it.
1973 if (_print_ascii_file(file, st)) {
1974 return;
1975 }
1976 }
1977
1978 if (file_exists("/etc/debian_version")) {
1979 st->print("Debian ");
1980 _print_ascii_file("/etc/debian_version", st);
1981 } else {
1982 st->print("Linux");
1983 }
1984 st->cr();
1985 }
1986
1987 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
1988 char buf[256];
1989 while (fgets(buf, sizeof(buf), fp)) {
1990 // Edit out extra stuff in expected format
1991 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
1992 char* ptr = strstr(buf, "\""); // the name is in quotes
1993 if (ptr != NULL) {
1994 ptr++; // go beyond first quote
1995 char* nl = strchr(ptr, '\"');
1996 if (nl != NULL) *nl = '\0';
1997 strncpy(distro, ptr, length);
1998 } else {
1999 ptr = strstr(buf, "=");
2000 ptr++; // go beyond equals then
2001 char* nl = strchr(ptr, '\n');
2002 if (nl != NULL) *nl = '\0';
2003 strncpy(distro, ptr, length);
2004 }
2005 return;
2006 } else if (get_first_line) {
2007 char* nl = strchr(buf, '\n');
2008 if (nl != NULL) *nl = '\0';
2009 strncpy(distro, buf, length);
2010 return;
2011 }
2012 }
2013 // print last line and close
2014 char* nl = strchr(buf, '\n');
2015 if (nl != NULL) *nl = '\0';
2016 strncpy(distro, buf, length);
2017 }
2018
2019 static void parse_os_info(char* distro, size_t length, const char* file) {
2020 FILE* fp = fopen(file, "r");
2021 if (fp != NULL) {
2022 // if suse format, print out first line
2023 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2024 parse_os_info_helper(fp, distro, length, get_first_line);
2025 fclose(fp);
2026 }
2027 }
2028
2029 void os::get_summary_os_info(char* buf, size_t buflen) {
2030 for (int i = 0;; i++) {
2031 const char* file = distro_files[i];
2032 if (file == NULL) {
2033 break; // ran out of distro_files
2034 }
2035 if (file_exists(file)) {
2036 parse_os_info(buf, buflen, file);
2037 return;
2038 }
2039 }
2040 // special case for debian
2041 if (file_exists("/etc/debian_version")) {
2042 strncpy(buf, "Debian ", buflen);
2043 if (buflen > 7) {
2044 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2045 }
2046 } else {
2047 strncpy(buf, "Linux", buflen);
2048 }
2049 }
2050
2051 void os::Linux::print_libversion_info(outputStream* st) {
2052 // libc, pthread
2053 st->print("libc:");
2054 st->print("%s ", os::Linux::glibc_version());
2055 st->print("%s ", os::Linux::libpthread_version());
2056 st->cr();
2057 }
2058
2059 void os::Linux::print_proc_sys_info(outputStream* st) {
2060 st->cr();
2061 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2062 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2063 st->cr();
2064 st->cr();
2065
2066 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2067 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2068 st->cr();
2069 st->cr();
2070
2071 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2072 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2073 st->cr();
2074 st->cr();
2075 }
2076
2077 void os::Linux::print_full_memory_info(outputStream* st) {
2078 st->print("\n/proc/meminfo:\n");
2079 _print_ascii_file("/proc/meminfo", st);
2080 st->cr();
2081 }
2082
2083 void os::Linux::print_ld_preload_file(outputStream* st) {
2084 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2085 st->cr();
2086 }
2087
2088 void os::Linux::print_container_info(outputStream* st) {
2089 if (!OSContainer::is_containerized()) {
2090 return;
2091 }
2092
2093 st->print("container (cgroup) information:\n");
2094
2095 const char *p_ct = OSContainer::container_type();
2096 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2097
2098 char *p = OSContainer::cpu_cpuset_cpus();
2099 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2100 free(p);
2101
2102 p = OSContainer::cpu_cpuset_memory_nodes();
2103 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2104 free(p);
2105
2106 int i = OSContainer::active_processor_count();
2107 if (i > 0) {
2108 st->print("active_processor_count: %d\n", i);
2109 } else {
2110 st->print("active_processor_count: failed\n");
2111 }
2112
2113 i = OSContainer::cpu_quota();
2114 st->print("cpu_quota: %d\n", i);
2115
2116 i = OSContainer::cpu_period();
2117 st->print("cpu_period: %d\n", i);
2118
2119 i = OSContainer::cpu_shares();
2120 st->print("cpu_shares: %d\n", i);
2121
2122 jlong j = OSContainer::memory_limit_in_bytes();
2123 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2124
2125 j = OSContainer::memory_and_swap_limit_in_bytes();
2126 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2127
2128 j = OSContainer::memory_soft_limit_in_bytes();
2129 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2130
2131 j = OSContainer::OSContainer::memory_usage_in_bytes();
2132 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2133
2134 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2135 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2136 st->cr();
2137 }
2138
2139 void os::print_memory_info(outputStream* st) {
2140
2141 st->print("Memory:");
2142 st->print(" %dk page", os::vm_page_size()>>10);
2143
2144 // values in struct sysinfo are "unsigned long"
2145 struct sysinfo si;
2146 sysinfo(&si);
2147
2148 st->print(", physical " UINT64_FORMAT "k",
2149 os::physical_memory() >> 10);
2150 st->print("(" UINT64_FORMAT "k free)",
2151 os::available_memory() >> 10);
2152 st->print(", swap " UINT64_FORMAT "k",
2153 ((jlong)si.totalswap * si.mem_unit) >> 10);
2154 st->print("(" UINT64_FORMAT "k free)",
2155 ((jlong)si.freeswap * si.mem_unit) >> 10);
2156 st->cr();
2157 }
2158
2159 // Print the first "model name" line and the first "flags" line
2160 // that we find and nothing more. We assume "model name" comes
2161 // before "flags" so if we find a second "model name", then the
2162 // "flags" field is considered missing.
2163 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2164 #if defined(IA32) || defined(AMD64)
2165 // Other platforms have less repetitive cpuinfo files
2166 FILE *fp = fopen("/proc/cpuinfo", "r");
2167 if (fp) {
2168 while (!feof(fp)) {
2169 if (fgets(buf, buflen, fp)) {
2170 // Assume model name comes before flags
2171 bool model_name_printed = false;
2172 if (strstr(buf, "model name") != NULL) {
2173 if (!model_name_printed) {
2174 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2175 st->print_raw(buf);
2176 model_name_printed = true;
2177 } else {
2178 // model name printed but not flags? Odd, just return
2179 fclose(fp);
2180 return true;
2181 }
2182 }
2183 // print the flags line too
2184 if (strstr(buf, "flags") != NULL) {
2185 st->print_raw(buf);
2186 fclose(fp);
2187 return true;
2188 }
2189 }
2190 }
2191 fclose(fp);
2192 }
2193 #endif // x86 platforms
2194 return false;
2195 }
2196
2197 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2198 // Only print the model name if the platform provides this as a summary
2199 if (!print_model_name_and_flags(st, buf, buflen)) {
2200 st->print("\n/proc/cpuinfo:\n");
2201 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2202 st->print_cr(" <Not Available>");
2203 }
2204 }
2205 }
2206
2207 #if defined(AMD64) || defined(IA32) || defined(X32)
2208 const char* search_string = "model name";
2209 #elif defined(M68K)
2210 const char* search_string = "CPU";
2211 #elif defined(PPC64)
2212 const char* search_string = "cpu";
2213 #elif defined(S390)
2214 const char* search_string = "processor";
2215 #elif defined(SPARC)
2216 const char* search_string = "cpu";
2217 #else
2218 const char* search_string = "Processor";
2219 #endif
2220
2221 // Parses the cpuinfo file for string representing the model name.
2222 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2223 FILE* fp = fopen("/proc/cpuinfo", "r");
2224 if (fp != NULL) {
2225 while (!feof(fp)) {
2226 char buf[256];
2227 if (fgets(buf, sizeof(buf), fp)) {
2228 char* start = strstr(buf, search_string);
2229 if (start != NULL) {
2230 char *ptr = start + strlen(search_string);
2231 char *end = buf + strlen(buf);
2232 while (ptr != end) {
2233 // skip whitespace and colon for the rest of the name.
2234 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2235 break;
2236 }
2237 ptr++;
2238 }
2239 if (ptr != end) {
2240 // reasonable string, get rid of newline and keep the rest
2241 char* nl = strchr(buf, '\n');
2242 if (nl != NULL) *nl = '\0';
2243 strncpy(cpuinfo, ptr, length);
2244 fclose(fp);
2245 return;
2246 }
2247 }
2248 }
2249 }
2250 fclose(fp);
2251 }
2252 // cpuinfo not found or parsing failed, just print generic string. The entire
2253 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2254 #if defined(AARCH64)
2255 strncpy(cpuinfo, "AArch64", length);
2256 #elif defined(AMD64)
2257 strncpy(cpuinfo, "x86_64", length);
2258 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2259 strncpy(cpuinfo, "ARM", length);
2260 #elif defined(IA32)
2261 strncpy(cpuinfo, "x86_32", length);
2262 #elif defined(IA64)
2263 strncpy(cpuinfo, "IA64", length);
2264 #elif defined(PPC)
2265 strncpy(cpuinfo, "PPC64", length);
2266 #elif defined(S390)
2267 strncpy(cpuinfo, "S390", length);
2268 #elif defined(SPARC)
2269 strncpy(cpuinfo, "sparcv9", length);
2270 #elif defined(ZERO_LIBARCH)
2271 strncpy(cpuinfo, ZERO_LIBARCH, length);
2272 #else
2273 strncpy(cpuinfo, "unknown", length);
2274 #endif
2275 }
2276
2277 static void print_signal_handler(outputStream* st, int sig,
2278 char* buf, size_t buflen);
2279
2280 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2281 st->print_cr("Signal Handlers:");
2282 print_signal_handler(st, SIGSEGV, buf, buflen);
2283 print_signal_handler(st, SIGBUS , buf, buflen);
2284 print_signal_handler(st, SIGFPE , buf, buflen);
2285 print_signal_handler(st, SIGPIPE, buf, buflen);
2286 print_signal_handler(st, SIGXFSZ, buf, buflen);
2287 print_signal_handler(st, SIGILL , buf, buflen);
2288 print_signal_handler(st, SR_signum, buf, buflen);
2289 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2290 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2291 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2292 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2293 #if defined(PPC64)
2294 print_signal_handler(st, SIGTRAP, buf, buflen);
2295 #endif
2296 }
2297
2298 static char saved_jvm_path[MAXPATHLEN] = {0};
2299
2300 // Find the full path to the current module, libjvm.so
2301 void os::jvm_path(char *buf, jint buflen) {
2302 // Error checking.
2303 if (buflen < MAXPATHLEN) {
2304 assert(false, "must use a large-enough buffer");
2305 buf[0] = '\0';
2306 return;
2307 }
2308 // Lazy resolve the path to current module.
2309 if (saved_jvm_path[0] != 0) {
2310 strcpy(buf, saved_jvm_path);
2311 return;
2312 }
2313
2314 char dli_fname[MAXPATHLEN];
2315 bool ret = dll_address_to_library_name(
2316 CAST_FROM_FN_PTR(address, os::jvm_path),
2317 dli_fname, sizeof(dli_fname), NULL);
2318 assert(ret, "cannot locate libjvm");
2319 char *rp = NULL;
2320 if (ret && dli_fname[0] != '\0') {
2321 rp = os::Posix::realpath(dli_fname, buf, buflen);
2322 }
2323 if (rp == NULL) {
2324 return;
2325 }
2326
2327 if (Arguments::sun_java_launcher_is_altjvm()) {
2328 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2329 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2330 // If "/jre/lib/" appears at the right place in the string, then
2331 // assume we are installed in a JDK and we're done. Otherwise, check
2332 // for a JAVA_HOME environment variable and fix up the path so it
2333 // looks like libjvm.so is installed there (append a fake suffix
2334 // hotspot/libjvm.so).
2335 const char *p = buf + strlen(buf) - 1;
2336 for (int count = 0; p > buf && count < 5; ++count) {
2337 for (--p; p > buf && *p != '/'; --p)
2338 /* empty */ ;
2339 }
2340
2341 if (strncmp(p, "/jre/lib/", 9) != 0) {
2342 // Look for JAVA_HOME in the environment.
2343 char* java_home_var = ::getenv("JAVA_HOME");
2344 if (java_home_var != NULL && java_home_var[0] != 0) {
2345 char* jrelib_p;
2346 int len;
2347
2348 // Check the current module name "libjvm.so".
2349 p = strrchr(buf, '/');
2350 if (p == NULL) {
2351 return;
2352 }
2353 assert(strstr(p, "/libjvm") == p, "invalid library name");
2354
2355 rp = os::Posix::realpath(java_home_var, buf, buflen);
2356 if (rp == NULL) {
2357 return;
2358 }
2359
2360 // determine if this is a legacy image or modules image
2361 // modules image doesn't have "jre" subdirectory
2362 len = strlen(buf);
2363 assert(len < buflen, "Ran out of buffer room");
2364 jrelib_p = buf + len;
2365 snprintf(jrelib_p, buflen-len, "/jre/lib");
2366 if (0 != access(buf, F_OK)) {
2367 snprintf(jrelib_p, buflen-len, "/lib");
2368 }
2369
2370 if (0 == access(buf, F_OK)) {
2371 // Use current module name "libjvm.so"
2372 len = strlen(buf);
2373 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2374 } else {
2375 // Go back to path of .so
2376 rp = os::Posix::realpath(dli_fname, buf, buflen);
2377 if (rp == NULL) {
2378 return;
2379 }
2380 }
2381 }
2382 }
2383 }
2384
2385 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2386 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2387 }
2388
2389 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2390 // no prefix required, not even "_"
2391 }
2392
2393 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2394 // no suffix required
2395 }
2396
2397 ////////////////////////////////////////////////////////////////////////////////
2398 // sun.misc.Signal support
2399
2400 static volatile jint sigint_count = 0;
2401
2402 static void UserHandler(int sig, void *siginfo, void *context) {
2403 // 4511530 - sem_post is serialized and handled by the manager thread. When
2404 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2405 // don't want to flood the manager thread with sem_post requests.
2406 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2407 return;
2408 }
2409
2410 // Ctrl-C is pressed during error reporting, likely because the error
2411 // handler fails to abort. Let VM die immediately.
2412 if (sig == SIGINT && VMError::is_error_reported()) {
2413 os::die();
2414 }
2415
2416 os::signal_notify(sig);
2417 }
2418
2419 void* os::user_handler() {
2420 return CAST_FROM_FN_PTR(void*, UserHandler);
2421 }
2422
2423 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2424 struct timespec ts;
2425 // Semaphore's are always associated with CLOCK_REALTIME
2426 os::Posix::clock_gettime(CLOCK_REALTIME, &ts);
2427 // see os_posix.cpp for discussion on overflow checking
2428 if (sec >= MAX_SECS) {
2429 ts.tv_sec += MAX_SECS;
2430 ts.tv_nsec = 0;
2431 } else {
2432 ts.tv_sec += sec;
2433 ts.tv_nsec += nsec;
2434 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2435 ts.tv_nsec -= NANOSECS_PER_SEC;
2436 ++ts.tv_sec; // note: this must be <= max_secs
2437 }
2438 }
2439
2440 return ts;
2441 }
2442
2443 extern "C" {
2444 typedef void (*sa_handler_t)(int);
2445 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2446 }
2447
2448 void* os::signal(int signal_number, void* handler) {
2449 struct sigaction sigAct, oldSigAct;
2450
2451 sigfillset(&(sigAct.sa_mask));
2452 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2453 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2454
2455 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2456 // -1 means registration failed
2457 return (void *)-1;
2458 }
2459
2460 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2461 }
2462
2463 void os::signal_raise(int signal_number) {
2464 ::raise(signal_number);
2465 }
2466
2467 // The following code is moved from os.cpp for making this
2468 // code platform specific, which it is by its very nature.
2469
2470 // Will be modified when max signal is changed to be dynamic
2471 int os::sigexitnum_pd() {
2472 return NSIG;
2473 }
2474
2475 // a counter for each possible signal value
2476 static volatile jint pending_signals[NSIG+1] = { 0 };
2477
2478 // Linux(POSIX) specific hand shaking semaphore.
2479 static Semaphore* sig_sem = NULL;
2480 static PosixSemaphore sr_semaphore;
2481
2482 static void jdk_misc_signal_init() {
2483 // Initialize signal structures
2484 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2485
2486 // Initialize signal semaphore
2487 sig_sem = new Semaphore();
2488 }
2489
2490 void os::signal_notify(int sig) {
2491 if (sig_sem != NULL) {
2492 Atomic::inc(&pending_signals[sig]);
2493 sig_sem->signal();
2494 } else {
2495 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2496 // initialization isn't called.
2497 assert(ReduceSignalUsage, "signal semaphore should be created");
2498 }
2499 }
2500
2501 static int check_pending_signals() {
2502 Atomic::store(0, &sigint_count);
2503 for (;;) {
2504 for (int i = 0; i < NSIG + 1; i++) {
2505 jint n = pending_signals[i];
2506 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2507 return i;
2508 }
2509 }
2510 JavaThread *thread = JavaThread::current();
2511 ThreadBlockInVM tbivm(thread);
2512
2513 bool threadIsSuspended;
2514 do {
2515 thread->set_suspend_equivalent();
2516 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2517 sig_sem->wait();
2518
2519 // were we externally suspended while we were waiting?
2520 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2521 if (threadIsSuspended) {
2522 // The semaphore has been incremented, but while we were waiting
2523 // another thread suspended us. We don't want to continue running
2524 // while suspended because that would surprise the thread that
2525 // suspended us.
2526 sig_sem->signal();
2527
2528 thread->java_suspend_self();
2529 }
2530 } while (threadIsSuspended);
2531 }
2532 }
2533
2534 int os::signal_wait() {
2535 return check_pending_signals();
2536 }
2537
2538 ////////////////////////////////////////////////////////////////////////////////
2539 // Virtual Memory
2540
2541 int os::vm_page_size() {
2542 // Seems redundant as all get out
2543 assert(os::Linux::page_size() != -1, "must call os::init");
2544 return os::Linux::page_size();
2545 }
2546
2547 // Solaris allocates memory by pages.
2548 int os::vm_allocation_granularity() {
2549 assert(os::Linux::page_size() != -1, "must call os::init");
2550 return os::Linux::page_size();
2551 }
2552
2553 // Rationale behind this function:
2554 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2555 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2556 // samples for JITted code. Here we create private executable mapping over the code cache
2557 // and then we can use standard (well, almost, as mapping can change) way to provide
2558 // info for the reporting script by storing timestamp and location of symbol
2559 void linux_wrap_code(char* base, size_t size) {
2560 static volatile jint cnt = 0;
2561
2562 if (!UseOprofile) {
2563 return;
2564 }
2565
2566 char buf[PATH_MAX+1];
2567 int num = Atomic::add(1, &cnt);
2568
2569 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2570 os::get_temp_directory(), os::current_process_id(), num);
2571 unlink(buf);
2572
2573 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2574
2575 if (fd != -1) {
2576 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2577 if (rv != (off_t)-1) {
2578 if (::write(fd, "", 1) == 1) {
2579 mmap(base, size,
2580 PROT_READ|PROT_WRITE|PROT_EXEC,
2581 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2582 }
2583 }
2584 ::close(fd);
2585 unlink(buf);
2586 }
2587 }
2588
2589 static bool recoverable_mmap_error(int err) {
2590 // See if the error is one we can let the caller handle. This
2591 // list of errno values comes from JBS-6843484. I can't find a
2592 // Linux man page that documents this specific set of errno
2593 // values so while this list currently matches Solaris, it may
2594 // change as we gain experience with this failure mode.
2595 switch (err) {
2596 case EBADF:
2597 case EINVAL:
2598 case ENOTSUP:
2599 // let the caller deal with these errors
2600 return true;
2601
2602 default:
2603 // Any remaining errors on this OS can cause our reserved mapping
2604 // to be lost. That can cause confusion where different data
2605 // structures think they have the same memory mapped. The worst
2606 // scenario is if both the VM and a library think they have the
2607 // same memory mapped.
2608 return false;
2609 }
2610 }
2611
2612 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2613 int err) {
2614 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2615 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2616 os::strerror(err), err);
2617 }
2618
2619 static void warn_fail_commit_memory(char* addr, size_t size,
2620 size_t alignment_hint, bool exec,
2621 int err) {
2622 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2623 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2624 alignment_hint, exec, os::strerror(err), err);
2625 }
2626
2627 // NOTE: Linux kernel does not really reserve the pages for us.
2628 // All it does is to check if there are enough free pages
2629 // left at the time of mmap(). This could be a potential
2630 // problem.
2631 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2632 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2633 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2634 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2635 if (res != (uintptr_t) MAP_FAILED) {
2636 if (UseNUMAInterleaving) {
2637 numa_make_global(addr, size);
2638 }
2639 return 0;
2640 }
2641
2642 int err = errno; // save errno from mmap() call above
2643
2644 if (!recoverable_mmap_error(err)) {
2645 warn_fail_commit_memory(addr, size, exec, err);
2646 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2647 }
2648
2649 return err;
2650 }
2651
2652 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2653 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2654 }
2655
2656 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2657 const char* mesg) {
2658 assert(mesg != NULL, "mesg must be specified");
2659 int err = os::Linux::commit_memory_impl(addr, size, exec);
2660 if (err != 0) {
2661 // the caller wants all commit errors to exit with the specified mesg:
2662 warn_fail_commit_memory(addr, size, exec, err);
2663 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2664 }
2665 }
2666
2667 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2668 #ifndef MAP_HUGETLB
2669 #define MAP_HUGETLB 0x40000
2670 #endif
2671
2672 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2673 #ifndef MADV_HUGEPAGE
2674 #define MADV_HUGEPAGE 14
2675 #endif
2676
2677 int os::Linux::commit_memory_impl(char* addr, size_t size,
2678 size_t alignment_hint, bool exec) {
2679 int err = os::Linux::commit_memory_impl(addr, size, exec);
2680 if (err == 0) {
2681 realign_memory(addr, size, alignment_hint);
2682 }
2683 return err;
2684 }
2685
2686 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2687 bool exec) {
2688 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2689 }
2690
2691 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2692 size_t alignment_hint, bool exec,
2693 const char* mesg) {
2694 assert(mesg != NULL, "mesg must be specified");
2695 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2696 if (err != 0) {
2697 // the caller wants all commit errors to exit with the specified mesg:
2698 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2699 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2700 }
2701 }
2702
2703 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2704 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2705 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2706 // be supported or the memory may already be backed by huge pages.
2707 ::madvise(addr, bytes, MADV_HUGEPAGE);
2708 }
2709 }
2710
2711 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2712 // This method works by doing an mmap over an existing mmaping and effectively discarding
2713 // the existing pages. However it won't work for SHM-based large pages that cannot be
2714 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2715 // small pages on top of the SHM segment. This method always works for small pages, so we
2716 // allow that in any case.
2717 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2718 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2719 }
2720 }
2721
2722 void os::numa_make_global(char *addr, size_t bytes) {
2723 Linux::numa_interleave_memory(addr, bytes);
2724 }
2725
2726 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2727 // bind policy to MPOL_PREFERRED for the current thread.
2728 #define USE_MPOL_PREFERRED 0
2729
2730 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2731 // To make NUMA and large pages more robust when both enabled, we need to ease
2732 // the requirements on where the memory should be allocated. MPOL_BIND is the
2733 // default policy and it will force memory to be allocated on the specified
2734 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2735 // the specified node, but will not force it. Using this policy will prevent
2736 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2737 // free large pages.
2738 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2739 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2740 }
2741
2742 bool os::numa_topology_changed() { return false; }
2743
2744 size_t os::numa_get_groups_num() {
2745 // Return just the number of nodes in which it's possible to allocate memory
2746 // (in numa terminology, configured nodes).
2747 return Linux::numa_num_configured_nodes();
2748 }
2749
2750 int os::numa_get_group_id() {
2751 int cpu_id = Linux::sched_getcpu();
2752 if (cpu_id != -1) {
2753 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2754 if (lgrp_id != -1) {
2755 return lgrp_id;
2756 }
2757 }
2758 return 0;
2759 }
2760
2761 int os::Linux::get_existing_num_nodes() {
2762 int node;
2763 int highest_node_number = Linux::numa_max_node();
2764 int num_nodes = 0;
2765
2766 // Get the total number of nodes in the system including nodes without memory.
2767 for (node = 0; node <= highest_node_number; node++) {
2768 if (isnode_in_existing_nodes(node)) {
2769 num_nodes++;
2770 }
2771 }
2772 return num_nodes;
2773 }
2774
2775 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2776 int highest_node_number = Linux::numa_max_node();
2777 size_t i = 0;
2778
2779 // Map all node ids in which it is possible to allocate memory. Also nodes are
2780 // not always consecutively available, i.e. available from 0 to the highest
2781 // node number. If the nodes have been bound explicitly using numactl membind,
2782 // then allocate memory from those nodes only.
2783 for (int node = 0; node <= highest_node_number; node++) {
2784 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2785 ids[i++] = node;
2786 }
2787 }
2788 return i;
2789 }
2790
2791 bool os::get_page_info(char *start, page_info* info) {
2792 return false;
2793 }
2794
2795 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2796 page_info* page_found) {
2797 return end;
2798 }
2799
2800
2801 int os::Linux::sched_getcpu_syscall(void) {
2802 unsigned int cpu = 0;
2803 int retval = -1;
2804
2805 #if defined(IA32)
2806 #ifndef SYS_getcpu
2807 #define SYS_getcpu 318
2808 #endif
2809 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2810 #elif defined(AMD64)
2811 // Unfortunately we have to bring all these macros here from vsyscall.h
2812 // to be able to compile on old linuxes.
2813 #define __NR_vgetcpu 2
2814 #define VSYSCALL_START (-10UL << 20)
2815 #define VSYSCALL_SIZE 1024
2816 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2817 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2818 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2819 retval = vgetcpu(&cpu, NULL, NULL);
2820 #endif
2821
2822 return (retval == -1) ? retval : cpu;
2823 }
2824
2825 void os::Linux::sched_getcpu_init() {
2826 // sched_getcpu() should be in libc.
2827 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2828 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2829
2830 // If it's not, try a direct syscall.
2831 if (sched_getcpu() == -1) {
2832 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2833 (void*)&sched_getcpu_syscall));
2834 }
2835
2836 if (sched_getcpu() == -1) {
2837 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2838 }
2839 }
2840
2841 // Something to do with the numa-aware allocator needs these symbols
2842 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2843 extern "C" JNIEXPORT void numa_error(char *where) { }
2844
2845 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2846 // load symbol from base version instead.
2847 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2848 void *f = dlvsym(handle, name, "libnuma_1.1");
2849 if (f == NULL) {
2850 f = dlsym(handle, name);
2851 }
2852 return f;
2853 }
2854
2855 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2856 // Return NULL if the symbol is not defined in this particular version.
2857 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2858 return dlvsym(handle, name, "libnuma_1.2");
2859 }
2860
2861 bool os::Linux::libnuma_init() {
2862 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2863 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2864 if (handle != NULL) {
2865 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2866 libnuma_dlsym(handle, "numa_node_to_cpus")));
2867 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2868 libnuma_dlsym(handle, "numa_max_node")));
2869 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2870 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2871 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2872 libnuma_dlsym(handle, "numa_available")));
2873 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2874 libnuma_dlsym(handle, "numa_tonode_memory")));
2875 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2876 libnuma_dlsym(handle, "numa_interleave_memory")));
2877 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2878 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2879 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2880 libnuma_dlsym(handle, "numa_set_bind_policy")));
2881 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2882 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2883 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2884 libnuma_dlsym(handle, "numa_distance")));
2885 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2886 libnuma_v2_dlsym(handle, "numa_get_membind")));
2887
2888 if (numa_available() != -1) {
2889 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2890 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2891 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2892 // Create an index -> node mapping, since nodes are not always consecutive
2893 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2894 rebuild_nindex_to_node_map();
2895 // Create a cpu -> node mapping
2896 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2897 rebuild_cpu_to_node_map();
2898 return true;
2899 }
2900 }
2901 }
2902 return false;
2903 }
2904
2905 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2906 // Creating guard page is very expensive. Java thread has HotSpot
2907 // guard pages, only enable glibc guard page for non-Java threads.
2908 // (Remember: compiler thread is a Java thread, too!)
2909 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2910 }
2911
2912 void os::Linux::rebuild_nindex_to_node_map() {
2913 int highest_node_number = Linux::numa_max_node();
2914
2915 nindex_to_node()->clear();
2916 for (int node = 0; node <= highest_node_number; node++) {
2917 if (Linux::isnode_in_existing_nodes(node)) {
2918 nindex_to_node()->append(node);
2919 }
2920 }
2921 }
2922
2923 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2924 // The table is later used in get_node_by_cpu().
2925 void os::Linux::rebuild_cpu_to_node_map() {
2926 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2927 // in libnuma (possible values are starting from 16,
2928 // and continuing up with every other power of 2, but less
2929 // than the maximum number of CPUs supported by kernel), and
2930 // is a subject to change (in libnuma version 2 the requirements
2931 // are more reasonable) we'll just hardcode the number they use
2932 // in the library.
2933 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2934
2935 size_t cpu_num = processor_count();
2936 size_t cpu_map_size = NCPUS / BitsPerCLong;
2937 size_t cpu_map_valid_size =
2938 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2939
2940 cpu_to_node()->clear();
2941 cpu_to_node()->at_grow(cpu_num - 1);
2942
2943 size_t node_num = get_existing_num_nodes();
2944
2945 int distance = 0;
2946 int closest_distance = INT_MAX;
2947 int closest_node = 0;
2948 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2949 for (size_t i = 0; i < node_num; i++) {
2950 // Check if node is configured (not a memory-less node). If it is not, find
2951 // the closest configured node. Check also if node is bound, i.e. it's allowed
2952 // to allocate memory from the node. If it's not allowed, map cpus in that node
2953 // to the closest node from which memory allocation is allowed.
2954 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
2955 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
2956 closest_distance = INT_MAX;
2957 // Check distance from all remaining nodes in the system. Ignore distance
2958 // from itself, from another non-configured node, and from another non-bound
2959 // node.
2960 for (size_t m = 0; m < node_num; m++) {
2961 if (m != i &&
2962 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
2963 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
2964 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2965 // If a closest node is found, update. There is always at least one
2966 // configured and bound node in the system so there is always at least
2967 // one node close.
2968 if (distance != 0 && distance < closest_distance) {
2969 closest_distance = distance;
2970 closest_node = nindex_to_node()->at(m);
2971 }
2972 }
2973 }
2974 } else {
2975 // Current node is already a configured node.
2976 closest_node = nindex_to_node()->at(i);
2977 }
2978
2979 // Get cpus from the original node and map them to the closest node. If node
2980 // is a configured node (not a memory-less node), then original node and
2981 // closest node are the same.
2982 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2983 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2984 if (cpu_map[j] != 0) {
2985 for (size_t k = 0; k < BitsPerCLong; k++) {
2986 if (cpu_map[j] & (1UL << k)) {
2987 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2988 }
2989 }
2990 }
2991 }
2992 }
2993 }
2994 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2995 }
2996
2997 int os::Linux::get_node_by_cpu(int cpu_id) {
2998 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2999 return cpu_to_node()->at(cpu_id);
3000 }
3001 return -1;
3002 }
3003
3004 GrowableArray<int>* os::Linux::_cpu_to_node;
3005 GrowableArray<int>* os::Linux::_nindex_to_node;
3006 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3007 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3008 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3009 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3010 os::Linux::numa_available_func_t os::Linux::_numa_available;
3011 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3012 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3013 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3014 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3015 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3016 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3017 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3018 unsigned long* os::Linux::_numa_all_nodes;
3019 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3020 struct bitmask* os::Linux::_numa_nodes_ptr;
3021
3022 bool os::pd_uncommit_memory(char* addr, size_t size) {
3023 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3024 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3025 return res != (uintptr_t) MAP_FAILED;
3026 }
3027
3028 static address get_stack_commited_bottom(address bottom, size_t size) {
3029 address nbot = bottom;
3030 address ntop = bottom + size;
3031
3032 size_t page_sz = os::vm_page_size();
3033 unsigned pages = size / page_sz;
3034
3035 unsigned char vec[1];
3036 unsigned imin = 1, imax = pages + 1, imid;
3037 int mincore_return_value = 0;
3038
3039 assert(imin <= imax, "Unexpected page size");
3040
3041 while (imin < imax) {
3042 imid = (imax + imin) / 2;
3043 nbot = ntop - (imid * page_sz);
3044
3045 // Use a trick with mincore to check whether the page is mapped or not.
3046 // mincore sets vec to 1 if page resides in memory and to 0 if page
3047 // is swapped output but if page we are asking for is unmapped
3048 // it returns -1,ENOMEM
3049 mincore_return_value = mincore(nbot, page_sz, vec);
3050
3051 if (mincore_return_value == -1) {
3052 // Page is not mapped go up
3053 // to find first mapped page
3054 if (errno != EAGAIN) {
3055 assert(errno == ENOMEM, "Unexpected mincore errno");
3056 imax = imid;
3057 }
3058 } else {
3059 // Page is mapped go down
3060 // to find first not mapped page
3061 imin = imid + 1;
3062 }
3063 }
3064
3065 nbot = nbot + page_sz;
3066
3067 // Adjust stack bottom one page up if last checked page is not mapped
3068 if (mincore_return_value == -1) {
3069 nbot = nbot + page_sz;
3070 }
3071
3072 return nbot;
3073 }
3074
3075 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3076 int mincore_return_value;
3077 const size_t stripe = 1024; // query this many pages each time
3078 unsigned char vec[stripe + 1];
3079 // set a guard
3080 vec[stripe] = 'X';
3081
3082 const size_t page_sz = os::vm_page_size();
3083 size_t pages = size / page_sz;
3084
3085 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3086 assert(is_aligned(size, page_sz), "Size must be page aligned");
3087
3088 committed_start = NULL;
3089
3090 int loops = (pages + stripe - 1) / stripe;
3091 int committed_pages = 0;
3092 address loop_base = start;
3093 bool found_range = false;
3094
3095 for (int index = 0; index < loops && !found_range; index ++) {
3096 assert(pages > 0, "Nothing to do");
3097 int pages_to_query = (pages >= stripe) ? stripe : pages;
3098 pages -= pages_to_query;
3099
3100 // Get stable read
3101 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3102
3103 // During shutdown, some memory goes away without properly notifying NMT,
3104 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3105 // Bailout and return as not committed for now.
3106 if (mincore_return_value == -1 && errno == ENOMEM) {
3107 return false;
3108 }
3109
3110 assert(vec[stripe] == 'X', "overflow guard");
3111 assert(mincore_return_value == 0, "Range must be valid");
3112 // Process this stripe
3113 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3114 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3115 // End of current contiguous region
3116 if (committed_start != NULL) {
3117 found_range = true;
3118 break;
3119 }
3120 } else { // committed
3121 // Start of region
3122 if (committed_start == NULL) {
3123 committed_start = loop_base + page_sz * vecIdx;
3124 }
3125 committed_pages ++;
3126 }
3127 }
3128
3129 loop_base += pages_to_query * page_sz;
3130 }
3131
3132 if (committed_start != NULL) {
3133 assert(committed_pages > 0, "Must have committed region");
3134 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3135 assert(committed_start >= start && committed_start < start + size, "Out of range");
3136 committed_size = page_sz * committed_pages;
3137 return true;
3138 } else {
3139 assert(committed_pages == 0, "Should not have committed region");
3140 return false;
3141 }
3142 }
3143
3144
3145 // Linux uses a growable mapping for the stack, and if the mapping for
3146 // the stack guard pages is not removed when we detach a thread the
3147 // stack cannot grow beyond the pages where the stack guard was
3148 // mapped. If at some point later in the process the stack expands to
3149 // that point, the Linux kernel cannot expand the stack any further
3150 // because the guard pages are in the way, and a segfault occurs.
3151 //
3152 // However, it's essential not to split the stack region by unmapping
3153 // a region (leaving a hole) that's already part of the stack mapping,
3154 // so if the stack mapping has already grown beyond the guard pages at
3155 // the time we create them, we have to truncate the stack mapping.
3156 // So, we need to know the extent of the stack mapping when
3157 // create_stack_guard_pages() is called.
3158
3159 // We only need this for stacks that are growable: at the time of
3160 // writing thread stacks don't use growable mappings (i.e. those
3161 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3162 // only applies to the main thread.
3163
3164 // If the (growable) stack mapping already extends beyond the point
3165 // where we're going to put our guard pages, truncate the mapping at
3166 // that point by munmap()ping it. This ensures that when we later
3167 // munmap() the guard pages we don't leave a hole in the stack
3168 // mapping. This only affects the main/primordial thread
3169
3170 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3171 if (os::is_primordial_thread()) {
3172 // As we manually grow stack up to bottom inside create_attached_thread(),
3173 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3174 // we don't need to do anything special.
3175 // Check it first, before calling heavy function.
3176 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3177 unsigned char vec[1];
3178
3179 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3180 // Fallback to slow path on all errors, including EAGAIN
3181 stack_extent = (uintptr_t) get_stack_commited_bottom(
3182 os::Linux::initial_thread_stack_bottom(),
3183 (size_t)addr - stack_extent);
3184 }
3185
3186 if (stack_extent < (uintptr_t)addr) {
3187 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3188 }
3189 }
3190
3191 return os::commit_memory(addr, size, !ExecMem);
3192 }
3193
3194 // If this is a growable mapping, remove the guard pages entirely by
3195 // munmap()ping them. If not, just call uncommit_memory(). This only
3196 // affects the main/primordial thread, but guard against future OS changes.
3197 // It's safe to always unmap guard pages for primordial thread because we
3198 // always place it right after end of the mapped region.
3199
3200 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3201 uintptr_t stack_extent, stack_base;
3202
3203 if (os::is_primordial_thread()) {
3204 return ::munmap(addr, size) == 0;
3205 }
3206
3207 return os::uncommit_memory(addr, size);
3208 }
3209
3210 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3211 // at 'requested_addr'. If there are existing memory mappings at the same
3212 // location, however, they will be overwritten. If 'fixed' is false,
3213 // 'requested_addr' is only treated as a hint, the return value may or
3214 // may not start from the requested address. Unlike Linux mmap(), this
3215 // function returns NULL to indicate failure.
3216 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3217 char * addr;
3218 int flags;
3219
3220 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3221 if (fixed) {
3222 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3223 flags |= MAP_FIXED;
3224 }
3225
3226 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3227 // touch an uncommitted page. Otherwise, the read/write might
3228 // succeed if we have enough swap space to back the physical page.
3229 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3230 flags, -1, 0);
3231
3232 return addr == MAP_FAILED ? NULL : addr;
3233 }
3234
3235 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3236 // (req_addr != NULL) or with a given alignment.
3237 // - bytes shall be a multiple of alignment.
3238 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3239 // - alignment sets the alignment at which memory shall be allocated.
3240 // It must be a multiple of allocation granularity.
3241 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3242 // req_addr or NULL.
3243 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3244
3245 size_t extra_size = bytes;
3246 if (req_addr == NULL && alignment > 0) {
3247 extra_size += alignment;
3248 }
3249
3250 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3251 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3252 -1, 0);
3253 if (start == MAP_FAILED) {
3254 start = NULL;
3255 } else {
3256 if (req_addr != NULL) {
3257 if (start != req_addr) {
3258 ::munmap(start, extra_size);
3259 start = NULL;
3260 }
3261 } else {
3262 char* const start_aligned = align_up(start, alignment);
3263 char* const end_aligned = start_aligned + bytes;
3264 char* const end = start + extra_size;
3265 if (start_aligned > start) {
3266 ::munmap(start, start_aligned - start);
3267 }
3268 if (end_aligned < end) {
3269 ::munmap(end_aligned, end - end_aligned);
3270 }
3271 start = start_aligned;
3272 }
3273 }
3274 return start;
3275 }
3276
3277 static int anon_munmap(char * addr, size_t size) {
3278 return ::munmap(addr, size) == 0;
3279 }
3280
3281 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3282 size_t alignment_hint) {
3283 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3284 }
3285
3286 bool os::pd_release_memory(char* addr, size_t size) {
3287 return anon_munmap(addr, size);
3288 }
3289
3290 static bool linux_mprotect(char* addr, size_t size, int prot) {
3291 // Linux wants the mprotect address argument to be page aligned.
3292 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3293
3294 // According to SUSv3, mprotect() should only be used with mappings
3295 // established by mmap(), and mmap() always maps whole pages. Unaligned
3296 // 'addr' likely indicates problem in the VM (e.g. trying to change
3297 // protection of malloc'ed or statically allocated memory). Check the
3298 // caller if you hit this assert.
3299 assert(addr == bottom, "sanity check");
3300
3301 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3302 return ::mprotect(bottom, size, prot) == 0;
3303 }
3304
3305 // Set protections specified
3306 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3307 bool is_committed) {
3308 unsigned int p = 0;
3309 switch (prot) {
3310 case MEM_PROT_NONE: p = PROT_NONE; break;
3311 case MEM_PROT_READ: p = PROT_READ; break;
3312 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3313 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3314 default:
3315 ShouldNotReachHere();
3316 }
3317 // is_committed is unused.
3318 return linux_mprotect(addr, bytes, p);
3319 }
3320
3321 bool os::guard_memory(char* addr, size_t size) {
3322 return linux_mprotect(addr, size, PROT_NONE);
3323 }
3324
3325 bool os::unguard_memory(char* addr, size_t size) {
3326 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3327 }
3328
3329 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3330 size_t page_size) {
3331 bool result = false;
3332 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3333 MAP_ANONYMOUS|MAP_PRIVATE,
3334 -1, 0);
3335 if (p != MAP_FAILED) {
3336 void *aligned_p = align_up(p, page_size);
3337
3338 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3339
3340 munmap(p, page_size * 2);
3341 }
3342
3343 if (warn && !result) {
3344 warning("TransparentHugePages is not supported by the operating system.");
3345 }
3346
3347 return result;
3348 }
3349
3350 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3351 bool result = false;
3352 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3353 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3354 -1, 0);
3355
3356 if (p != MAP_FAILED) {
3357 // We don't know if this really is a huge page or not.
3358 FILE *fp = fopen("/proc/self/maps", "r");
3359 if (fp) {
3360 while (!feof(fp)) {
3361 char chars[257];
3362 long x = 0;
3363 if (fgets(chars, sizeof(chars), fp)) {
3364 if (sscanf(chars, "%lx-%*x", &x) == 1
3365 && x == (long)p) {
3366 if (strstr (chars, "hugepage")) {
3367 result = true;
3368 break;
3369 }
3370 }
3371 }
3372 }
3373 fclose(fp);
3374 }
3375 munmap(p, page_size);
3376 }
3377
3378 if (warn && !result) {
3379 warning("HugeTLBFS is not supported by the operating system.");
3380 }
3381
3382 return result;
3383 }
3384
3385 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3386 //
3387 // From the coredump_filter documentation:
3388 //
3389 // - (bit 0) anonymous private memory
3390 // - (bit 1) anonymous shared memory
3391 // - (bit 2) file-backed private memory
3392 // - (bit 3) file-backed shared memory
3393 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3394 // effective only if the bit 2 is cleared)
3395 // - (bit 5) hugetlb private memory
3396 // - (bit 6) hugetlb shared memory
3397 // - (bit 7) dax private memory
3398 // - (bit 8) dax shared memory
3399 //
3400 static void set_coredump_filter(bool largepages, bool dax_shared) {
3401 FILE *f;
3402 long cdm;
3403 bool filter_changed = false;
3404
3405 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3406 return;
3407 }
3408
3409 if (fscanf(f, "%lx", &cdm) != 1) {
3410 fclose(f);
3411 return;
3412 }
3413
3414 rewind(f);
3415
3416 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3417 cdm |= LARGEPAGES_BIT;
3418 filter_changed = true;
3419 }
3420 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3421 cdm |= DAX_SHARED_BIT;
3422 filter_changed = true;
3423 }
3424 if (filter_changed) {
3425 fprintf(f, "%#lx", cdm);
3426 }
3427
3428 fclose(f);
3429 }
3430
3431 // Large page support
3432
3433 static size_t _large_page_size = 0;
3434
3435 size_t os::Linux::find_large_page_size() {
3436 size_t large_page_size = 0;
3437
3438 // large_page_size on Linux is used to round up heap size. x86 uses either
3439 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3440 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3441 // page as large as 256M.
3442 //
3443 // Here we try to figure out page size by parsing /proc/meminfo and looking
3444 // for a line with the following format:
3445 // Hugepagesize: 2048 kB
3446 //
3447 // If we can't determine the value (e.g. /proc is not mounted, or the text
3448 // format has been changed), we'll use the largest page size supported by
3449 // the processor.
3450
3451 #ifndef ZERO
3452 large_page_size =
3453 AARCH64_ONLY(2 * M)
3454 AMD64_ONLY(2 * M)
3455 ARM32_ONLY(2 * M)
3456 IA32_ONLY(4 * M)
3457 IA64_ONLY(256 * M)
3458 PPC_ONLY(4 * M)
3459 S390_ONLY(1 * M)
3460 SPARC_ONLY(4 * M);
3461 #endif // ZERO
3462
3463 FILE *fp = fopen("/proc/meminfo", "r");
3464 if (fp) {
3465 while (!feof(fp)) {
3466 int x = 0;
3467 char buf[16];
3468 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3469 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3470 large_page_size = x * K;
3471 break;
3472 }
3473 } else {
3474 // skip to next line
3475 for (;;) {
3476 int ch = fgetc(fp);
3477 if (ch == EOF || ch == (int)'\n') break;
3478 }
3479 }
3480 }
3481 fclose(fp);
3482 }
3483
3484 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3485 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3486 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3487 proper_unit_for_byte_size(large_page_size));
3488 }
3489
3490 return large_page_size;
3491 }
3492
3493 size_t os::Linux::setup_large_page_size() {
3494 _large_page_size = Linux::find_large_page_size();
3495 const size_t default_page_size = (size_t)Linux::page_size();
3496 if (_large_page_size > default_page_size) {
3497 _page_sizes[0] = _large_page_size;
3498 _page_sizes[1] = default_page_size;
3499 _page_sizes[2] = 0;
3500 }
3501
3502 return _large_page_size;
3503 }
3504
3505 bool os::Linux::setup_large_page_type(size_t page_size) {
3506 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3507 FLAG_IS_DEFAULT(UseSHM) &&
3508 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3509
3510 // The type of large pages has not been specified by the user.
3511
3512 // Try UseHugeTLBFS and then UseSHM.
3513 UseHugeTLBFS = UseSHM = true;
3514
3515 // Don't try UseTransparentHugePages since there are known
3516 // performance issues with it turned on. This might change in the future.
3517 UseTransparentHugePages = false;
3518 }
3519
3520 if (UseTransparentHugePages) {
3521 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3522 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3523 UseHugeTLBFS = false;
3524 UseSHM = false;
3525 return true;
3526 }
3527 UseTransparentHugePages = false;
3528 }
3529
3530 if (UseHugeTLBFS) {
3531 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3532 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3533 UseSHM = false;
3534 return true;
3535 }
3536 UseHugeTLBFS = false;
3537 }
3538
3539 return UseSHM;
3540 }
3541
3542 void os::large_page_init() {
3543 if (!UseLargePages &&
3544 !UseTransparentHugePages &&
3545 !UseHugeTLBFS &&
3546 !UseSHM) {
3547 // Not using large pages.
3548 return;
3549 }
3550
3551 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3552 // The user explicitly turned off large pages.
3553 // Ignore the rest of the large pages flags.
3554 UseTransparentHugePages = false;
3555 UseHugeTLBFS = false;
3556 UseSHM = false;
3557 return;
3558 }
3559
3560 size_t large_page_size = Linux::setup_large_page_size();
3561 UseLargePages = Linux::setup_large_page_type(large_page_size);
3562
3563 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3564 }
3565
3566 #ifndef SHM_HUGETLB
3567 #define SHM_HUGETLB 04000
3568 #endif
3569
3570 #define shm_warning_format(format, ...) \
3571 do { \
3572 if (UseLargePages && \
3573 (!FLAG_IS_DEFAULT(UseLargePages) || \
3574 !FLAG_IS_DEFAULT(UseSHM) || \
3575 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3576 warning(format, __VA_ARGS__); \
3577 } \
3578 } while (0)
3579
3580 #define shm_warning(str) shm_warning_format("%s", str)
3581
3582 #define shm_warning_with_errno(str) \
3583 do { \
3584 int err = errno; \
3585 shm_warning_format(str " (error = %d)", err); \
3586 } while (0)
3587
3588 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3589 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3590
3591 if (!is_aligned(alignment, SHMLBA)) {
3592 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3593 return NULL;
3594 }
3595
3596 // To ensure that we get 'alignment' aligned memory from shmat,
3597 // we pre-reserve aligned virtual memory and then attach to that.
3598
3599 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3600 if (pre_reserved_addr == NULL) {
3601 // Couldn't pre-reserve aligned memory.
3602 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3603 return NULL;
3604 }
3605
3606 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3607 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3608
3609 if ((intptr_t)addr == -1) {
3610 int err = errno;
3611 shm_warning_with_errno("Failed to attach shared memory.");
3612
3613 assert(err != EACCES, "Unexpected error");
3614 assert(err != EIDRM, "Unexpected error");
3615 assert(err != EINVAL, "Unexpected error");
3616
3617 // Since we don't know if the kernel unmapped the pre-reserved memory area
3618 // we can't unmap it, since that would potentially unmap memory that was
3619 // mapped from other threads.
3620 return NULL;
3621 }
3622
3623 return addr;
3624 }
3625
3626 static char* shmat_at_address(int shmid, char* req_addr) {
3627 if (!is_aligned(req_addr, SHMLBA)) {
3628 assert(false, "Requested address needs to be SHMLBA aligned");
3629 return NULL;
3630 }
3631
3632 char* addr = (char*)shmat(shmid, req_addr, 0);
3633
3634 if ((intptr_t)addr == -1) {
3635 shm_warning_with_errno("Failed to attach shared memory.");
3636 return NULL;
3637 }
3638
3639 return addr;
3640 }
3641
3642 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3643 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3644 if (req_addr != NULL) {
3645 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3646 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3647 return shmat_at_address(shmid, req_addr);
3648 }
3649
3650 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3651 // return large page size aligned memory addresses when req_addr == NULL.
3652 // However, if the alignment is larger than the large page size, we have
3653 // to manually ensure that the memory returned is 'alignment' aligned.
3654 if (alignment > os::large_page_size()) {
3655 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3656 return shmat_with_alignment(shmid, bytes, alignment);
3657 } else {
3658 return shmat_at_address(shmid, NULL);
3659 }
3660 }
3661
3662 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3663 char* req_addr, bool exec) {
3664 // "exec" is passed in but not used. Creating the shared image for
3665 // the code cache doesn't have an SHM_X executable permission to check.
3666 assert(UseLargePages && UseSHM, "only for SHM large pages");
3667 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3668 assert(is_aligned(req_addr, alignment), "Unaligned address");
3669
3670 if (!is_aligned(bytes, os::large_page_size())) {
3671 return NULL; // Fallback to small pages.
3672 }
3673
3674 // Create a large shared memory region to attach to based on size.
3675 // Currently, size is the total size of the heap.
3676 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3677 if (shmid == -1) {
3678 // Possible reasons for shmget failure:
3679 // 1. shmmax is too small for Java heap.
3680 // > check shmmax value: cat /proc/sys/kernel/shmmax
3681 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3682 // 2. not enough large page memory.
3683 // > check available large pages: cat /proc/meminfo
3684 // > increase amount of large pages:
3685 // echo new_value > /proc/sys/vm/nr_hugepages
3686 // Note 1: different Linux may use different name for this property,
3687 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3688 // Note 2: it's possible there's enough physical memory available but
3689 // they are so fragmented after a long run that they can't
3690 // coalesce into large pages. Try to reserve large pages when
3691 // the system is still "fresh".
3692 shm_warning_with_errno("Failed to reserve shared memory.");
3693 return NULL;
3694 }
3695
3696 // Attach to the region.
3697 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3698
3699 // Remove shmid. If shmat() is successful, the actual shared memory segment
3700 // will be deleted when it's detached by shmdt() or when the process
3701 // terminates. If shmat() is not successful this will remove the shared
3702 // segment immediately.
3703 shmctl(shmid, IPC_RMID, NULL);
3704
3705 return addr;
3706 }
3707
3708 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3709 int error) {
3710 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3711
3712 bool warn_on_failure = UseLargePages &&
3713 (!FLAG_IS_DEFAULT(UseLargePages) ||
3714 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3715 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3716
3717 if (warn_on_failure) {
3718 char msg[128];
3719 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3720 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3721 warning("%s", msg);
3722 }
3723 }
3724
3725 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3726 char* req_addr,
3727 bool exec) {
3728 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3729 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3730 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3731
3732 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3733 char* addr = (char*)::mmap(req_addr, bytes, prot,
3734 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3735 -1, 0);
3736
3737 if (addr == MAP_FAILED) {
3738 warn_on_large_pages_failure(req_addr, bytes, errno);
3739 return NULL;
3740 }
3741
3742 assert(is_aligned(addr, os::large_page_size()), "Must be");
3743
3744 return addr;
3745 }
3746
3747 // Reserve memory using mmap(MAP_HUGETLB).
3748 // - bytes shall be a multiple of alignment.
3749 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3750 // - alignment sets the alignment at which memory shall be allocated.
3751 // It must be a multiple of allocation granularity.
3752 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3753 // req_addr or NULL.
3754 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3755 size_t alignment,
3756 char* req_addr,
3757 bool exec) {
3758 size_t large_page_size = os::large_page_size();
3759 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3760
3761 assert(is_aligned(req_addr, alignment), "Must be");
3762 assert(is_aligned(bytes, alignment), "Must be");
3763
3764 // First reserve - but not commit - the address range in small pages.
3765 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3766
3767 if (start == NULL) {
3768 return NULL;
3769 }
3770
3771 assert(is_aligned(start, alignment), "Must be");
3772
3773 char* end = start + bytes;
3774
3775 // Find the regions of the allocated chunk that can be promoted to large pages.
3776 char* lp_start = align_up(start, large_page_size);
3777 char* lp_end = align_down(end, large_page_size);
3778
3779 size_t lp_bytes = lp_end - lp_start;
3780
3781 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3782
3783 if (lp_bytes == 0) {
3784 // The mapped region doesn't even span the start and the end of a large page.
3785 // Fall back to allocate a non-special area.
3786 ::munmap(start, end - start);
3787 return NULL;
3788 }
3789
3790 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3791
3792 void* result;
3793
3794 // Commit small-paged leading area.
3795 if (start != lp_start) {
3796 result = ::mmap(start, lp_start - start, prot,
3797 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3798 -1, 0);
3799 if (result == MAP_FAILED) {
3800 ::munmap(lp_start, end - lp_start);
3801 return NULL;
3802 }
3803 }
3804
3805 // Commit large-paged area.
3806 result = ::mmap(lp_start, lp_bytes, prot,
3807 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3808 -1, 0);
3809 if (result == MAP_FAILED) {
3810 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3811 // If the mmap above fails, the large pages region will be unmapped and we
3812 // have regions before and after with small pages. Release these regions.
3813 //
3814 // | mapped | unmapped | mapped |
3815 // ^ ^ ^ ^
3816 // start lp_start lp_end end
3817 //
3818 ::munmap(start, lp_start - start);
3819 ::munmap(lp_end, end - lp_end);
3820 return NULL;
3821 }
3822
3823 // Commit small-paged trailing area.
3824 if (lp_end != end) {
3825 result = ::mmap(lp_end, end - lp_end, prot,
3826 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3827 -1, 0);
3828 if (result == MAP_FAILED) {
3829 ::munmap(start, lp_end - start);
3830 return NULL;
3831 }
3832 }
3833
3834 return start;
3835 }
3836
3837 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3838 size_t alignment,
3839 char* req_addr,
3840 bool exec) {
3841 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3842 assert(is_aligned(req_addr, alignment), "Must be");
3843 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3844 assert(is_power_of_2(os::large_page_size()), "Must be");
3845 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3846
3847 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3848 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3849 } else {
3850 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3851 }
3852 }
3853
3854 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3855 char* req_addr, bool exec) {
3856 assert(UseLargePages, "only for large pages");
3857
3858 char* addr;
3859 if (UseSHM) {
3860 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3861 } else {
3862 assert(UseHugeTLBFS, "must be");
3863 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3864 }
3865
3866 if (addr != NULL) {
3867 if (UseNUMAInterleaving) {
3868 numa_make_global(addr, bytes);
3869 }
3870
3871 // The memory is committed
3872 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3873 }
3874
3875 return addr;
3876 }
3877
3878 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3879 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3880 return shmdt(base) == 0;
3881 }
3882
3883 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3884 return pd_release_memory(base, bytes);
3885 }
3886
3887 bool os::release_memory_special(char* base, size_t bytes) {
3888 bool res;
3889 if (MemTracker::tracking_level() > NMT_minimal) {
3890 Tracker tkr(Tracker::release);
3891 res = os::Linux::release_memory_special_impl(base, bytes);
3892 if (res) {
3893 tkr.record((address)base, bytes);
3894 }
3895
3896 } else {
3897 res = os::Linux::release_memory_special_impl(base, bytes);
3898 }
3899 return res;
3900 }
3901
3902 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3903 assert(UseLargePages, "only for large pages");
3904 bool res;
3905
3906 if (UseSHM) {
3907 res = os::Linux::release_memory_special_shm(base, bytes);
3908 } else {
3909 assert(UseHugeTLBFS, "must be");
3910 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3911 }
3912 return res;
3913 }
3914
3915 size_t os::large_page_size() {
3916 return _large_page_size;
3917 }
3918
3919 // With SysV SHM the entire memory region must be allocated as shared
3920 // memory.
3921 // HugeTLBFS allows application to commit large page memory on demand.
3922 // However, when committing memory with HugeTLBFS fails, the region
3923 // that was supposed to be committed will lose the old reservation
3924 // and allow other threads to steal that memory region. Because of this
3925 // behavior we can't commit HugeTLBFS memory.
3926 bool os::can_commit_large_page_memory() {
3927 return UseTransparentHugePages;
3928 }
3929
3930 bool os::can_execute_large_page_memory() {
3931 return UseTransparentHugePages || UseHugeTLBFS;
3932 }
3933
3934 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3935 assert(file_desc >= 0, "file_desc is not valid");
3936 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3937 if (result != NULL) {
3938 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3939 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3940 }
3941 }
3942 return result;
3943 }
3944
3945 // Reserve memory at an arbitrary address, only if that area is
3946 // available (and not reserved for something else).
3947
3948 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3949 const int max_tries = 10;
3950 char* base[max_tries];
3951 size_t size[max_tries];
3952 const size_t gap = 0x000000;
3953
3954 // Assert only that the size is a multiple of the page size, since
3955 // that's all that mmap requires, and since that's all we really know
3956 // about at this low abstraction level. If we need higher alignment,
3957 // we can either pass an alignment to this method or verify alignment
3958 // in one of the methods further up the call chain. See bug 5044738.
3959 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3960
3961 // Repeatedly allocate blocks until the block is allocated at the
3962 // right spot.
3963
3964 // Linux mmap allows caller to pass an address as hint; give it a try first,
3965 // if kernel honors the hint then we can return immediately.
3966 char * addr = anon_mmap(requested_addr, bytes, false);
3967 if (addr == requested_addr) {
3968 return requested_addr;
3969 }
3970
3971 if (addr != NULL) {
3972 // mmap() is successful but it fails to reserve at the requested address
3973 anon_munmap(addr, bytes);
3974 }
3975
3976 int i;
3977 for (i = 0; i < max_tries; ++i) {
3978 base[i] = reserve_memory(bytes);
3979
3980 if (base[i] != NULL) {
3981 // Is this the block we wanted?
3982 if (base[i] == requested_addr) {
3983 size[i] = bytes;
3984 break;
3985 }
3986
3987 // Does this overlap the block we wanted? Give back the overlapped
3988 // parts and try again.
3989
3990 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3991 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3992 unmap_memory(base[i], top_overlap);
3993 base[i] += top_overlap;
3994 size[i] = bytes - top_overlap;
3995 } else {
3996 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3997 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3998 unmap_memory(requested_addr, bottom_overlap);
3999 size[i] = bytes - bottom_overlap;
4000 } else {
4001 size[i] = bytes;
4002 }
4003 }
4004 }
4005 }
4006
4007 // Give back the unused reserved pieces.
4008
4009 for (int j = 0; j < i; ++j) {
4010 if (base[j] != NULL) {
4011 unmap_memory(base[j], size[j]);
4012 }
4013 }
4014
4015 if (i < max_tries) {
4016 return requested_addr;
4017 } else {
4018 return NULL;
4019 }
4020 }
4021
4022 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4023 return ::read(fd, buf, nBytes);
4024 }
4025
4026 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4027 return ::pread(fd, buf, nBytes, offset);
4028 }
4029
4030 // Short sleep, direct OS call.
4031 //
4032 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4033 // sched_yield(2) will actually give up the CPU:
4034 //
4035 // * Alone on this pariticular CPU, keeps running.
4036 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4037 // (pre 2.6.39).
4038 //
4039 // So calling this with 0 is an alternative.
4040 //
4041 void os::naked_short_sleep(jlong ms) {
4042 struct timespec req;
4043
4044 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4045 req.tv_sec = 0;
4046 if (ms > 0) {
4047 req.tv_nsec = (ms % 1000) * 1000000;
4048 } else {
4049 req.tv_nsec = 1;
4050 }
4051
4052 nanosleep(&req, NULL);
4053
4054 return;
4055 }
4056
4057 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4058 void os::infinite_sleep() {
4059 while (true) { // sleep forever ...
4060 ::sleep(100); // ... 100 seconds at a time
4061 }
4062 }
4063
4064 // Used to convert frequent JVM_Yield() to nops
4065 bool os::dont_yield() {
4066 return DontYieldALot;
4067 }
4068
4069 void os::naked_yield() {
4070 sched_yield();
4071 }
4072
4073 ////////////////////////////////////////////////////////////////////////////////
4074 // thread priority support
4075
4076 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4077 // only supports dynamic priority, static priority must be zero. For real-time
4078 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4079 // However, for large multi-threaded applications, SCHED_RR is not only slower
4080 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4081 // of 5 runs - Sep 2005).
4082 //
4083 // The following code actually changes the niceness of kernel-thread/LWP. It
4084 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4085 // not the entire user process, and user level threads are 1:1 mapped to kernel
4086 // threads. It has always been the case, but could change in the future. For
4087 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4088 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4089
4090 int os::java_to_os_priority[CriticalPriority + 1] = {
4091 19, // 0 Entry should never be used
4092
4093 4, // 1 MinPriority
4094 3, // 2
4095 2, // 3
4096
4097 1, // 4
4098 0, // 5 NormPriority
4099 -1, // 6
4100
4101 -2, // 7
4102 -3, // 8
4103 -4, // 9 NearMaxPriority
4104
4105 -5, // 10 MaxPriority
4106
4107 -5 // 11 CriticalPriority
4108 };
4109
4110 static int prio_init() {
4111 if (ThreadPriorityPolicy == 1) {
4112 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4113 // if effective uid is not root. Perhaps, a more elegant way of doing
4114 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4115 if (geteuid() != 0) {
4116 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4117 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4118 }
4119 ThreadPriorityPolicy = 0;
4120 }
4121 }
4122 if (UseCriticalJavaThreadPriority) {
4123 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4124 }
4125 return 0;
4126 }
4127
4128 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4129 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4130
4131 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4132 return (ret == 0) ? OS_OK : OS_ERR;
4133 }
4134
4135 OSReturn os::get_native_priority(const Thread* const thread,
4136 int *priority_ptr) {
4137 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4138 *priority_ptr = java_to_os_priority[NormPriority];
4139 return OS_OK;
4140 }
4141
4142 errno = 0;
4143 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4144 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4145 }
4146
4147 ////////////////////////////////////////////////////////////////////////////////
4148 // suspend/resume support
4149
4150 // The low-level signal-based suspend/resume support is a remnant from the
4151 // old VM-suspension that used to be for java-suspension, safepoints etc,
4152 // within hotspot. Currently used by JFR's OSThreadSampler
4153 //
4154 // The remaining code is greatly simplified from the more general suspension
4155 // code that used to be used.
4156 //
4157 // The protocol is quite simple:
4158 // - suspend:
4159 // - sends a signal to the target thread
4160 // - polls the suspend state of the osthread using a yield loop
4161 // - target thread signal handler (SR_handler) sets suspend state
4162 // and blocks in sigsuspend until continued
4163 // - resume:
4164 // - sets target osthread state to continue
4165 // - sends signal to end the sigsuspend loop in the SR_handler
4166 //
4167 // Note that the SR_lock plays no role in this suspend/resume protocol,
4168 // but is checked for NULL in SR_handler as a thread termination indicator.
4169 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4170 //
4171 // Note that resume_clear_context() and suspend_save_context() are needed
4172 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4173 // which in part is used by:
4174 // - Forte Analyzer: AsyncGetCallTrace()
4175 // - StackBanging: get_frame_at_stack_banging_point()
4176
4177 static void resume_clear_context(OSThread *osthread) {
4178 osthread->set_ucontext(NULL);
4179 osthread->set_siginfo(NULL);
4180 }
4181
4182 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4183 ucontext_t* context) {
4184 osthread->set_ucontext(context);
4185 osthread->set_siginfo(siginfo);
4186 }
4187
4188 // Handler function invoked when a thread's execution is suspended or
4189 // resumed. We have to be careful that only async-safe functions are
4190 // called here (Note: most pthread functions are not async safe and
4191 // should be avoided.)
4192 //
4193 // Note: sigwait() is a more natural fit than sigsuspend() from an
4194 // interface point of view, but sigwait() prevents the signal hander
4195 // from being run. libpthread would get very confused by not having
4196 // its signal handlers run and prevents sigwait()'s use with the
4197 // mutex granting granting signal.
4198 //
4199 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4200 //
4201 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4202 // Save and restore errno to avoid confusing native code with EINTR
4203 // after sigsuspend.
4204 int old_errno = errno;
4205
4206 Thread* thread = Thread::current_or_null_safe();
4207 assert(thread != NULL, "Missing current thread in SR_handler");
4208
4209 // On some systems we have seen signal delivery get "stuck" until the signal
4210 // mask is changed as part of thread termination. Check that the current thread
4211 // has not already terminated (via SR_lock()) - else the following assertion
4212 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4213 // destructor has completed.
4214
4215 if (thread->SR_lock() == NULL) {
4216 return;
4217 }
4218
4219 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4220
4221 OSThread* osthread = thread->osthread();
4222
4223 os::SuspendResume::State current = osthread->sr.state();
4224 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4225 suspend_save_context(osthread, siginfo, context);
4226
4227 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4228 os::SuspendResume::State state = osthread->sr.suspended();
4229 if (state == os::SuspendResume::SR_SUSPENDED) {
4230 sigset_t suspend_set; // signals for sigsuspend()
4231 sigemptyset(&suspend_set);
4232 // get current set of blocked signals and unblock resume signal
4233 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4234 sigdelset(&suspend_set, SR_signum);
4235
4236 sr_semaphore.signal();
4237 // wait here until we are resumed
4238 while (1) {
4239 sigsuspend(&suspend_set);
4240
4241 os::SuspendResume::State result = osthread->sr.running();
4242 if (result == os::SuspendResume::SR_RUNNING) {
4243 sr_semaphore.signal();
4244 break;
4245 }
4246 }
4247
4248 } else if (state == os::SuspendResume::SR_RUNNING) {
4249 // request was cancelled, continue
4250 } else {
4251 ShouldNotReachHere();
4252 }
4253
4254 resume_clear_context(osthread);
4255 } else if (current == os::SuspendResume::SR_RUNNING) {
4256 // request was cancelled, continue
4257 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4258 // ignore
4259 } else {
4260 // ignore
4261 }
4262
4263 errno = old_errno;
4264 }
4265
4266 static int SR_initialize() {
4267 struct sigaction act;
4268 char *s;
4269
4270 // Get signal number to use for suspend/resume
4271 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4272 int sig = ::strtol(s, 0, 10);
4273 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4274 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4275 SR_signum = sig;
4276 } else {
4277 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4278 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4279 }
4280 }
4281
4282 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4283 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4284
4285 sigemptyset(&SR_sigset);
4286 sigaddset(&SR_sigset, SR_signum);
4287
4288 // Set up signal handler for suspend/resume
4289 act.sa_flags = SA_RESTART|SA_SIGINFO;
4290 act.sa_handler = (void (*)(int)) SR_handler;
4291
4292 // SR_signum is blocked by default.
4293 // 4528190 - We also need to block pthread restart signal (32 on all
4294 // supported Linux platforms). Note that LinuxThreads need to block
4295 // this signal for all threads to work properly. So we don't have
4296 // to use hard-coded signal number when setting up the mask.
4297 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4298
4299 if (sigaction(SR_signum, &act, 0) == -1) {
4300 return -1;
4301 }
4302
4303 // Save signal flag
4304 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4305 return 0;
4306 }
4307
4308 static int sr_notify(OSThread* osthread) {
4309 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4310 assert_status(status == 0, status, "pthread_kill");
4311 return status;
4312 }
4313
4314 // "Randomly" selected value for how long we want to spin
4315 // before bailing out on suspending a thread, also how often
4316 // we send a signal to a thread we want to resume
4317 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4318 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4319
4320 // returns true on success and false on error - really an error is fatal
4321 // but this seems the normal response to library errors
4322 static bool do_suspend(OSThread* osthread) {
4323 assert(osthread->sr.is_running(), "thread should be running");
4324 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4325
4326 // mark as suspended and send signal
4327 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4328 // failed to switch, state wasn't running?
4329 ShouldNotReachHere();
4330 return false;
4331 }
4332
4333 if (sr_notify(osthread) != 0) {
4334 ShouldNotReachHere();
4335 }
4336
4337 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4338 while (true) {
4339 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4340 break;
4341 } else {
4342 // timeout
4343 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4344 if (cancelled == os::SuspendResume::SR_RUNNING) {
4345 return false;
4346 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4347 // make sure that we consume the signal on the semaphore as well
4348 sr_semaphore.wait();
4349 break;
4350 } else {
4351 ShouldNotReachHere();
4352 return false;
4353 }
4354 }
4355 }
4356
4357 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4358 return true;
4359 }
4360
4361 static void do_resume(OSThread* osthread) {
4362 assert(osthread->sr.is_suspended(), "thread should be suspended");
4363 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4364
4365 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4366 // failed to switch to WAKEUP_REQUEST
4367 ShouldNotReachHere();
4368 return;
4369 }
4370
4371 while (true) {
4372 if (sr_notify(osthread) == 0) {
4373 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4374 if (osthread->sr.is_running()) {
4375 return;
4376 }
4377 }
4378 } else {
4379 ShouldNotReachHere();
4380 }
4381 }
4382
4383 guarantee(osthread->sr.is_running(), "Must be running!");
4384 }
4385
4386 ///////////////////////////////////////////////////////////////////////////////////
4387 // signal handling (except suspend/resume)
4388
4389 // This routine may be used by user applications as a "hook" to catch signals.
4390 // The user-defined signal handler must pass unrecognized signals to this
4391 // routine, and if it returns true (non-zero), then the signal handler must
4392 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4393 // routine will never retun false (zero), but instead will execute a VM panic
4394 // routine kill the process.
4395 //
4396 // If this routine returns false, it is OK to call it again. This allows
4397 // the user-defined signal handler to perform checks either before or after
4398 // the VM performs its own checks. Naturally, the user code would be making
4399 // a serious error if it tried to handle an exception (such as a null check
4400 // or breakpoint) that the VM was generating for its own correct operation.
4401 //
4402 // This routine may recognize any of the following kinds of signals:
4403 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4404 // It should be consulted by handlers for any of those signals.
4405 //
4406 // The caller of this routine must pass in the three arguments supplied
4407 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4408 // field of the structure passed to sigaction(). This routine assumes that
4409 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4410 //
4411 // Note that the VM will print warnings if it detects conflicting signal
4412 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4413 //
4414 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4415 siginfo_t* siginfo,
4416 void* ucontext,
4417 int abort_if_unrecognized);
4418
4419 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4420 assert(info != NULL && uc != NULL, "it must be old kernel");
4421 int orig_errno = errno; // Preserve errno value over signal handler.
4422 JVM_handle_linux_signal(sig, info, uc, true);
4423 errno = orig_errno;
4424 }
4425
4426
4427 // This boolean allows users to forward their own non-matching signals
4428 // to JVM_handle_linux_signal, harmlessly.
4429 bool os::Linux::signal_handlers_are_installed = false;
4430
4431 // For signal-chaining
4432 struct sigaction sigact[NSIG];
4433 uint64_t sigs = 0;
4434 #if (64 < NSIG-1)
4435 #error "Not all signals can be encoded in sigs. Adapt its type!"
4436 #endif
4437 bool os::Linux::libjsig_is_loaded = false;
4438 typedef struct sigaction *(*get_signal_t)(int);
4439 get_signal_t os::Linux::get_signal_action = NULL;
4440
4441 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4442 struct sigaction *actp = NULL;
4443
4444 if (libjsig_is_loaded) {
4445 // Retrieve the old signal handler from libjsig
4446 actp = (*get_signal_action)(sig);
4447 }
4448 if (actp == NULL) {
4449 // Retrieve the preinstalled signal handler from jvm
4450 actp = get_preinstalled_handler(sig);
4451 }
4452
4453 return actp;
4454 }
4455
4456 static bool call_chained_handler(struct sigaction *actp, int sig,
4457 siginfo_t *siginfo, void *context) {
4458 // Call the old signal handler
4459 if (actp->sa_handler == SIG_DFL) {
4460 // It's more reasonable to let jvm treat it as an unexpected exception
4461 // instead of taking the default action.
4462 return false;
4463 } else if (actp->sa_handler != SIG_IGN) {
4464 if ((actp->sa_flags & SA_NODEFER) == 0) {
4465 // automaticlly block the signal
4466 sigaddset(&(actp->sa_mask), sig);
4467 }
4468
4469 sa_handler_t hand = NULL;
4470 sa_sigaction_t sa = NULL;
4471 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4472 // retrieve the chained handler
4473 if (siginfo_flag_set) {
4474 sa = actp->sa_sigaction;
4475 } else {
4476 hand = actp->sa_handler;
4477 }
4478
4479 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4480 actp->sa_handler = SIG_DFL;
4481 }
4482
4483 // try to honor the signal mask
4484 sigset_t oset;
4485 sigemptyset(&oset);
4486 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4487
4488 // call into the chained handler
4489 if (siginfo_flag_set) {
4490 (*sa)(sig, siginfo, context);
4491 } else {
4492 (*hand)(sig);
4493 }
4494
4495 // restore the signal mask
4496 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4497 }
4498 // Tell jvm's signal handler the signal is taken care of.
4499 return true;
4500 }
4501
4502 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4503 bool chained = false;
4504 // signal-chaining
4505 if (UseSignalChaining) {
4506 struct sigaction *actp = get_chained_signal_action(sig);
4507 if (actp != NULL) {
4508 chained = call_chained_handler(actp, sig, siginfo, context);
4509 }
4510 }
4511 return chained;
4512 }
4513
4514 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4515 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4516 return &sigact[sig];
4517 }
4518 return NULL;
4519 }
4520
4521 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4522 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4523 sigact[sig] = oldAct;
4524 sigs |= (uint64_t)1 << (sig-1);
4525 }
4526
4527 // for diagnostic
4528 int sigflags[NSIG];
4529
4530 int os::Linux::get_our_sigflags(int sig) {
4531 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4532 return sigflags[sig];
4533 }
4534
4535 void os::Linux::set_our_sigflags(int sig, int flags) {
4536 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4537 if (sig > 0 && sig < NSIG) {
4538 sigflags[sig] = flags;
4539 }
4540 }
4541
4542 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4543 // Check for overwrite.
4544 struct sigaction oldAct;
4545 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4546
4547 void* oldhand = oldAct.sa_sigaction
4548 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4549 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4550 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4551 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4552 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4553 if (AllowUserSignalHandlers || !set_installed) {
4554 // Do not overwrite; user takes responsibility to forward to us.
4555 return;
4556 } else if (UseSignalChaining) {
4557 // save the old handler in jvm
4558 save_preinstalled_handler(sig, oldAct);
4559 // libjsig also interposes the sigaction() call below and saves the
4560 // old sigaction on it own.
4561 } else {
4562 fatal("Encountered unexpected pre-existing sigaction handler "
4563 "%#lx for signal %d.", (long)oldhand, sig);
4564 }
4565 }
4566
4567 struct sigaction sigAct;
4568 sigfillset(&(sigAct.sa_mask));
4569 sigAct.sa_handler = SIG_DFL;
4570 if (!set_installed) {
4571 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4572 } else {
4573 sigAct.sa_sigaction = signalHandler;
4574 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4575 }
4576 // Save flags, which are set by ours
4577 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4578 sigflags[sig] = sigAct.sa_flags;
4579
4580 int ret = sigaction(sig, &sigAct, &oldAct);
4581 assert(ret == 0, "check");
4582
4583 void* oldhand2 = oldAct.sa_sigaction
4584 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4585 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4586 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4587 }
4588
4589 // install signal handlers for signals that HotSpot needs to
4590 // handle in order to support Java-level exception handling.
4591
4592 void os::Linux::install_signal_handlers() {
4593 if (!signal_handlers_are_installed) {
4594 signal_handlers_are_installed = true;
4595
4596 // signal-chaining
4597 typedef void (*signal_setting_t)();
4598 signal_setting_t begin_signal_setting = NULL;
4599 signal_setting_t end_signal_setting = NULL;
4600 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4601 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4602 if (begin_signal_setting != NULL) {
4603 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4604 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4605 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4606 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4607 libjsig_is_loaded = true;
4608 assert(UseSignalChaining, "should enable signal-chaining");
4609 }
4610 if (libjsig_is_loaded) {
4611 // Tell libjsig jvm is setting signal handlers
4612 (*begin_signal_setting)();
4613 }
4614
4615 set_signal_handler(SIGSEGV, true);
4616 set_signal_handler(SIGPIPE, true);
4617 set_signal_handler(SIGBUS, true);
4618 set_signal_handler(SIGILL, true);
4619 set_signal_handler(SIGFPE, true);
4620 #if defined(PPC64)
4621 set_signal_handler(SIGTRAP, true);
4622 #endif
4623 set_signal_handler(SIGXFSZ, true);
4624
4625 if (libjsig_is_loaded) {
4626 // Tell libjsig jvm finishes setting signal handlers
4627 (*end_signal_setting)();
4628 }
4629
4630 // We don't activate signal checker if libjsig is in place, we trust ourselves
4631 // and if UserSignalHandler is installed all bets are off.
4632 // Log that signal checking is off only if -verbose:jni is specified.
4633 if (CheckJNICalls) {
4634 if (libjsig_is_loaded) {
4635 if (PrintJNIResolving) {
4636 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4637 }
4638 check_signals = false;
4639 }
4640 if (AllowUserSignalHandlers) {
4641 if (PrintJNIResolving) {
4642 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4643 }
4644 check_signals = false;
4645 }
4646 }
4647 }
4648 }
4649
4650 // This is the fastest way to get thread cpu time on Linux.
4651 // Returns cpu time (user+sys) for any thread, not only for current.
4652 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4653 // It might work on 2.6.10+ with a special kernel/glibc patch.
4654 // For reference, please, see IEEE Std 1003.1-2004:
4655 // http://www.unix.org/single_unix_specification
4656
4657 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4658 struct timespec tp;
4659 int rc = os::Posix::clock_gettime(clockid, &tp);
4660 assert(rc == 0, "clock_gettime is expected to return 0 code");
4661
4662 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4663 }
4664
4665 void os::Linux::initialize_os_info() {
4666 assert(_os_version == 0, "OS info already initialized");
4667
4668 struct utsname _uname;
4669
4670 uint32_t major;
4671 uint32_t minor;
4672 uint32_t fix;
4673
4674 int rc;
4675
4676 // Kernel version is unknown if
4677 // verification below fails.
4678 _os_version = 0x01000000;
4679
4680 rc = uname(&_uname);
4681 if (rc != -1) {
4682
4683 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4684 if (rc == 3) {
4685
4686 if (major < 256 && minor < 256 && fix < 256) {
4687 // Kernel version format is as expected,
4688 // set it overriding unknown state.
4689 _os_version = (major << 16) |
4690 (minor << 8 ) |
4691 (fix << 0 ) ;
4692 }
4693 }
4694 }
4695 }
4696
4697 uint32_t os::Linux::os_version() {
4698 assert(_os_version != 0, "not initialized");
4699 return _os_version & 0x00FFFFFF;
4700 }
4701
4702 bool os::Linux::os_version_is_known() {
4703 assert(_os_version != 0, "not initialized");
4704 return _os_version & 0x01000000 ? false : true;
4705 }
4706
4707 /////
4708 // glibc on Linux platform uses non-documented flag
4709 // to indicate, that some special sort of signal
4710 // trampoline is used.
4711 // We will never set this flag, and we should
4712 // ignore this flag in our diagnostic
4713 #ifdef SIGNIFICANT_SIGNAL_MASK
4714 #undef SIGNIFICANT_SIGNAL_MASK
4715 #endif
4716 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4717
4718 static const char* get_signal_handler_name(address handler,
4719 char* buf, int buflen) {
4720 int offset = 0;
4721 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4722 if (found) {
4723 // skip directory names
4724 const char *p1, *p2;
4725 p1 = buf;
4726 size_t len = strlen(os::file_separator());
4727 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4728 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4729 } else {
4730 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4731 }
4732 return buf;
4733 }
4734
4735 static void print_signal_handler(outputStream* st, int sig,
4736 char* buf, size_t buflen) {
4737 struct sigaction sa;
4738
4739 sigaction(sig, NULL, &sa);
4740
4741 // See comment for SIGNIFICANT_SIGNAL_MASK define
4742 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4743
4744 st->print("%s: ", os::exception_name(sig, buf, buflen));
4745
4746 address handler = (sa.sa_flags & SA_SIGINFO)
4747 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4748 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4749
4750 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4751 st->print("SIG_DFL");
4752 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4753 st->print("SIG_IGN");
4754 } else {
4755 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4756 }
4757
4758 st->print(", sa_mask[0]=");
4759 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4760
4761 address rh = VMError::get_resetted_sighandler(sig);
4762 // May be, handler was resetted by VMError?
4763 if (rh != NULL) {
4764 handler = rh;
4765 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4766 }
4767
4768 st->print(", sa_flags=");
4769 os::Posix::print_sa_flags(st, sa.sa_flags);
4770
4771 // Check: is it our handler?
4772 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4773 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4774 // It is our signal handler
4775 // check for flags, reset system-used one!
4776 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4777 st->print(
4778 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4779 os::Linux::get_our_sigflags(sig));
4780 }
4781 }
4782 st->cr();
4783 }
4784
4785
4786 #define DO_SIGNAL_CHECK(sig) \
4787 do { \
4788 if (!sigismember(&check_signal_done, sig)) { \
4789 os::Linux::check_signal_handler(sig); \
4790 } \
4791 } while (0)
4792
4793 // This method is a periodic task to check for misbehaving JNI applications
4794 // under CheckJNI, we can add any periodic checks here
4795
4796 void os::run_periodic_checks() {
4797 if (check_signals == false) return;
4798
4799 // SEGV and BUS if overridden could potentially prevent
4800 // generation of hs*.log in the event of a crash, debugging
4801 // such a case can be very challenging, so we absolutely
4802 // check the following for a good measure:
4803 DO_SIGNAL_CHECK(SIGSEGV);
4804 DO_SIGNAL_CHECK(SIGILL);
4805 DO_SIGNAL_CHECK(SIGFPE);
4806 DO_SIGNAL_CHECK(SIGBUS);
4807 DO_SIGNAL_CHECK(SIGPIPE);
4808 DO_SIGNAL_CHECK(SIGXFSZ);
4809 #if defined(PPC64)
4810 DO_SIGNAL_CHECK(SIGTRAP);
4811 #endif
4812
4813 // ReduceSignalUsage allows the user to override these handlers
4814 // see comments at the very top and jvm_md.h
4815 if (!ReduceSignalUsage) {
4816 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4817 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4818 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4819 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4820 }
4821
4822 DO_SIGNAL_CHECK(SR_signum);
4823 }
4824
4825 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4826
4827 static os_sigaction_t os_sigaction = NULL;
4828
4829 void os::Linux::check_signal_handler(int sig) {
4830 char buf[O_BUFLEN];
4831 address jvmHandler = NULL;
4832
4833
4834 struct sigaction act;
4835 if (os_sigaction == NULL) {
4836 // only trust the default sigaction, in case it has been interposed
4837 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4838 if (os_sigaction == NULL) return;
4839 }
4840
4841 os_sigaction(sig, (struct sigaction*)NULL, &act);
4842
4843
4844 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4845
4846 address thisHandler = (act.sa_flags & SA_SIGINFO)
4847 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4848 : CAST_FROM_FN_PTR(address, act.sa_handler);
4849
4850
4851 switch (sig) {
4852 case SIGSEGV:
4853 case SIGBUS:
4854 case SIGFPE:
4855 case SIGPIPE:
4856 case SIGILL:
4857 case SIGXFSZ:
4858 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4859 break;
4860
4861 case SHUTDOWN1_SIGNAL:
4862 case SHUTDOWN2_SIGNAL:
4863 case SHUTDOWN3_SIGNAL:
4864 case BREAK_SIGNAL:
4865 jvmHandler = (address)user_handler();
4866 break;
4867
4868 default:
4869 if (sig == SR_signum) {
4870 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4871 } else {
4872 return;
4873 }
4874 break;
4875 }
4876
4877 if (thisHandler != jvmHandler) {
4878 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4879 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4880 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4881 // No need to check this sig any longer
4882 sigaddset(&check_signal_done, sig);
4883 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4884 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4885 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4886 exception_name(sig, buf, O_BUFLEN));
4887 }
4888 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4889 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4890 tty->print("expected:");
4891 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4892 tty->cr();
4893 tty->print(" found:");
4894 os::Posix::print_sa_flags(tty, act.sa_flags);
4895 tty->cr();
4896 // No need to check this sig any longer
4897 sigaddset(&check_signal_done, sig);
4898 }
4899
4900 // Dump all the signal
4901 if (sigismember(&check_signal_done, sig)) {
4902 print_signal_handlers(tty, buf, O_BUFLEN);
4903 }
4904 }
4905
4906 extern void report_error(char* file_name, int line_no, char* title,
4907 char* format, ...);
4908
4909 // this is called _before_ most of the global arguments have been parsed
4910 void os::init(void) {
4911 char dummy; // used to get a guess on initial stack address
4912
4913 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4914
4915 init_random(1234567);
4916
4917 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4918 if (Linux::page_size() == -1) {
4919 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4920 os::strerror(errno));
4921 }
4922 init_page_sizes((size_t) Linux::page_size());
4923
4924 Linux::initialize_system_info();
4925
4926 Linux::initialize_os_info();
4927
4928 // _main_thread points to the thread that created/loaded the JVM.
4929 Linux::_main_thread = pthread_self();
4930
4931 // retrieve entry point for pthread_setname_np
4932 Linux::_pthread_setname_np =
4933 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4934
4935 os::Posix::init();
4936
4937 initial_time_count = javaTimeNanos();
4938
4939 // Always warn if no monotonic clock available
4940 if (!os::Posix::supports_monotonic_clock()) {
4941 warning("No monotonic clock was available - timed services may " \
4942 "be adversely affected if the time-of-day clock changes");
4943 }
4944 }
4945
4946 // To install functions for atexit system call
4947 extern "C" {
4948 static void perfMemory_exit_helper() {
4949 perfMemory_exit();
4950 }
4951 }
4952
4953 void os::pd_init_container_support() {
4954 OSContainer::init();
4955 }
4956
4957 // this is called _after_ the global arguments have been parsed
4958 jint os::init_2(void) {
4959
4960 os::Posix::init_2();
4961
4962 Linux::fast_thread_clock_init();
4963
4964 // initialize suspend/resume support - must do this before signal_sets_init()
4965 if (SR_initialize() != 0) {
4966 perror("SR_initialize failed");
4967 return JNI_ERR;
4968 }
4969
4970 Linux::signal_sets_init();
4971 Linux::install_signal_handlers();
4972 // Initialize data for jdk.internal.misc.Signal
4973 if (!ReduceSignalUsage) {
4974 jdk_misc_signal_init();
4975 }
4976
4977 // Check and sets minimum stack sizes against command line options
4978 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4979 return JNI_ERR;
4980 }
4981
4982 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
4983 if (!suppress_primordial_thread_resolution) {
4984 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4985 }
4986
4987 #if defined(IA32)
4988 workaround_expand_exec_shield_cs_limit();
4989 #endif
4990
4991 Linux::libpthread_init();
4992 Linux::sched_getcpu_init();
4993 log_info(os)("HotSpot is running with %s, %s",
4994 Linux::glibc_version(), Linux::libpthread_version());
4995
4996 if (UseNUMA) {
4997 if (!Linux::libnuma_init()) {
4998 UseNUMA = false;
4999 } else {
5000 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5001 // If there's only one node (they start from 0) or if the process
5002 // is bound explicitly to a single node using membind, disable NUMA.
5003 UseNUMA = false;
5004 }
5005 }
5006
5007 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5008 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5009 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5010 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5011 // and disable adaptive resizing.
5012 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5013 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5014 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5015 UseAdaptiveSizePolicy = false;
5016 UseAdaptiveNUMAChunkSizing = false;
5017 }
5018 }
5019
5020 if (!UseNUMA && ForceNUMA) {
5021 UseNUMA = true;
5022 }
5023 }
5024
5025 if (MaxFDLimit) {
5026 // set the number of file descriptors to max. print out error
5027 // if getrlimit/setrlimit fails but continue regardless.
5028 struct rlimit nbr_files;
5029 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5030 if (status != 0) {
5031 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5032 } else {
5033 nbr_files.rlim_cur = nbr_files.rlim_max;
5034 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5035 if (status != 0) {
5036 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5037 }
5038 }
5039 }
5040
5041 // Initialize lock used to serialize thread creation (see os::create_thread)
5042 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5043
5044 // at-exit methods are called in the reverse order of their registration.
5045 // atexit functions are called on return from main or as a result of a
5046 // call to exit(3C). There can be only 32 of these functions registered
5047 // and atexit() does not set errno.
5048
5049 if (PerfAllowAtExitRegistration) {
5050 // only register atexit functions if PerfAllowAtExitRegistration is set.
5051 // atexit functions can be delayed until process exit time, which
5052 // can be problematic for embedded VM situations. Embedded VMs should
5053 // call DestroyJavaVM() to assure that VM resources are released.
5054
5055 // note: perfMemory_exit_helper atexit function may be removed in
5056 // the future if the appropriate cleanup code can be added to the
5057 // VM_Exit VMOperation's doit method.
5058 if (atexit(perfMemory_exit_helper) != 0) {
5059 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5060 }
5061 }
5062
5063 // initialize thread priority policy
5064 prio_init();
5065
5066 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5067 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5068 }
5069 return JNI_OK;
5070 }
5071
5072 // Mark the polling page as unreadable
5073 void os::make_polling_page_unreadable(void) {
5074 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5075 fatal("Could not disable polling page");
5076 }
5077 }
5078
5079 // Mark the polling page as readable
5080 void os::make_polling_page_readable(void) {
5081 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5082 fatal("Could not enable polling page");
5083 }
5084 }
5085
5086 // older glibc versions don't have this macro (which expands to
5087 // an optimized bit-counting function) so we have to roll our own
5088 #ifndef CPU_COUNT
5089
5090 static int _cpu_count(const cpu_set_t* cpus) {
5091 int count = 0;
5092 // only look up to the number of configured processors
5093 for (int i = 0; i < os::processor_count(); i++) {
5094 if (CPU_ISSET(i, cpus)) {
5095 count++;
5096 }
5097 }
5098 return count;
5099 }
5100
5101 #define CPU_COUNT(cpus) _cpu_count(cpus)
5102
5103 #endif // CPU_COUNT
5104
5105 // Get the current number of available processors for this process.
5106 // This value can change at any time during a process's lifetime.
5107 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5108 // If it appears there may be more than 1024 processors then we do a
5109 // dynamic check - see 6515172 for details.
5110 // If anything goes wrong we fallback to returning the number of online
5111 // processors - which can be greater than the number available to the process.
5112 int os::Linux::active_processor_count() {
5113 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5114 cpu_set_t* cpus_p = &cpus;
5115 int cpus_size = sizeof(cpu_set_t);
5116
5117 int configured_cpus = os::processor_count(); // upper bound on available cpus
5118 int cpu_count = 0;
5119
5120 // old build platforms may not support dynamic cpu sets
5121 #ifdef CPU_ALLOC
5122
5123 // To enable easy testing of the dynamic path on different platforms we
5124 // introduce a diagnostic flag: UseCpuAllocPath
5125 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5126 // kernel may use a mask bigger than cpu_set_t
5127 log_trace(os)("active_processor_count: using dynamic path %s"
5128 "- configured processors: %d",
5129 UseCpuAllocPath ? "(forced) " : "",
5130 configured_cpus);
5131 cpus_p = CPU_ALLOC(configured_cpus);
5132 if (cpus_p != NULL) {
5133 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5134 // zero it just to be safe
5135 CPU_ZERO_S(cpus_size, cpus_p);
5136 }
5137 else {
5138 // failed to allocate so fallback to online cpus
5139 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5140 log_trace(os)("active_processor_count: "
5141 "CPU_ALLOC failed (%s) - using "
5142 "online processor count: %d",
5143 os::strerror(errno), online_cpus);
5144 return online_cpus;
5145 }
5146 }
5147 else {
5148 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5149 configured_cpus);
5150 }
5151 #else // CPU_ALLOC
5152 // these stubs won't be executed
5153 #define CPU_COUNT_S(size, cpus) -1
5154 #define CPU_FREE(cpus)
5155
5156 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5157 configured_cpus);
5158 #endif // CPU_ALLOC
5159
5160 // pid 0 means the current thread - which we have to assume represents the process
5161 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5162 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5163 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5164 }
5165 else {
5166 cpu_count = CPU_COUNT(cpus_p);
5167 }
5168 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5169 }
5170 else {
5171 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5172 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5173 "which may exceed available processors", os::strerror(errno), cpu_count);
5174 }
5175
5176 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5177 CPU_FREE(cpus_p);
5178 }
5179
5180 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5181 return cpu_count;
5182 }
5183
5184 // Determine the active processor count from one of
5185 // three different sources:
5186 //
5187 // 1. User option -XX:ActiveProcessorCount
5188 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5189 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5190 //
5191 // Option 1, if specified, will always override.
5192 // If the cgroup subsystem is active and configured, we
5193 // will return the min of the cgroup and option 2 results.
5194 // This is required since tools, such as numactl, that
5195 // alter cpu affinity do not update cgroup subsystem
5196 // cpuset configuration files.
5197 int os::active_processor_count() {
5198 // User has overridden the number of active processors
5199 if (ActiveProcessorCount > 0) {
5200 log_trace(os)("active_processor_count: "
5201 "active processor count set by user : %d",
5202 ActiveProcessorCount);
5203 return ActiveProcessorCount;
5204 }
5205
5206 int active_cpus;
5207 if (OSContainer::is_containerized()) {
5208 active_cpus = OSContainer::active_processor_count();
5209 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5210 active_cpus);
5211 } else {
5212 active_cpus = os::Linux::active_processor_count();
5213 }
5214
5215 return active_cpus;
5216 }
5217
5218 uint os::processor_id() {
5219 const int id = Linux::sched_getcpu();
5220 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5221 return (uint)id;
5222 }
5223
5224 void os::set_native_thread_name(const char *name) {
5225 if (Linux::_pthread_setname_np) {
5226 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5227 snprintf(buf, sizeof(buf), "%s", name);
5228 buf[sizeof(buf) - 1] = '\0';
5229 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5230 // ERANGE should not happen; all other errors should just be ignored.
5231 assert(rc != ERANGE, "pthread_setname_np failed");
5232 }
5233 }
5234
5235 bool os::distribute_processes(uint length, uint* distribution) {
5236 // Not yet implemented.
5237 return false;
5238 }
5239
5240 bool os::bind_to_processor(uint processor_id) {
5241 // Not yet implemented.
5242 return false;
5243 }
5244
5245 ///
5246
5247 void os::SuspendedThreadTask::internal_do_task() {
5248 if (do_suspend(_thread->osthread())) {
5249 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5250 do_task(context);
5251 do_resume(_thread->osthread());
5252 }
5253 }
5254
5255 ////////////////////////////////////////////////////////////////////////////////
5256 // debug support
5257
5258 bool os::find(address addr, outputStream* st) {
5259 Dl_info dlinfo;
5260 memset(&dlinfo, 0, sizeof(dlinfo));
5261 if (dladdr(addr, &dlinfo) != 0) {
5262 st->print(PTR_FORMAT ": ", p2i(addr));
5263 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5264 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5265 p2i(addr) - p2i(dlinfo.dli_saddr));
5266 } else if (dlinfo.dli_fbase != NULL) {
5267 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5268 } else {
5269 st->print("<absolute address>");
5270 }
5271 if (dlinfo.dli_fname != NULL) {
5272 st->print(" in %s", dlinfo.dli_fname);
5273 }
5274 if (dlinfo.dli_fbase != NULL) {
5275 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5276 }
5277 st->cr();
5278
5279 if (Verbose) {
5280 // decode some bytes around the PC
5281 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5282 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5283 address lowest = (address) dlinfo.dli_sname;
5284 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5285 if (begin < lowest) begin = lowest;
5286 Dl_info dlinfo2;
5287 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5288 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5289 end = (address) dlinfo2.dli_saddr;
5290 }
5291 Disassembler::decode(begin, end, st);
5292 }
5293 return true;
5294 }
5295 return false;
5296 }
5297
5298 ////////////////////////////////////////////////////////////////////////////////
5299 // misc
5300
5301 // This does not do anything on Linux. This is basically a hook for being
5302 // able to use structured exception handling (thread-local exception filters)
5303 // on, e.g., Win32.
5304 void
5305 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5306 JavaCallArguments* args, Thread* thread) {
5307 f(value, method, args, thread);
5308 }
5309
5310 void os::print_statistics() {
5311 }
5312
5313 bool os::message_box(const char* title, const char* message) {
5314 int i;
5315 fdStream err(defaultStream::error_fd());
5316 for (i = 0; i < 78; i++) err.print_raw("=");
5317 err.cr();
5318 err.print_raw_cr(title);
5319 for (i = 0; i < 78; i++) err.print_raw("-");
5320 err.cr();
5321 err.print_raw_cr(message);
5322 for (i = 0; i < 78; i++) err.print_raw("=");
5323 err.cr();
5324
5325 char buf[16];
5326 // Prevent process from exiting upon "read error" without consuming all CPU
5327 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5328
5329 return buf[0] == 'y' || buf[0] == 'Y';
5330 }
5331
5332 // Is a (classpath) directory empty?
5333 bool os::dir_is_empty(const char* path) {
5334 DIR *dir = NULL;
5335 struct dirent *ptr;
5336
5337 dir = opendir(path);
5338 if (dir == NULL) return true;
5339
5340 // Scan the directory
5341 bool result = true;
5342 while (result && (ptr = readdir(dir)) != NULL) {
5343 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5344 result = false;
5345 }
5346 }
5347 closedir(dir);
5348 return result;
5349 }
5350
5351 // This code originates from JDK's sysOpen and open64_w
5352 // from src/solaris/hpi/src/system_md.c
5353
5354 int os::open(const char *path, int oflag, int mode) {
5355 if (strlen(path) > MAX_PATH - 1) {
5356 errno = ENAMETOOLONG;
5357 return -1;
5358 }
5359
5360 // All file descriptors that are opened in the Java process and not
5361 // specifically destined for a subprocess should have the close-on-exec
5362 // flag set. If we don't set it, then careless 3rd party native code
5363 // might fork and exec without closing all appropriate file descriptors
5364 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5365 // turn might:
5366 //
5367 // - cause end-of-file to fail to be detected on some file
5368 // descriptors, resulting in mysterious hangs, or
5369 //
5370 // - might cause an fopen in the subprocess to fail on a system
5371 // suffering from bug 1085341.
5372 //
5373 // (Yes, the default setting of the close-on-exec flag is a Unix
5374 // design flaw)
5375 //
5376 // See:
5377 // 1085341: 32-bit stdio routines should support file descriptors >255
5378 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5379 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5380 //
5381 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5382 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5383 // because it saves a system call and removes a small window where the flag
5384 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5385 // and we fall back to using FD_CLOEXEC (see below).
5386 #ifdef O_CLOEXEC
5387 oflag |= O_CLOEXEC;
5388 #endif
5389
5390 int fd = ::open64(path, oflag, mode);
5391 if (fd == -1) return -1;
5392
5393 //If the open succeeded, the file might still be a directory
5394 {
5395 struct stat64 buf64;
5396 int ret = ::fstat64(fd, &buf64);
5397 int st_mode = buf64.st_mode;
5398
5399 if (ret != -1) {
5400 if ((st_mode & S_IFMT) == S_IFDIR) {
5401 errno = EISDIR;
5402 ::close(fd);
5403 return -1;
5404 }
5405 } else {
5406 ::close(fd);
5407 return -1;
5408 }
5409 }
5410
5411 #ifdef FD_CLOEXEC
5412 // Validate that the use of the O_CLOEXEC flag on open above worked.
5413 // With recent kernels, we will perform this check exactly once.
5414 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5415 if (!O_CLOEXEC_is_known_to_work) {
5416 int flags = ::fcntl(fd, F_GETFD);
5417 if (flags != -1) {
5418 if ((flags & FD_CLOEXEC) != 0)
5419 O_CLOEXEC_is_known_to_work = 1;
5420 else
5421 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5422 }
5423 }
5424 #endif
5425
5426 return fd;
5427 }
5428
5429
5430 // create binary file, rewriting existing file if required
5431 int os::create_binary_file(const char* path, bool rewrite_existing) {
5432 int oflags = O_WRONLY | O_CREAT;
5433 if (!rewrite_existing) {
5434 oflags |= O_EXCL;
5435 }
5436 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5437 }
5438
5439 // return current position of file pointer
5440 jlong os::current_file_offset(int fd) {
5441 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5442 }
5443
5444 // move file pointer to the specified offset
5445 jlong os::seek_to_file_offset(int fd, jlong offset) {
5446 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5447 }
5448
5449 // This code originates from JDK's sysAvailable
5450 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5451
5452 int os::available(int fd, jlong *bytes) {
5453 jlong cur, end;
5454 int mode;
5455 struct stat64 buf64;
5456
5457 if (::fstat64(fd, &buf64) >= 0) {
5458 mode = buf64.st_mode;
5459 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5460 int n;
5461 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5462 *bytes = n;
5463 return 1;
5464 }
5465 }
5466 }
5467 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5468 return 0;
5469 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5470 return 0;
5471 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5472 return 0;
5473 }
5474 *bytes = end - cur;
5475 return 1;
5476 }
5477
5478 // Map a block of memory.
5479 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5480 char *addr, size_t bytes, bool read_only,
5481 bool allow_exec) {
5482 int prot;
5483 int flags = MAP_PRIVATE;
5484
5485 if (read_only) {
5486 prot = PROT_READ;
5487 } else {
5488 prot = PROT_READ | PROT_WRITE;
5489 }
5490
5491 if (allow_exec) {
5492 prot |= PROT_EXEC;
5493 }
5494
5495 if (addr != NULL) {
5496 flags |= MAP_FIXED;
5497 }
5498
5499 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5500 fd, file_offset);
5501 if (mapped_address == MAP_FAILED) {
5502 return NULL;
5503 }
5504 return mapped_address;
5505 }
5506
5507
5508 // Remap a block of memory.
5509 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5510 char *addr, size_t bytes, bool read_only,
5511 bool allow_exec) {
5512 // same as map_memory() on this OS
5513 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5514 allow_exec);
5515 }
5516
5517
5518 // Unmap a block of memory.
5519 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5520 return munmap(addr, bytes) == 0;
5521 }
5522
5523 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5524
5525 static jlong fast_cpu_time(Thread *thread) {
5526 clockid_t clockid;
5527 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5528 &clockid);
5529 if (rc == 0) {
5530 return os::Linux::fast_thread_cpu_time(clockid);
5531 } else {
5532 // It's possible to encounter a terminated native thread that failed
5533 // to detach itself from the VM - which should result in ESRCH.
5534 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5535 return -1;
5536 }
5537 }
5538
5539 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5540 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5541 // of a thread.
5542 //
5543 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5544 // the fast estimate available on the platform.
5545
5546 jlong os::current_thread_cpu_time() {
5547 if (os::Linux::supports_fast_thread_cpu_time()) {
5548 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5549 } else {
5550 // return user + sys since the cost is the same
5551 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5552 }
5553 }
5554
5555 jlong os::thread_cpu_time(Thread* thread) {
5556 // consistent with what current_thread_cpu_time() returns
5557 if (os::Linux::supports_fast_thread_cpu_time()) {
5558 return fast_cpu_time(thread);
5559 } else {
5560 return slow_thread_cpu_time(thread, true /* user + sys */);
5561 }
5562 }
5563
5564 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5565 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5566 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5567 } else {
5568 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5569 }
5570 }
5571
5572 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5573 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5574 return fast_cpu_time(thread);
5575 } else {
5576 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5577 }
5578 }
5579
5580 // -1 on error.
5581 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5582 pid_t tid = thread->osthread()->thread_id();
5583 char *s;
5584 char stat[2048];
5585 int statlen;
5586 char proc_name[64];
5587 int count;
5588 long sys_time, user_time;
5589 char cdummy;
5590 int idummy;
5591 long ldummy;
5592 FILE *fp;
5593
5594 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5595 fp = fopen(proc_name, "r");
5596 if (fp == NULL) return -1;
5597 statlen = fread(stat, 1, 2047, fp);
5598 stat[statlen] = '\0';
5599 fclose(fp);
5600
5601 // Skip pid and the command string. Note that we could be dealing with
5602 // weird command names, e.g. user could decide to rename java launcher
5603 // to "java 1.4.2 :)", then the stat file would look like
5604 // 1234 (java 1.4.2 :)) R ... ...
5605 // We don't really need to know the command string, just find the last
5606 // occurrence of ")" and then start parsing from there. See bug 4726580.
5607 s = strrchr(stat, ')');
5608 if (s == NULL) return -1;
5609
5610 // Skip blank chars
5611 do { s++; } while (s && isspace(*s));
5612
5613 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5614 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5615 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5616 &user_time, &sys_time);
5617 if (count != 13) return -1;
5618 if (user_sys_cpu_time) {
5619 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5620 } else {
5621 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5622 }
5623 }
5624
5625 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5626 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5627 info_ptr->may_skip_backward = false; // elapsed time not wall time
5628 info_ptr->may_skip_forward = false; // elapsed time not wall time
5629 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5630 }
5631
5632 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5633 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5634 info_ptr->may_skip_backward = false; // elapsed time not wall time
5635 info_ptr->may_skip_forward = false; // elapsed time not wall time
5636 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5637 }
5638
5639 bool os::is_thread_cpu_time_supported() {
5640 return true;
5641 }
5642
5643 // System loadavg support. Returns -1 if load average cannot be obtained.
5644 // Linux doesn't yet have a (official) notion of processor sets,
5645 // so just return the system wide load average.
5646 int os::loadavg(double loadavg[], int nelem) {
5647 return ::getloadavg(loadavg, nelem);
5648 }
5649
5650 void os::pause() {
5651 char filename[MAX_PATH];
5652 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5653 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5654 } else {
5655 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5656 }
5657
5658 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5659 if (fd != -1) {
5660 struct stat buf;
5661 ::close(fd);
5662 while (::stat(filename, &buf) == 0) {
5663 (void)::poll(NULL, 0, 100);
5664 }
5665 } else {
5666 jio_fprintf(stderr,
5667 "Could not open pause file '%s', continuing immediately.\n", filename);
5668 }
5669 }
5670
5671 extern char** environ;
5672
5673 // Run the specified command in a separate process. Return its exit value,
5674 // or -1 on failure (e.g. can't fork a new process).
5675 // Unlike system(), this function can be called from signal handler. It
5676 // doesn't block SIGINT et al.
5677 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5678 const char * argv[4] = {"sh", "-c", cmd, NULL};
5679
5680 pid_t pid ;
5681
5682 if (use_vfork_if_available) {
5683 pid = vfork();
5684 } else {
5685 pid = fork();
5686 }
5687
5688 if (pid < 0) {
5689 // fork failed
5690 return -1;
5691
5692 } else if (pid == 0) {
5693 // child process
5694
5695 execve("/bin/sh", (char* const*)argv, environ);
5696
5697 // execve failed
5698 _exit(-1);
5699
5700 } else {
5701 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5702 // care about the actual exit code, for now.
5703
5704 int status;
5705
5706 // Wait for the child process to exit. This returns immediately if
5707 // the child has already exited. */
5708 while (waitpid(pid, &status, 0) < 0) {
5709 switch (errno) {
5710 case ECHILD: return 0;
5711 case EINTR: break;
5712 default: return -1;
5713 }
5714 }
5715
5716 if (WIFEXITED(status)) {
5717 // The child exited normally; get its exit code.
5718 return WEXITSTATUS(status);
5719 } else if (WIFSIGNALED(status)) {
5720 // The child exited because of a signal
5721 // The best value to return is 0x80 + signal number,
5722 // because that is what all Unix shells do, and because
5723 // it allows callers to distinguish between process exit and
5724 // process death by signal.
5725 return 0x80 + WTERMSIG(status);
5726 } else {
5727 // Unknown exit code; pass it through
5728 return status;
5729 }
5730 }
5731 }
5732
5733 // Get the default path to the core file
5734 // Returns the length of the string
5735 int os::get_core_path(char* buffer, size_t bufferSize) {
5736 /*
5737 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5738 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5739 */
5740 const int core_pattern_len = 129;
5741 char core_pattern[core_pattern_len] = {0};
5742
5743 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5744 if (core_pattern_file == -1) {
5745 return -1;
5746 }
5747
5748 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5749 ::close(core_pattern_file);
5750 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5751 return -1;
5752 }
5753 if (core_pattern[ret-1] == '\n') {
5754 core_pattern[ret-1] = '\0';
5755 } else {
5756 core_pattern[ret] = '\0';
5757 }
5758
5759 // Replace the %p in the core pattern with the process id. NOTE: we do this
5760 // only if the pattern doesn't start with "|", and we support only one %p in
5761 // the pattern.
5762 char *pid_pos = strstr(core_pattern, "%p");
5763 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5764 int written;
5765
5766 if (core_pattern[0] == '/') {
5767 if (pid_pos != NULL) {
5768 *pid_pos = '\0';
5769 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5770 current_process_id(), tail);
5771 } else {
5772 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5773 }
5774 } else {
5775 char cwd[PATH_MAX];
5776
5777 const char* p = get_current_directory(cwd, PATH_MAX);
5778 if (p == NULL) {
5779 return -1;
5780 }
5781
5782 if (core_pattern[0] == '|') {
5783 written = jio_snprintf(buffer, bufferSize,
5784 "\"%s\" (or dumping to %s/core.%d)",
5785 &core_pattern[1], p, current_process_id());
5786 } else if (pid_pos != NULL) {
5787 *pid_pos = '\0';
5788 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5789 current_process_id(), tail);
5790 } else {
5791 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5792 }
5793 }
5794
5795 if (written < 0) {
5796 return -1;
5797 }
5798
5799 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5800 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5801
5802 if (core_uses_pid_file != -1) {
5803 char core_uses_pid = 0;
5804 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5805 ::close(core_uses_pid_file);
5806
5807 if (core_uses_pid == '1') {
5808 jio_snprintf(buffer + written, bufferSize - written,
5809 ".%d", current_process_id());
5810 }
5811 }
5812 }
5813
5814 return strlen(buffer);
5815 }
5816
5817 bool os::start_debugging(char *buf, int buflen) {
5818 int len = (int)strlen(buf);
5819 char *p = &buf[len];
5820
5821 jio_snprintf(p, buflen-len,
5822 "\n\n"
5823 "Do you want to debug the problem?\n\n"
5824 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5825 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5826 "Otherwise, press RETURN to abort...",
5827 os::current_process_id(), os::current_process_id(),
5828 os::current_thread_id(), os::current_thread_id());
5829
5830 bool yes = os::message_box("Unexpected Error", buf);
5831
5832 if (yes) {
5833 // yes, user asked VM to launch debugger
5834 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5835 os::current_process_id(), os::current_process_id());
5836
5837 os::fork_and_exec(buf);
5838 yes = false;
5839 }
5840 return yes;
5841 }
5842
5843
5844 // Java/Compiler thread:
5845 //
5846 // Low memory addresses
5847 // P0 +------------------------+
5848 // | |\ Java thread created by VM does not have glibc
5849 // | glibc guard page | - guard page, attached Java thread usually has
5850 // | |/ 1 glibc guard page.
5851 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5852 // | |\
5853 // | HotSpot Guard Pages | - red, yellow and reserved pages
5854 // | |/
5855 // +------------------------+ JavaThread::stack_reserved_zone_base()
5856 // | |\
5857 // | Normal Stack | -
5858 // | |/
5859 // P2 +------------------------+ Thread::stack_base()
5860 //
5861 // Non-Java thread:
5862 //
5863 // Low memory addresses
5864 // P0 +------------------------+
5865 // | |\
5866 // | glibc guard page | - usually 1 page
5867 // | |/
5868 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5869 // | |\
5870 // | Normal Stack | -
5871 // | |/
5872 // P2 +------------------------+ Thread::stack_base()
5873 //
5874 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5875 // returned from pthread_attr_getstack().
5876 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5877 // of the stack size given in pthread_attr. We work around this for
5878 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5879 //
5880 #ifndef ZERO
5881 static void current_stack_region(address * bottom, size_t * size) {
5882 if (os::is_primordial_thread()) {
5883 // primordial thread needs special handling because pthread_getattr_np()
5884 // may return bogus value.
5885 *bottom = os::Linux::initial_thread_stack_bottom();
5886 *size = os::Linux::initial_thread_stack_size();
5887 } else {
5888 pthread_attr_t attr;
5889
5890 int rslt = pthread_getattr_np(pthread_self(), &attr);
5891
5892 // JVM needs to know exact stack location, abort if it fails
5893 if (rslt != 0) {
5894 if (rslt == ENOMEM) {
5895 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5896 } else {
5897 fatal("pthread_getattr_np failed with error = %d", rslt);
5898 }
5899 }
5900
5901 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5902 fatal("Cannot locate current stack attributes!");
5903 }
5904
5905 // Work around NPTL stack guard error.
5906 size_t guard_size = 0;
5907 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5908 if (rslt != 0) {
5909 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5910 }
5911 *bottom += guard_size;
5912 *size -= guard_size;
5913
5914 pthread_attr_destroy(&attr);
5915
5916 }
5917 assert(os::current_stack_pointer() >= *bottom &&
5918 os::current_stack_pointer() < *bottom + *size, "just checking");
5919 }
5920
5921 address os::current_stack_base() {
5922 address bottom;
5923 size_t size;
5924 current_stack_region(&bottom, &size);
5925 return (bottom + size);
5926 }
5927
5928 size_t os::current_stack_size() {
5929 // This stack size includes the usable stack and HotSpot guard pages
5930 // (for the threads that have Hotspot guard pages).
5931 address bottom;
5932 size_t size;
5933 current_stack_region(&bottom, &size);
5934 return size;
5935 }
5936 #endif
5937
5938 static inline struct timespec get_mtime(const char* filename) {
5939 struct stat st;
5940 int ret = os::stat(filename, &st);
5941 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5942 return st.st_mtim;
5943 }
5944
5945 int os::compare_file_modified_times(const char* file1, const char* file2) {
5946 struct timespec filetime1 = get_mtime(file1);
5947 struct timespec filetime2 = get_mtime(file2);
5948 int diff = filetime1.tv_sec - filetime2.tv_sec;
5949 if (diff == 0) {
5950 return filetime1.tv_nsec - filetime2.tv_nsec;
5951 }
5952 return diff;
5953 }
5954
5955 /////////////// Unit tests ///////////////
5956
5957 #ifndef PRODUCT
5958
5959 #define test_log(...) \
5960 do { \
5961 } while (false)
5962
5963 class TestReserveMemorySpecial : AllStatic {
5964 public:
5965 static void small_page_write(void* addr, size_t size) {
5966 size_t page_size = os::vm_page_size();
5967
5968 char* end = (char*)addr + size;
5969 for (char* p = (char*)addr; p < end; p += page_size) {
5970 *p = 1;
5971 }
5972 }
5973
5974 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5975 if (!UseHugeTLBFS) {
5976 return;
5977 }
5978
5979 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5980
5981 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5982
5983 if (addr != NULL) {
5984 small_page_write(addr, size);
5985
5986 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5987 }
5988 }
5989
5990 static void test_reserve_memory_special_huge_tlbfs_only() {
5991 if (!UseHugeTLBFS) {
5992 return;
5993 }
5994
5995 size_t lp = os::large_page_size();
5996
5997 for (size_t size = lp; size <= lp * 10; size += lp) {
5998 test_reserve_memory_special_huge_tlbfs_only(size);
5999 }
6000 }
6001
6002 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6003 size_t lp = os::large_page_size();
6004 size_t ag = os::vm_allocation_granularity();
6005
6006 // sizes to test
6007 const size_t sizes[] = {
6008 lp, lp + ag, lp + lp / 2, lp * 2,
6009 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6010 lp * 10, lp * 10 + lp / 2
6011 };
6012 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6013
6014 // For each size/alignment combination, we test three scenarios:
6015 // 1) with req_addr == NULL
6016 // 2) with a non-null req_addr at which we expect to successfully allocate
6017 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6018 // expect the allocation to either fail or to ignore req_addr
6019
6020 // Pre-allocate two areas; they shall be as large as the largest allocation
6021 // and aligned to the largest alignment we will be testing.
6022 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6023 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6024 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6025 -1, 0);
6026 assert(mapping1 != MAP_FAILED, "should work");
6027
6028 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6029 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6030 -1, 0);
6031 assert(mapping2 != MAP_FAILED, "should work");
6032
6033 // Unmap the first mapping, but leave the second mapping intact: the first
6034 // mapping will serve as a value for a "good" req_addr (case 2). The second
6035 // mapping, still intact, as "bad" req_addr (case 3).
6036 ::munmap(mapping1, mapping_size);
6037
6038 // Case 1
6039 test_log("%s, req_addr NULL:", __FUNCTION__);
6040 test_log("size align result");
6041
6042 for (int i = 0; i < num_sizes; i++) {
6043 const size_t size = sizes[i];
6044 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6045 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6046 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6047 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6048 if (p != NULL) {
6049 assert(is_aligned(p, alignment), "must be");
6050 small_page_write(p, size);
6051 os::Linux::release_memory_special_huge_tlbfs(p, size);
6052 }
6053 }
6054 }
6055
6056 // Case 2
6057 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6058 test_log("size align req_addr result");
6059
6060 for (int i = 0; i < num_sizes; i++) {
6061 const size_t size = sizes[i];
6062 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6063 char* const req_addr = align_up(mapping1, alignment);
6064 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6065 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6066 size, alignment, p2i(req_addr), p2i(p),
6067 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6068 if (p != NULL) {
6069 assert(p == req_addr, "must be");
6070 small_page_write(p, size);
6071 os::Linux::release_memory_special_huge_tlbfs(p, size);
6072 }
6073 }
6074 }
6075
6076 // Case 3
6077 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6078 test_log("size align req_addr result");
6079
6080 for (int i = 0; i < num_sizes; i++) {
6081 const size_t size = sizes[i];
6082 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6083 char* const req_addr = align_up(mapping2, alignment);
6084 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6085 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6086 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6087 // as the area around req_addr contains already existing mappings, the API should always
6088 // return NULL (as per contract, it cannot return another address)
6089 assert(p == NULL, "must be");
6090 }
6091 }
6092
6093 ::munmap(mapping2, mapping_size);
6094
6095 }
6096
6097 static void test_reserve_memory_special_huge_tlbfs() {
6098 if (!UseHugeTLBFS) {
6099 return;
6100 }
6101
6102 test_reserve_memory_special_huge_tlbfs_only();
6103 test_reserve_memory_special_huge_tlbfs_mixed();
6104 }
6105
6106 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6107 if (!UseSHM) {
6108 return;
6109 }
6110
6111 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6112
6113 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6114
6115 if (addr != NULL) {
6116 assert(is_aligned(addr, alignment), "Check");
6117 assert(is_aligned(addr, os::large_page_size()), "Check");
6118
6119 small_page_write(addr, size);
6120
6121 os::Linux::release_memory_special_shm(addr, size);
6122 }
6123 }
6124
6125 static void test_reserve_memory_special_shm() {
6126 size_t lp = os::large_page_size();
6127 size_t ag = os::vm_allocation_granularity();
6128
6129 for (size_t size = ag; size < lp * 3; size += ag) {
6130 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6131 test_reserve_memory_special_shm(size, alignment);
6132 }
6133 }
6134 }
6135
6136 static void test() {
6137 test_reserve_memory_special_huge_tlbfs();
6138 test_reserve_memory_special_shm();
6139 }
6140 };
6141
6142 void TestReserveMemorySpecial_test() {
6143 TestReserveMemorySpecial::test();
6144 }
6145
6146 #endif