1 /*
2 * Copyright (c) 1999, 2019, 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 "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_posix.inline.hpp"
42 #include "os_share_linux.hpp"
43 #include "osContainer_linux.hpp"
44 #include "prims/jniFastGetField.hpp"
45 #include "prims/jvm_misc.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/atomic.hpp"
48 #include "runtime/extendedPC.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/interfaceSupport.inline.hpp"
51 #include "runtime/init.hpp"
52 #include "runtime/java.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/mutexLocker.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/orderAccess.hpp"
57 #include "runtime/osThread.hpp"
58 #include "runtime/perfMemory.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/statSampler.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "runtime/thread.inline.hpp"
63 #include "runtime/threadCritical.hpp"
64 #include "runtime/threadSMR.hpp"
65 #include "runtime/timer.hpp"
66 #include "semaphore_posix.hpp"
67 #include "services/attachListener.hpp"
68 #include "services/memTracker.hpp"
69 #include "services/runtimeService.hpp"
70 #include "utilities/align.hpp"
71 #include "utilities/decoder.hpp"
72 #include "utilities/defaultStream.hpp"
73 #include "utilities/events.hpp"
74 #include "utilities/elfFile.hpp"
75 #include "utilities/growableArray.hpp"
76 #include "utilities/macros.hpp"
77 #include "utilities/vmError.hpp"
78
79 // put OS-includes here
80 # include <sys/types.h>
81 # include <sys/mman.h>
82 # include <sys/stat.h>
83 # include <sys/select.h>
84 # include <pthread.h>
85 # include <signal.h>
86 # include <errno.h>
87 # include <dlfcn.h>
88 # include <stdio.h>
89 # include <unistd.h>
90 # include <sys/resource.h>
91 # include <pthread.h>
92 # include <sys/stat.h>
93 # include <sys/time.h>
94 # include <sys/times.h>
95 # include <sys/utsname.h>
96 # include <sys/socket.h>
97 # include <sys/wait.h>
98 # include <pwd.h>
99 # include <poll.h>
100 # include <fcntl.h>
101 # include <string.h>
102 # include <syscall.h>
103 # include <sys/sysinfo.h>
104 # include <gnu/libc-version.h>
105 # include <sys/ipc.h>
106 # include <sys/shm.h>
107 # include <link.h>
108 # include <stdint.h>
109 # include <inttypes.h>
110 # include <sys/ioctl.h>
111
112 #ifndef _GNU_SOURCE
113 #define _GNU_SOURCE
114 #include <sched.h>
115 #undef _GNU_SOURCE
116 #else
117 #include <sched.h>
118 #endif
119
120 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
121 // getrusage() is prepared to handle the associated failure.
122 #ifndef RUSAGE_THREAD
123 #define RUSAGE_THREAD (1) /* only the calling thread */
124 #endif
125
126 #define MAX_PATH (2 * K)
127
128 #define MAX_SECS 100000000
129
130 // for timer info max values which include all bits
131 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
132
133 enum CoredumpFilterBit {
134 FILE_BACKED_PVT_BIT = 1 << 2,
135 FILE_BACKED_SHARED_BIT = 1 << 3,
136 LARGEPAGES_BIT = 1 << 6,
137 DAX_SHARED_BIT = 1 << 8
138 };
139
140 ////////////////////////////////////////////////////////////////////////////////
141 // global variables
142 julong os::Linux::_physical_memory = 0;
143
144 address os::Linux::_initial_thread_stack_bottom = NULL;
145 uintptr_t os::Linux::_initial_thread_stack_size = 0;
146
147 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
148 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
149 Mutex* os::Linux::_createThread_lock = NULL;
150 pthread_t os::Linux::_main_thread;
151 int os::Linux::_page_size = -1;
152 bool os::Linux::_supports_fast_thread_cpu_time = false;
153 uint32_t os::Linux::_os_version = 0;
154 const char * os::Linux::_glibc_version = NULL;
155 const char * os::Linux::_libpthread_version = NULL;
156
157 static jlong initial_time_count=0;
158
159 static int clock_tics_per_sec = 100;
160
161 // If the VM might have been created on the primordial thread, we need to resolve the
162 // primordial thread stack bounds and check if the current thread might be the
163 // primordial thread in places. If we know that the primordial thread is never used,
164 // such as when the VM was created by one of the standard java launchers, we can
165 // avoid this
166 static bool suppress_primordial_thread_resolution = false;
167
168 // For diagnostics to print a message once. see run_periodic_checks
169 static sigset_t check_signal_done;
170 static bool check_signals = true;
171
172 // Signal number used to suspend/resume a thread
173
174 // do not use any signal number less than SIGSEGV, see 4355769
175 static int SR_signum = SIGUSR2;
176 sigset_t SR_sigset;
177
178 // utility functions
179
180 static int SR_initialize();
181
182 julong os::available_memory() {
183 return Linux::available_memory();
184 }
185
186 julong os::Linux::available_memory() {
187 // values in struct sysinfo are "unsigned long"
188 struct sysinfo si;
189 julong avail_mem;
190
191 if (OSContainer::is_containerized()) {
192 jlong mem_limit, mem_usage;
193 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
194 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
195 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
196 }
197 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
198 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
199 }
200 if (mem_limit > 0 && mem_usage > 0 ) {
201 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
202 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
203 return avail_mem;
204 }
205 }
206
207 sysinfo(&si);
208 avail_mem = (julong)si.freeram * si.mem_unit;
209 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
210 return avail_mem;
211 }
212
213 julong os::physical_memory() {
214 jlong phys_mem = 0;
215 if (OSContainer::is_containerized()) {
216 jlong mem_limit;
217 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
218 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
219 return mem_limit;
220 }
221 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
222 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
223 }
224
225 phys_mem = Linux::physical_memory();
226 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
227 return phys_mem;
228 }
229
230 // Return true if user is running as root.
231
232 bool os::have_special_privileges() {
233 static bool init = false;
234 static bool privileges = false;
235 if (!init) {
236 privileges = (getuid() != geteuid()) || (getgid() != getegid());
237 init = true;
238 }
239 return privileges;
240 }
241
242
243 #ifndef SYS_gettid
244 // i386: 224, ia64: 1105, amd64: 186, sparc 143
245 #ifdef __ia64__
246 #define SYS_gettid 1105
247 #else
248 #ifdef __i386__
249 #define SYS_gettid 224
250 #else
251 #ifdef __amd64__
252 #define SYS_gettid 186
253 #else
254 #ifdef __sparc__
255 #define SYS_gettid 143
256 #else
257 #error define gettid for the arch
258 #endif
259 #endif
260 #endif
261 #endif
262 #endif
263
264
265 // pid_t gettid()
266 //
267 // Returns the kernel thread id of the currently running thread. Kernel
268 // thread id is used to access /proc.
269 pid_t os::Linux::gettid() {
270 int rslt = syscall(SYS_gettid);
271 assert(rslt != -1, "must be."); // old linuxthreads implementation?
272 return (pid_t)rslt;
273 }
274
275 // Most versions of linux have a bug where the number of processors are
276 // determined by looking at the /proc file system. In a chroot environment,
277 // the system call returns 1.
278 static bool unsafe_chroot_detected = false;
279 static const char *unstable_chroot_error = "/proc file system not found.\n"
280 "Java may be unstable running multithreaded in a chroot "
281 "environment on Linux when /proc filesystem is not mounted.";
282
283 void os::Linux::initialize_system_info() {
284 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
285 if (processor_count() == 1) {
286 pid_t pid = os::Linux::gettid();
287 char fname[32];
288 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
289 FILE *fp = fopen(fname, "r");
290 if (fp == NULL) {
291 unsafe_chroot_detected = true;
292 } else {
293 fclose(fp);
294 }
295 }
296 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
297 assert(processor_count() > 0, "linux error");
298 }
299
300 void os::init_system_properties_values() {
301 // The next steps are taken in the product version:
302 //
303 // Obtain the JAVA_HOME value from the location of libjvm.so.
304 // This library should be located at:
305 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
306 //
307 // If "/jre/lib/" appears at the right place in the path, then we
308 // assume libjvm.so is installed in a JDK and we use this path.
309 //
310 // Otherwise exit with message: "Could not create the Java virtual machine."
311 //
312 // The following extra steps are taken in the debugging version:
313 //
314 // If "/jre/lib/" does NOT appear at the right place in the path
315 // instead of exit check for $JAVA_HOME environment variable.
316 //
317 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
318 // then we append a fake suffix "hotspot/libjvm.so" to this path so
319 // it looks like libjvm.so is installed there
320 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
321 //
322 // Otherwise exit.
323 //
324 // Important note: if the location of libjvm.so changes this
325 // code needs to be changed accordingly.
326
327 // See ld(1):
328 // The linker uses the following search paths to locate required
329 // shared libraries:
330 // 1: ...
331 // ...
332 // 7: The default directories, normally /lib and /usr/lib.
333 #ifndef OVERRIDE_LIBPATH
334 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
335 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
336 #else
337 #define DEFAULT_LIBPATH "/lib:/usr/lib"
338 #endif
339 #else
340 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
341 #endif
342
343 // Base path of extensions installed on the system.
344 #define SYS_EXT_DIR "/usr/java/packages"
345 #define EXTENSIONS_DIR "/lib/ext"
346
347 // Buffer that fits several sprintfs.
348 // Note that the space for the colon and the trailing null are provided
349 // by the nulls included by the sizeof operator.
350 const size_t bufsize =
351 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
352 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
353 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
354
355 // sysclasspath, java_home, dll_dir
356 {
357 char *pslash;
358 os::jvm_path(buf, bufsize);
359
360 // Found the full path to libjvm.so.
361 // Now cut the path to <java_home>/jre if we can.
362 pslash = strrchr(buf, '/');
363 if (pslash != NULL) {
364 *pslash = '\0'; // Get rid of /libjvm.so.
365 }
366 pslash = strrchr(buf, '/');
367 if (pslash != NULL) {
368 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
369 }
370 Arguments::set_dll_dir(buf);
371
372 if (pslash != NULL) {
373 pslash = strrchr(buf, '/');
374 if (pslash != NULL) {
375 *pslash = '\0'; // Get rid of /lib.
376 }
377 }
378 Arguments::set_java_home(buf);
379 if (!set_boot_path('/', ':')) {
380 vm_exit_during_initialization("Failed setting boot class path.", NULL);
381 }
382 }
383
384 // Where to look for native libraries.
385 //
386 // Note: Due to a legacy implementation, most of the library path
387 // is set in the launcher. This was to accomodate linking restrictions
388 // on legacy Linux implementations (which are no longer supported).
389 // Eventually, all the library path setting will be done here.
390 //
391 // However, to prevent the proliferation of improperly built native
392 // libraries, the new path component /usr/java/packages is added here.
393 // Eventually, all the library path setting will be done here.
394 {
395 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
396 // should always exist (until the legacy problem cited above is
397 // addressed).
398 const char *v = ::getenv("LD_LIBRARY_PATH");
399 const char *v_colon = ":";
400 if (v == NULL) { v = ""; v_colon = ""; }
401 // That's +1 for the colon and +1 for the trailing '\0'.
402 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
403 strlen(v) + 1 +
404 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
405 mtInternal);
406 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
407 Arguments::set_library_path(ld_library_path);
408 FREE_C_HEAP_ARRAY(char, ld_library_path);
409 }
410
411 // Extensions directories.
412 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
413 Arguments::set_ext_dirs(buf);
414
415 FREE_C_HEAP_ARRAY(char, buf);
416
417 #undef DEFAULT_LIBPATH
418 #undef SYS_EXT_DIR
419 #undef EXTENSIONS_DIR
420 }
421
422 ////////////////////////////////////////////////////////////////////////////////
423 // breakpoint support
424
425 void os::breakpoint() {
426 BREAKPOINT;
427 }
428
429 extern "C" void breakpoint() {
430 // use debugger to set breakpoint here
431 }
432
433 ////////////////////////////////////////////////////////////////////////////////
434 // signal support
435
436 debug_only(static bool signal_sets_initialized = false);
437 static sigset_t unblocked_sigs, vm_sigs;
438
439 void os::Linux::signal_sets_init() {
440 // Should also have an assertion stating we are still single-threaded.
441 assert(!signal_sets_initialized, "Already initialized");
442 // Fill in signals that are necessarily unblocked for all threads in
443 // the VM. Currently, we unblock the following signals:
444 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
445 // by -Xrs (=ReduceSignalUsage));
446 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
447 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
448 // the dispositions or masks wrt these signals.
449 // Programs embedding the VM that want to use the above signals for their
450 // own purposes must, at this time, use the "-Xrs" option to prevent
451 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
452 // (See bug 4345157, and other related bugs).
453 // In reality, though, unblocking these signals is really a nop, since
454 // these signals are not blocked by default.
455 sigemptyset(&unblocked_sigs);
456 sigaddset(&unblocked_sigs, SIGILL);
457 sigaddset(&unblocked_sigs, SIGSEGV);
458 sigaddset(&unblocked_sigs, SIGBUS);
459 sigaddset(&unblocked_sigs, SIGFPE);
460 #if defined(PPC64)
461 sigaddset(&unblocked_sigs, SIGTRAP);
462 #endif
463 sigaddset(&unblocked_sigs, SR_signum);
464
465 if (!ReduceSignalUsage) {
466 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
468 }
469 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
470 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
471 }
472 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
473 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
474 }
475 }
476 // Fill in signals that are blocked by all but the VM thread.
477 sigemptyset(&vm_sigs);
478 if (!ReduceSignalUsage) {
479 sigaddset(&vm_sigs, BREAK_SIGNAL);
480 }
481 debug_only(signal_sets_initialized = true);
482
483 }
484
485 // These are signals that are unblocked while a thread is running Java.
486 // (For some reason, they get blocked by default.)
487 sigset_t* os::Linux::unblocked_signals() {
488 assert(signal_sets_initialized, "Not initialized");
489 return &unblocked_sigs;
490 }
491
492 // These are the signals that are blocked while a (non-VM) thread is
493 // running Java. Only the VM thread handles these signals.
494 sigset_t* os::Linux::vm_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &vm_sigs;
497 }
498
499 void os::Linux::hotspot_sigmask(Thread* thread) {
500
501 //Save caller's signal mask before setting VM signal mask
502 sigset_t caller_sigmask;
503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
504
505 OSThread* osthread = thread->osthread();
506 osthread->set_caller_sigmask(caller_sigmask);
507
508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
509
510 if (!ReduceSignalUsage) {
511 if (thread->is_VM_thread()) {
512 // Only the VM thread handles BREAK_SIGNAL ...
513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
514 } else {
515 // ... all other threads block BREAK_SIGNAL
516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
517 }
518 }
519 }
520
521 //////////////////////////////////////////////////////////////////////////////
522 // detecting pthread library
523
524 void os::Linux::libpthread_init() {
525 // Save glibc and pthread version strings.
526 #if !defined(_CS_GNU_LIBC_VERSION) || \
527 !defined(_CS_GNU_LIBPTHREAD_VERSION)
528 #error "glibc too old (< 2.3.2)"
529 #endif
530
531 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
532 assert(n > 0, "cannot retrieve glibc version");
533 char *str = (char *)malloc(n, mtInternal);
534 confstr(_CS_GNU_LIBC_VERSION, str, n);
535 os::Linux::set_glibc_version(str);
536
537 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
538 assert(n > 0, "cannot retrieve pthread version");
539 str = (char *)malloc(n, mtInternal);
540 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
541 os::Linux::set_libpthread_version(str);
542 }
543
544 /////////////////////////////////////////////////////////////////////////////
545 // thread stack expansion
546
547 // os::Linux::manually_expand_stack() takes care of expanding the thread
548 // stack. Note that this is normally not needed: pthread stacks allocate
549 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
550 // committed. Therefore it is not necessary to expand the stack manually.
551 //
552 // Manually expanding the stack was historically needed on LinuxThreads
553 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
554 // it is kept to deal with very rare corner cases:
555 //
556 // For one, user may run the VM on an own implementation of threads
557 // whose stacks are - like the old LinuxThreads - implemented using
558 // mmap(MAP_GROWSDOWN).
559 //
560 // Also, this coding may be needed if the VM is running on the primordial
561 // thread. Normally we avoid running on the primordial thread; however,
562 // user may still invoke the VM on the primordial thread.
563 //
564 // The following historical comment describes the details about running
565 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
566
567
568 // Force Linux kernel to expand current thread stack. If "bottom" is close
569 // to the stack guard, caller should block all signals.
570 //
571 // MAP_GROWSDOWN:
572 // A special mmap() flag that is used to implement thread stacks. It tells
573 // kernel that the memory region should extend downwards when needed. This
574 // allows early versions of LinuxThreads to only mmap the first few pages
575 // when creating a new thread. Linux kernel will automatically expand thread
576 // stack as needed (on page faults).
577 //
578 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
579 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
580 // region, it's hard to tell if the fault is due to a legitimate stack
581 // access or because of reading/writing non-exist memory (e.g. buffer
582 // overrun). As a rule, if the fault happens below current stack pointer,
583 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
584 // application (see Linux kernel fault.c).
585 //
586 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
587 // stack overflow detection.
588 //
589 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
590 // not use MAP_GROWSDOWN.
591 //
592 // To get around the problem and allow stack banging on Linux, we need to
593 // manually expand thread stack after receiving the SIGSEGV.
594 //
595 // There are two ways to expand thread stack to address "bottom", we used
596 // both of them in JVM before 1.5:
597 // 1. adjust stack pointer first so that it is below "bottom", and then
598 // touch "bottom"
599 // 2. mmap() the page in question
600 //
601 // Now alternate signal stack is gone, it's harder to use 2. For instance,
602 // if current sp is already near the lower end of page 101, and we need to
603 // call mmap() to map page 100, it is possible that part of the mmap() frame
604 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
605 // That will destroy the mmap() frame and cause VM to crash.
606 //
607 // The following code works by adjusting sp first, then accessing the "bottom"
608 // page to force a page fault. Linux kernel will then automatically expand the
609 // stack mapping.
610 //
611 // _expand_stack_to() assumes its frame size is less than page size, which
612 // should always be true if the function is not inlined.
613
614 static void NOINLINE _expand_stack_to(address bottom) {
615 address sp;
616 size_t size;
617 volatile char *p;
618
619 // Adjust bottom to point to the largest address within the same page, it
620 // gives us a one-page buffer if alloca() allocates slightly more memory.
621 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
622 bottom += os::Linux::page_size() - 1;
623
624 // sp might be slightly above current stack pointer; if that's the case, we
625 // will alloca() a little more space than necessary, which is OK. Don't use
626 // os::current_stack_pointer(), as its result can be slightly below current
627 // stack pointer, causing us to not alloca enough to reach "bottom".
628 sp = (address)&sp;
629
630 if (sp > bottom) {
631 size = sp - bottom;
632 p = (volatile char *)alloca(size);
633 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
634 p[0] = '\0';
635 }
636 }
637
638 void os::Linux::expand_stack_to(address bottom) {
639 _expand_stack_to(bottom);
640 }
641
642 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
643 assert(t!=NULL, "just checking");
644 assert(t->osthread()->expanding_stack(), "expand should be set");
645 assert(t->stack_base() != NULL, "stack_base was not initialized");
646
647 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
648 sigset_t mask_all, old_sigset;
649 sigfillset(&mask_all);
650 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
651 _expand_stack_to(addr);
652 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
653 return true;
654 }
655 return false;
656 }
657
658 //////////////////////////////////////////////////////////////////////////////
659 // create new thread
660
661 // Thread start routine for all newly created threads
662 static void *thread_native_entry(Thread *thread) {
663
664 thread->record_stack_base_and_size();
665
666 // Try to randomize the cache line index of hot stack frames.
667 // This helps when threads of the same stack traces evict each other's
668 // cache lines. The threads can be either from the same JVM instance, or
669 // from different JVM instances. The benefit is especially true for
670 // processors with hyperthreading technology.
671 static int counter = 0;
672 int pid = os::current_process_id();
673 alloca(((pid ^ counter++) & 7) * 128);
674
675 thread->initialize_thread_current();
676
677 OSThread* osthread = thread->osthread();
678 Monitor* sync = osthread->startThread_lock();
679
680 osthread->set_thread_id(os::current_thread_id());
681
682 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
683 os::current_thread_id(), (uintx) pthread_self());
684
685 if (UseNUMA) {
686 int lgrp_id = os::numa_get_group_id();
687 if (lgrp_id != -1) {
688 thread->set_lgrp_id(lgrp_id);
689 }
690 }
691 // initialize signal mask for this thread
692 os::Linux::hotspot_sigmask(thread);
693
694 // initialize floating point control register
695 os::Linux::init_thread_fpu_state();
696
697 // handshaking with parent thread
698 {
699 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
700
701 // notify parent thread
702 osthread->set_state(INITIALIZED);
703 sync->notify_all();
704
705 // wait until os::start_thread()
706 while (osthread->get_state() == INITIALIZED) {
707 sync->wait(Mutex::_no_safepoint_check_flag);
708 }
709 }
710
711 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
712
713 // call one more level start routine
714 thread->call_run();
715
716 // Note: at this point the thread object may already have deleted itself.
717 // Prevent dereferencing it from here on out.
718 thread = NULL;
719
720 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
721 os::current_thread_id(), (uintx) pthread_self());
722
723 return 0;
724 }
725
726 bool os::create_thread(Thread* thread, ThreadType thr_type,
727 size_t req_stack_size) {
728 assert(thread->osthread() == NULL, "caller responsible");
729
730 // Allocate the OSThread object
731 OSThread* osthread = new OSThread(NULL, NULL);
732 if (osthread == NULL) {
733 return false;
734 }
735
736 // set the correct thread state
737 osthread->set_thread_type(thr_type);
738
739 // Initial state is ALLOCATED but not INITIALIZED
740 osthread->set_state(ALLOCATED);
741
742 thread->set_osthread(osthread);
743
744 // init thread attributes
745 pthread_attr_t attr;
746 pthread_attr_init(&attr);
747 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
748
749 // Calculate stack size if it's not specified by caller.
750 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
751 // In the Linux NPTL pthread implementation the guard size mechanism
752 // is not implemented properly. The posix standard requires adding
753 // the size of the guard pages to the stack size, instead Linux
754 // takes the space out of 'stacksize'. Thus we adapt the requested
755 // stack_size by the size of the guard pages to mimick proper
756 // behaviour. However, be careful not to end up with a size
757 // of zero due to overflow. Don't add the guard page in that case.
758 size_t guard_size = os::Linux::default_guard_size(thr_type);
759 if (stack_size <= SIZE_MAX - guard_size) {
760 stack_size += guard_size;
761 }
762 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
763
764 int status = pthread_attr_setstacksize(&attr, stack_size);
765 assert_status(status == 0, status, "pthread_attr_setstacksize");
766
767 // Configure glibc guard page.
768 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
769
770 ThreadState state;
771
772 {
773 pthread_t tid;
774 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
775
776 char buf[64];
777 if (ret == 0) {
778 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
779 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
780 } else {
781 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
782 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
783 }
784
785 pthread_attr_destroy(&attr);
786
787 if (ret != 0) {
788 // Need to clean up stuff we've allocated so far
789 thread->set_osthread(NULL);
790 delete osthread;
791 return false;
792 }
793
794 // Store pthread info into the OSThread
795 osthread->set_pthread_id(tid);
796
797 // Wait until child thread is either initialized or aborted
798 {
799 Monitor* sync_with_child = osthread->startThread_lock();
800 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
801 while ((state = osthread->get_state()) == ALLOCATED) {
802 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
803 }
804 }
805 }
806
807 // Aborted due to thread limit being reached
808 if (state == ZOMBIE) {
809 thread->set_osthread(NULL);
810 delete osthread;
811 return false;
812 }
813
814 // The thread is returned suspended (in state INITIALIZED),
815 // and is started higher up in the call chain
816 assert(state == INITIALIZED, "race condition");
817 return true;
818 }
819
820 /////////////////////////////////////////////////////////////////////////////
821 // attach existing thread
822
823 // bootstrap the main thread
824 bool os::create_main_thread(JavaThread* thread) {
825 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
826 return create_attached_thread(thread);
827 }
828
829 bool os::create_attached_thread(JavaThread* thread) {
830 #ifdef ASSERT
831 thread->verify_not_published();
832 #endif
833
834 // Allocate the OSThread object
835 OSThread* osthread = new OSThread(NULL, NULL);
836
837 if (osthread == NULL) {
838 return false;
839 }
840
841 // Store pthread info into the OSThread
842 osthread->set_thread_id(os::Linux::gettid());
843 osthread->set_pthread_id(::pthread_self());
844
845 // initialize floating point control register
846 os::Linux::init_thread_fpu_state();
847
848 // Initial thread state is RUNNABLE
849 osthread->set_state(RUNNABLE);
850
851 thread->set_osthread(osthread);
852
853 if (UseNUMA) {
854 int lgrp_id = os::numa_get_group_id();
855 if (lgrp_id != -1) {
856 thread->set_lgrp_id(lgrp_id);
857 }
858 }
859
860 if (os::is_primordial_thread()) {
861 // If current thread is primordial thread, its stack is mapped on demand,
862 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
863 // the entire stack region to avoid SEGV in stack banging.
864 // It is also useful to get around the heap-stack-gap problem on SuSE
865 // kernel (see 4821821 for details). We first expand stack to the top
866 // of yellow zone, then enable stack yellow zone (order is significant,
867 // enabling yellow zone first will crash JVM on SuSE Linux), so there
868 // is no gap between the last two virtual memory regions.
869
870 JavaThread *jt = (JavaThread *)thread;
871 address addr = jt->stack_reserved_zone_base();
872 assert(addr != NULL, "initialization problem?");
873 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
874
875 osthread->set_expanding_stack();
876 os::Linux::manually_expand_stack(jt, addr);
877 osthread->clear_expanding_stack();
878 }
879
880 // initialize signal mask for this thread
881 // and save the caller's signal mask
882 os::Linux::hotspot_sigmask(thread);
883
884 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
885 os::current_thread_id(), (uintx) pthread_self());
886
887 return true;
888 }
889
890 void os::pd_start_thread(Thread* thread) {
891 OSThread * osthread = thread->osthread();
892 assert(osthread->get_state() != INITIALIZED, "just checking");
893 Monitor* sync_with_child = osthread->startThread_lock();
894 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
895 sync_with_child->notify();
896 }
897
898 // Free Linux resources related to the OSThread
899 void os::free_thread(OSThread* osthread) {
900 assert(osthread != NULL, "osthread not set");
901
902 // We are told to free resources of the argument thread,
903 // but we can only really operate on the current thread.
904 assert(Thread::current()->osthread() == osthread,
905 "os::free_thread but not current thread");
906
907 #ifdef ASSERT
908 sigset_t current;
909 sigemptyset(¤t);
910 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
911 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
912 #endif
913
914 // Restore caller's signal mask
915 sigset_t sigmask = osthread->caller_sigmask();
916 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
917
918 delete osthread;
919 }
920
921 //////////////////////////////////////////////////////////////////////////////
922 // primordial thread
923
924 // Check if current thread is the primordial thread, similar to Solaris thr_main.
925 bool os::is_primordial_thread(void) {
926 if (suppress_primordial_thread_resolution) {
927 return false;
928 }
929 char dummy;
930 // If called before init complete, thread stack bottom will be null.
931 // Can be called if fatal error occurs before initialization.
932 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
933 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
934 os::Linux::initial_thread_stack_size() != 0,
935 "os::init did not locate primordial thread's stack region");
936 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
937 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
938 os::Linux::initial_thread_stack_size()) {
939 return true;
940 } else {
941 return false;
942 }
943 }
944
945 // Find the virtual memory area that contains addr
946 static bool find_vma(address addr, address* vma_low, address* vma_high) {
947 FILE *fp = fopen("/proc/self/maps", "r");
948 if (fp) {
949 address low, high;
950 while (!feof(fp)) {
951 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
952 if (low <= addr && addr < high) {
953 if (vma_low) *vma_low = low;
954 if (vma_high) *vma_high = high;
955 fclose(fp);
956 return true;
957 }
958 }
959 for (;;) {
960 int ch = fgetc(fp);
961 if (ch == EOF || ch == (int)'\n') break;
962 }
963 }
964 fclose(fp);
965 }
966 return false;
967 }
968
969 // Locate primordial thread stack. This special handling of primordial thread stack
970 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
971 // bogus value for the primordial process thread. While the launcher has created
972 // the VM in a new thread since JDK 6, we still have to allow for the use of the
973 // JNI invocation API from a primordial thread.
974 void os::Linux::capture_initial_stack(size_t max_size) {
975
976 // max_size is either 0 (which means accept OS default for thread stacks) or
977 // a user-specified value known to be at least the minimum needed. If we
978 // are actually on the primordial thread we can make it appear that we have a
979 // smaller max_size stack by inserting the guard pages at that location. But we
980 // cannot do anything to emulate a larger stack than what has been provided by
981 // the OS or threading library. In fact if we try to use a stack greater than
982 // what is set by rlimit then we will crash the hosting process.
983
984 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
985 // If this is "unlimited" then it will be a huge value.
986 struct rlimit rlim;
987 getrlimit(RLIMIT_STACK, &rlim);
988 size_t stack_size = rlim.rlim_cur;
989
990 // 6308388: a bug in ld.so will relocate its own .data section to the
991 // lower end of primordial stack; reduce ulimit -s value a little bit
992 // so we won't install guard page on ld.so's data section.
993 // But ensure we don't underflow the stack size - allow 1 page spare
994 if (stack_size >= (size_t)(3 * page_size())) {
995 stack_size -= 2 * page_size();
996 }
997
998 // Try to figure out where the stack base (top) is. This is harder.
999 //
1000 // When an application is started, glibc saves the initial stack pointer in
1001 // a global variable "__libc_stack_end", which is then used by system
1002 // libraries. __libc_stack_end should be pretty close to stack top. The
1003 // variable is available since the very early days. However, because it is
1004 // a private interface, it could disappear in the future.
1005 //
1006 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1007 // to __libc_stack_end, it is very close to stack top, but isn't the real
1008 // stack top. Note that /proc may not exist if VM is running as a chroot
1009 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1010 // /proc/<pid>/stat could change in the future (though unlikely).
1011 //
1012 // We try __libc_stack_end first. If that doesn't work, look for
1013 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1014 // as a hint, which should work well in most cases.
1015
1016 uintptr_t stack_start;
1017
1018 // try __libc_stack_end first
1019 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1020 if (p && *p) {
1021 stack_start = *p;
1022 } else {
1023 // see if we can get the start_stack field from /proc/self/stat
1024 FILE *fp;
1025 int pid;
1026 char state;
1027 int ppid;
1028 int pgrp;
1029 int session;
1030 int nr;
1031 int tpgrp;
1032 unsigned long flags;
1033 unsigned long minflt;
1034 unsigned long cminflt;
1035 unsigned long majflt;
1036 unsigned long cmajflt;
1037 unsigned long utime;
1038 unsigned long stime;
1039 long cutime;
1040 long cstime;
1041 long prio;
1042 long nice;
1043 long junk;
1044 long it_real;
1045 uintptr_t start;
1046 uintptr_t vsize;
1047 intptr_t rss;
1048 uintptr_t rsslim;
1049 uintptr_t scodes;
1050 uintptr_t ecode;
1051 int i;
1052
1053 // Figure what the primordial thread stack base is. Code is inspired
1054 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1055 // followed by command name surrounded by parentheses, state, etc.
1056 char stat[2048];
1057 int statlen;
1058
1059 fp = fopen("/proc/self/stat", "r");
1060 if (fp) {
1061 statlen = fread(stat, 1, 2047, fp);
1062 stat[statlen] = '\0';
1063 fclose(fp);
1064
1065 // Skip pid and the command string. Note that we could be dealing with
1066 // weird command names, e.g. user could decide to rename java launcher
1067 // to "java 1.4.2 :)", then the stat file would look like
1068 // 1234 (java 1.4.2 :)) R ... ...
1069 // We don't really need to know the command string, just find the last
1070 // occurrence of ")" and then start parsing from there. See bug 4726580.
1071 char * s = strrchr(stat, ')');
1072
1073 i = 0;
1074 if (s) {
1075 // Skip blank chars
1076 do { s++; } while (s && isspace(*s));
1077
1078 #define _UFM UINTX_FORMAT
1079 #define _DFM INTX_FORMAT
1080
1081 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1082 // 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
1083 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,
1084 &state, // 3 %c
1085 &ppid, // 4 %d
1086 &pgrp, // 5 %d
1087 &session, // 6 %d
1088 &nr, // 7 %d
1089 &tpgrp, // 8 %d
1090 &flags, // 9 %lu
1091 &minflt, // 10 %lu
1092 &cminflt, // 11 %lu
1093 &majflt, // 12 %lu
1094 &cmajflt, // 13 %lu
1095 &utime, // 14 %lu
1096 &stime, // 15 %lu
1097 &cutime, // 16 %ld
1098 &cstime, // 17 %ld
1099 &prio, // 18 %ld
1100 &nice, // 19 %ld
1101 &junk, // 20 %ld
1102 &it_real, // 21 %ld
1103 &start, // 22 UINTX_FORMAT
1104 &vsize, // 23 UINTX_FORMAT
1105 &rss, // 24 INTX_FORMAT
1106 &rsslim, // 25 UINTX_FORMAT
1107 &scodes, // 26 UINTX_FORMAT
1108 &ecode, // 27 UINTX_FORMAT
1109 &stack_start); // 28 UINTX_FORMAT
1110 }
1111
1112 #undef _UFM
1113 #undef _DFM
1114
1115 if (i != 28 - 2) {
1116 assert(false, "Bad conversion from /proc/self/stat");
1117 // product mode - assume we are the primordial thread, good luck in the
1118 // embedded case.
1119 warning("Can't detect primordial thread stack location - bad conversion");
1120 stack_start = (uintptr_t) &rlim;
1121 }
1122 } else {
1123 // For some reason we can't open /proc/self/stat (for example, running on
1124 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1125 // most cases, so don't abort:
1126 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1127 stack_start = (uintptr_t) &rlim;
1128 }
1129 }
1130
1131 // Now we have a pointer (stack_start) very close to the stack top, the
1132 // next thing to do is to figure out the exact location of stack top. We
1133 // can find out the virtual memory area that contains stack_start by
1134 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1135 // and its upper limit is the real stack top. (again, this would fail if
1136 // running inside chroot, because /proc may not exist.)
1137
1138 uintptr_t stack_top;
1139 address low, high;
1140 if (find_vma((address)stack_start, &low, &high)) {
1141 // success, "high" is the true stack top. (ignore "low", because initial
1142 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1143 stack_top = (uintptr_t)high;
1144 } else {
1145 // failed, likely because /proc/self/maps does not exist
1146 warning("Can't detect primordial thread stack location - find_vma failed");
1147 // best effort: stack_start is normally within a few pages below the real
1148 // stack top, use it as stack top, and reduce stack size so we won't put
1149 // guard page outside stack.
1150 stack_top = stack_start;
1151 stack_size -= 16 * page_size();
1152 }
1153
1154 // stack_top could be partially down the page so align it
1155 stack_top = align_up(stack_top, page_size());
1156
1157 // Allowed stack value is minimum of max_size and what we derived from rlimit
1158 if (max_size > 0) {
1159 _initial_thread_stack_size = MIN2(max_size, stack_size);
1160 } else {
1161 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1162 // clamp it at 8MB as we do on Solaris
1163 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1164 }
1165 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1166 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1167
1168 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1169
1170 if (log_is_enabled(Info, os, thread)) {
1171 // See if we seem to be on primordial process thread
1172 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1173 uintptr_t(&rlim) < stack_top;
1174
1175 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1176 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1177 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1178 stack_top, intptr_t(_initial_thread_stack_bottom));
1179 }
1180 }
1181
1182 ////////////////////////////////////////////////////////////////////////////////
1183 // time support
1184
1185 #ifndef SUPPORTS_CLOCK_MONOTONIC
1186 #error "Build platform doesn't support clock_gettime and related functionality"
1187 #endif
1188
1189 // Time since start-up in seconds to a fine granularity.
1190 // Used by VMSelfDestructTimer and the MemProfiler.
1191 double os::elapsedTime() {
1192
1193 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1194 }
1195
1196 jlong os::elapsed_counter() {
1197 return javaTimeNanos() - initial_time_count;
1198 }
1199
1200 jlong os::elapsed_frequency() {
1201 return NANOSECS_PER_SEC; // nanosecond resolution
1202 }
1203
1204 bool os::supports_vtime() { return true; }
1205 bool os::enable_vtime() { return false; }
1206 bool os::vtime_enabled() { return false; }
1207
1208 double os::elapsedVTime() {
1209 struct rusage usage;
1210 int retval = getrusage(RUSAGE_THREAD, &usage);
1211 if (retval == 0) {
1212 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);
1213 } else {
1214 // better than nothing, but not much
1215 return elapsedTime();
1216 }
1217 }
1218
1219 jlong os::javaTimeMillis() {
1220 timeval time;
1221 int status = gettimeofday(&time, NULL);
1222 assert(status != -1, "linux error");
1223 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1224 }
1225
1226 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1227 timeval time;
1228 int status = gettimeofday(&time, NULL);
1229 assert(status != -1, "linux error");
1230 seconds = jlong(time.tv_sec);
1231 nanos = jlong(time.tv_usec) * 1000;
1232 }
1233
1234 void os::Linux::fast_thread_clock_init() {
1235 if (!UseLinuxPosixThreadCPUClocks) {
1236 return;
1237 }
1238 clockid_t clockid;
1239 struct timespec tp;
1240 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1241 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1242
1243 // Switch to using fast clocks for thread cpu time if
1244 // the clock_getres() returns 0 error code.
1245 // Note, that some kernels may support the current thread
1246 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1247 // returned by the pthread_getcpuclockid().
1248 // If the fast Posix clocks are supported then the clock_getres()
1249 // must return at least tp.tv_sec == 0 which means a resolution
1250 // better than 1 sec. This is extra check for reliability.
1251
1252 if (pthread_getcpuclockid_func &&
1253 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1254 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1255 _supports_fast_thread_cpu_time = true;
1256 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1257 }
1258 }
1259
1260 jlong os::javaTimeNanos() {
1261 if (os::supports_monotonic_clock()) {
1262 struct timespec tp;
1263 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1264 assert(status == 0, "gettime error");
1265 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1266 return result;
1267 } else {
1268 timeval time;
1269 int status = gettimeofday(&time, NULL);
1270 assert(status != -1, "linux error");
1271 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1272 return 1000 * usecs;
1273 }
1274 }
1275
1276 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1277 if (os::supports_monotonic_clock()) {
1278 info_ptr->max_value = ALL_64_BITS;
1279
1280 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1281 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1282 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1283 } else {
1284 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1285 info_ptr->max_value = ALL_64_BITS;
1286
1287 // gettimeofday is a real time clock so it skips
1288 info_ptr->may_skip_backward = true;
1289 info_ptr->may_skip_forward = true;
1290 }
1291
1292 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1293 }
1294
1295 // Return the real, user, and system times in seconds from an
1296 // arbitrary fixed point in the past.
1297 bool os::getTimesSecs(double* process_real_time,
1298 double* process_user_time,
1299 double* process_system_time) {
1300 struct tms ticks;
1301 clock_t real_ticks = times(&ticks);
1302
1303 if (real_ticks == (clock_t) (-1)) {
1304 return false;
1305 } else {
1306 double ticks_per_second = (double) clock_tics_per_sec;
1307 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1308 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1309 *process_real_time = ((double) real_ticks) / ticks_per_second;
1310
1311 return true;
1312 }
1313 }
1314
1315
1316 char * os::local_time_string(char *buf, size_t buflen) {
1317 struct tm t;
1318 time_t long_time;
1319 time(&long_time);
1320 localtime_r(&long_time, &t);
1321 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1322 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1323 t.tm_hour, t.tm_min, t.tm_sec);
1324 return buf;
1325 }
1326
1327 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1328 return localtime_r(clock, res);
1329 }
1330
1331 ////////////////////////////////////////////////////////////////////////////////
1332 // runtime exit support
1333
1334 // Note: os::shutdown() might be called very early during initialization, or
1335 // called from signal handler. Before adding something to os::shutdown(), make
1336 // sure it is async-safe and can handle partially initialized VM.
1337 void os::shutdown() {
1338
1339 // allow PerfMemory to attempt cleanup of any persistent resources
1340 perfMemory_exit();
1341
1342 // needs to remove object in file system
1343 AttachListener::abort();
1344
1345 // flush buffered output, finish log files
1346 ostream_abort();
1347
1348 // Check for abort hook
1349 abort_hook_t abort_hook = Arguments::abort_hook();
1350 if (abort_hook != NULL) {
1351 abort_hook();
1352 }
1353
1354 }
1355
1356 // Note: os::abort() might be called very early during initialization, or
1357 // called from signal handler. Before adding something to os::abort(), make
1358 // sure it is async-safe and can handle partially initialized VM.
1359 void os::abort(bool dump_core, void* siginfo, const void* context) {
1360 os::shutdown();
1361 if (dump_core) {
1362 if (DumpPrivateMappingsInCore) {
1363 ClassLoader::close_jrt_image();
1364 }
1365 #ifndef PRODUCT
1366 fdStream out(defaultStream::output_fd());
1367 out.print_raw("Current thread is ");
1368 char buf[16];
1369 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1370 out.print_raw_cr(buf);
1371 out.print_raw_cr("Dumping core ...");
1372 #endif
1373 ::abort(); // dump core
1374 }
1375
1376 ::exit(1);
1377 }
1378
1379 // Die immediately, no exit hook, no abort hook, no cleanup.
1380 void os::die() {
1381 ::abort();
1382 }
1383
1384 // thread_id is kernel thread id (similar to Solaris LWP id)
1385 intx os::current_thread_id() { return os::Linux::gettid(); }
1386 int os::current_process_id() {
1387 return ::getpid();
1388 }
1389
1390 // DLL functions
1391
1392 const char* os::dll_file_extension() { return ".so"; }
1393
1394 // This must be hard coded because it's the system's temporary
1395 // directory not the java application's temp directory, ala java.io.tmpdir.
1396 const char* os::get_temp_directory() { return "/tmp"; }
1397
1398 static bool file_exists(const char* filename) {
1399 struct stat statbuf;
1400 if (filename == NULL || strlen(filename) == 0) {
1401 return false;
1402 }
1403 return os::stat(filename, &statbuf) == 0;
1404 }
1405
1406 // check if addr is inside libjvm.so
1407 bool os::address_is_in_vm(address addr) {
1408 static address libjvm_base_addr;
1409 Dl_info dlinfo;
1410
1411 if (libjvm_base_addr == NULL) {
1412 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1413 libjvm_base_addr = (address)dlinfo.dli_fbase;
1414 }
1415 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1416 }
1417
1418 if (dladdr((void *)addr, &dlinfo) != 0) {
1419 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1420 }
1421
1422 return false;
1423 }
1424
1425 bool os::dll_address_to_function_name(address addr, char *buf,
1426 int buflen, int *offset,
1427 bool demangle) {
1428 // buf is not optional, but offset is optional
1429 assert(buf != NULL, "sanity check");
1430
1431 Dl_info dlinfo;
1432
1433 if (dladdr((void*)addr, &dlinfo) != 0) {
1434 // see if we have a matching symbol
1435 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1436 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1437 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1438 }
1439 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1440 return true;
1441 }
1442 // no matching symbol so try for just file info
1443 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1444 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1445 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1446 return true;
1447 }
1448 }
1449 }
1450
1451 buf[0] = '\0';
1452 if (offset != NULL) *offset = -1;
1453 return false;
1454 }
1455
1456 struct _address_to_library_name {
1457 address addr; // input : memory address
1458 size_t buflen; // size of fname
1459 char* fname; // output: library name
1460 address base; // library base addr
1461 };
1462
1463 static int address_to_library_name_callback(struct dl_phdr_info *info,
1464 size_t size, void *data) {
1465 int i;
1466 bool found = false;
1467 address libbase = NULL;
1468 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1469
1470 // iterate through all loadable segments
1471 for (i = 0; i < info->dlpi_phnum; i++) {
1472 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1473 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1474 // base address of a library is the lowest address of its loaded
1475 // segments.
1476 if (libbase == NULL || libbase > segbase) {
1477 libbase = segbase;
1478 }
1479 // see if 'addr' is within current segment
1480 if (segbase <= d->addr &&
1481 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1482 found = true;
1483 }
1484 }
1485 }
1486
1487 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1488 // so dll_address_to_library_name() can fall through to use dladdr() which
1489 // can figure out executable name from argv[0].
1490 if (found && info->dlpi_name && info->dlpi_name[0]) {
1491 d->base = libbase;
1492 if (d->fname) {
1493 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1494 }
1495 return 1;
1496 }
1497 return 0;
1498 }
1499
1500 bool os::dll_address_to_library_name(address addr, char* buf,
1501 int buflen, int* offset) {
1502 // buf is not optional, but offset is optional
1503 assert(buf != NULL, "sanity check");
1504
1505 Dl_info dlinfo;
1506 struct _address_to_library_name data;
1507
1508 // There is a bug in old glibc dladdr() implementation that it could resolve
1509 // to wrong library name if the .so file has a base address != NULL. Here
1510 // we iterate through the program headers of all loaded libraries to find
1511 // out which library 'addr' really belongs to. This workaround can be
1512 // removed once the minimum requirement for glibc is moved to 2.3.x.
1513 data.addr = addr;
1514 data.fname = buf;
1515 data.buflen = buflen;
1516 data.base = NULL;
1517 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1518
1519 if (rslt) {
1520 // buf already contains library name
1521 if (offset) *offset = addr - data.base;
1522 return true;
1523 }
1524 if (dladdr((void*)addr, &dlinfo) != 0) {
1525 if (dlinfo.dli_fname != NULL) {
1526 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1527 }
1528 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1529 *offset = addr - (address)dlinfo.dli_fbase;
1530 }
1531 return true;
1532 }
1533
1534 buf[0] = '\0';
1535 if (offset) *offset = -1;
1536 return false;
1537 }
1538
1539 // Loads .dll/.so and
1540 // in case of error it checks if .dll/.so was built for the
1541 // same architecture as Hotspot is running on
1542
1543
1544 // Remember the stack's state. The Linux dynamic linker will change
1545 // the stack to 'executable' at most once, so we must safepoint only once.
1546 bool os::Linux::_stack_is_executable = false;
1547
1548 // VM operation that loads a library. This is necessary if stack protection
1549 // of the Java stacks can be lost during loading the library. If we
1550 // do not stop the Java threads, they can stack overflow before the stacks
1551 // are protected again.
1552 class VM_LinuxDllLoad: public VM_Operation {
1553 private:
1554 const char *_filename;
1555 char *_ebuf;
1556 int _ebuflen;
1557 void *_lib;
1558 public:
1559 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1560 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1561 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1562 void doit() {
1563 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1564 os::Linux::_stack_is_executable = true;
1565 }
1566 void* loaded_library() { return _lib; }
1567 };
1568
1569 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1570 void * result = NULL;
1571 bool load_attempted = false;
1572
1573 // Check whether the library to load might change execution rights
1574 // of the stack. If they are changed, the protection of the stack
1575 // guard pages will be lost. We need a safepoint to fix this.
1576 //
1577 // See Linux man page execstack(8) for more info.
1578 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1579 if (!ElfFile::specifies_noexecstack(filename)) {
1580 if (!is_init_completed()) {
1581 os::Linux::_stack_is_executable = true;
1582 // This is OK - No Java threads have been created yet, and hence no
1583 // stack guard pages to fix.
1584 //
1585 // Dynamic loader will make all stacks executable after
1586 // this function returns, and will not do that again.
1587 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1588 } else {
1589 warning("You have loaded library %s which might have disabled stack guard. "
1590 "The VM will try to fix the stack guard now.\n"
1591 "It's highly recommended that you fix the library with "
1592 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1593 filename);
1594
1595 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1596 JavaThread *jt = JavaThread::current();
1597 if (jt->thread_state() != _thread_in_native) {
1598 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1599 // that requires ExecStack. Cannot enter safe point. Let's give up.
1600 warning("Unable to fix stack guard. Giving up.");
1601 } else {
1602 if (!LoadExecStackDllInVMThread) {
1603 // This is for the case where the DLL has an static
1604 // constructor function that executes JNI code. We cannot
1605 // load such DLLs in the VMThread.
1606 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1607 }
1608
1609 ThreadInVMfromNative tiv(jt);
1610 debug_only(VMNativeEntryWrapper vew;)
1611
1612 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1613 VMThread::execute(&op);
1614 if (LoadExecStackDllInVMThread) {
1615 result = op.loaded_library();
1616 }
1617 load_attempted = true;
1618 }
1619 }
1620 }
1621 }
1622
1623 if (!load_attempted) {
1624 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1625 }
1626
1627 if (result != NULL) {
1628 // Successful loading
1629 return result;
1630 }
1631
1632 Elf32_Ehdr elf_head;
1633 int diag_msg_max_length=ebuflen-strlen(ebuf);
1634 char* diag_msg_buf=ebuf+strlen(ebuf);
1635
1636 if (diag_msg_max_length==0) {
1637 // No more space in ebuf for additional diagnostics message
1638 return NULL;
1639 }
1640
1641
1642 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1643
1644 if (file_descriptor < 0) {
1645 // Can't open library, report dlerror() message
1646 return NULL;
1647 }
1648
1649 bool failed_to_read_elf_head=
1650 (sizeof(elf_head)!=
1651 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1652
1653 ::close(file_descriptor);
1654 if (failed_to_read_elf_head) {
1655 // file i/o error - report dlerror() msg
1656 return NULL;
1657 }
1658
1659 typedef struct {
1660 Elf32_Half code; // Actual value as defined in elf.h
1661 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1662 unsigned char elf_class; // 32 or 64 bit
1663 unsigned char endianess; // MSB or LSB
1664 char* name; // String representation
1665 } arch_t;
1666
1667 #ifndef EM_486
1668 #define EM_486 6 /* Intel 80486 */
1669 #endif
1670 #ifndef EM_AARCH64
1671 #define EM_AARCH64 183 /* ARM AARCH64 */
1672 #endif
1673
1674 static const arch_t arch_array[]={
1675 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1676 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1677 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1678 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1679 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1680 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1681 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1682 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1683 #if defined(VM_LITTLE_ENDIAN)
1684 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1685 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1686 #else
1687 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1688 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1689 #endif
1690 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1691 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1692 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1693 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1694 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1695 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1696 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1697 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1698 };
1699
1700 #if (defined IA32)
1701 static Elf32_Half running_arch_code=EM_386;
1702 #elif (defined AMD64) || (defined X32)
1703 static Elf32_Half running_arch_code=EM_X86_64;
1704 #elif (defined IA64)
1705 static Elf32_Half running_arch_code=EM_IA_64;
1706 #elif (defined __sparc) && (defined _LP64)
1707 static Elf32_Half running_arch_code=EM_SPARCV9;
1708 #elif (defined __sparc) && (!defined _LP64)
1709 static Elf32_Half running_arch_code=EM_SPARC;
1710 #elif (defined __powerpc64__)
1711 static Elf32_Half running_arch_code=EM_PPC64;
1712 #elif (defined __powerpc__)
1713 static Elf32_Half running_arch_code=EM_PPC;
1714 #elif (defined AARCH64)
1715 static Elf32_Half running_arch_code=EM_AARCH64;
1716 #elif (defined ARM)
1717 static Elf32_Half running_arch_code=EM_ARM;
1718 #elif (defined S390)
1719 static Elf32_Half running_arch_code=EM_S390;
1720 #elif (defined ALPHA)
1721 static Elf32_Half running_arch_code=EM_ALPHA;
1722 #elif (defined MIPSEL)
1723 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1724 #elif (defined PARISC)
1725 static Elf32_Half running_arch_code=EM_PARISC;
1726 #elif (defined MIPS)
1727 static Elf32_Half running_arch_code=EM_MIPS;
1728 #elif (defined M68K)
1729 static Elf32_Half running_arch_code=EM_68K;
1730 #elif (defined SH)
1731 static Elf32_Half running_arch_code=EM_SH;
1732 #else
1733 #error Method os::dll_load requires that one of following is defined:\
1734 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1735 #endif
1736
1737 // Identify compatability class for VM's architecture and library's architecture
1738 // Obtain string descriptions for architectures
1739
1740 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1741 int running_arch_index=-1;
1742
1743 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1744 if (running_arch_code == arch_array[i].code) {
1745 running_arch_index = i;
1746 }
1747 if (lib_arch.code == arch_array[i].code) {
1748 lib_arch.compat_class = arch_array[i].compat_class;
1749 lib_arch.name = arch_array[i].name;
1750 }
1751 }
1752
1753 assert(running_arch_index != -1,
1754 "Didn't find running architecture code (running_arch_code) in arch_array");
1755 if (running_arch_index == -1) {
1756 // Even though running architecture detection failed
1757 // we may still continue with reporting dlerror() message
1758 return NULL;
1759 }
1760
1761 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1762 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1763 return NULL;
1764 }
1765
1766 #ifndef S390
1767 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1768 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1769 return NULL;
1770 }
1771 #endif // !S390
1772
1773 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1774 if (lib_arch.name!=NULL) {
1775 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1776 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1777 lib_arch.name, arch_array[running_arch_index].name);
1778 } else {
1779 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1780 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1781 lib_arch.code,
1782 arch_array[running_arch_index].name);
1783 }
1784 }
1785
1786 return NULL;
1787 }
1788
1789 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1790 int ebuflen) {
1791 void * result = ::dlopen(filename, RTLD_LAZY);
1792 if (result == NULL) {
1793 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1794 ebuf[ebuflen-1] = '\0';
1795 }
1796 return result;
1797 }
1798
1799 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1800 int ebuflen) {
1801 void * result = NULL;
1802 if (LoadExecStackDllInVMThread) {
1803 result = dlopen_helper(filename, ebuf, ebuflen);
1804 }
1805
1806 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1807 // library that requires an executable stack, or which does not have this
1808 // stack attribute set, dlopen changes the stack attribute to executable. The
1809 // read protection of the guard pages gets lost.
1810 //
1811 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1812 // may have been queued at the same time.
1813
1814 if (!_stack_is_executable) {
1815 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1816 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1817 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1818 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1819 warning("Attempt to reguard stack yellow zone failed.");
1820 }
1821 }
1822 }
1823 }
1824
1825 return result;
1826 }
1827
1828 void* os::dll_lookup(void* handle, const char* name) {
1829 void* res = dlsym(handle, name);
1830 return res;
1831 }
1832
1833 void* os::get_default_process_handle() {
1834 return (void*)::dlopen(NULL, RTLD_LAZY);
1835 }
1836
1837 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1838 int fd = ::open(filename, O_RDONLY);
1839 if (fd == -1) {
1840 return false;
1841 }
1842
1843 if (hdr != NULL) {
1844 st->print_cr("%s", hdr);
1845 }
1846
1847 char buf[33];
1848 int bytes;
1849 buf[32] = '\0';
1850 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1851 st->print_raw(buf, bytes);
1852 }
1853
1854 ::close(fd);
1855
1856 return true;
1857 }
1858
1859 #if defined(S390)
1860 // keywords_to_match - NULL terminated array of keywords
1861 static bool print_matching_lines_from_sysinfo_file(outputStream* st, const char* keywords_to_match[]) {
1862 const char* filename = "/proc/sysinfo";
1863 char* line = NULL;
1864 size_t length = 0;
1865 FILE* fp = fopen(filename, "r");
1866 if (fp == NULL) {
1867 return false;
1868 }
1869
1870 st->print_cr("Virtualization information:");
1871 while (getline(&line, &length, fp) != -1) {
1872 int i = 0;
1873 while (keywords_to_match[i] != NULL) {
1874 if (strncmp(line, keywords_to_match[i], strlen(keywords_to_match[i])) == 0) {
1875 st->print("%s", line);
1876 break;
1877 }
1878 i++;
1879 }
1880 }
1881
1882 free(line);
1883 fclose(fp);
1884
1885 return true;
1886 }
1887 #endif
1888
1889 void os::print_dll_info(outputStream *st) {
1890 st->print_cr("Dynamic libraries:");
1891
1892 char fname[32];
1893 pid_t pid = os::Linux::gettid();
1894
1895 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1896
1897 if (!_print_ascii_file(fname, st)) {
1898 st->print("Can not get library information for pid = %d\n", pid);
1899 }
1900 }
1901
1902 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1903 FILE *procmapsFile = NULL;
1904
1905 // Open the procfs maps file for the current process
1906 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1907 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1908 char line[PATH_MAX + 100];
1909
1910 // Read line by line from 'file'
1911 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1912 u8 base, top, offset, inode;
1913 char permissions[5];
1914 char device[6];
1915 char name[PATH_MAX + 1];
1916
1917 // Parse fields from line
1918 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
1919 &base, &top, permissions, &offset, device, &inode, name);
1920
1921 // Filter by device id '00:00' so that we only get file system mapped files.
1922 if (strcmp(device, "00:00") != 0) {
1923
1924 // Call callback with the fields of interest
1925 if(callback(name, (address)base, (address)top, param)) {
1926 // Oops abort, callback aborted
1927 fclose(procmapsFile);
1928 return 1;
1929 }
1930 }
1931 }
1932 fclose(procmapsFile);
1933 }
1934 return 0;
1935 }
1936
1937 void os::print_os_info_brief(outputStream* st) {
1938 os::Linux::print_distro_info(st);
1939
1940 os::Posix::print_uname_info(st);
1941
1942 os::Linux::print_libversion_info(st);
1943
1944 }
1945
1946 void os::print_os_info(outputStream* st) {
1947 st->print("OS:");
1948
1949 os::Linux::print_distro_info(st);
1950
1951 os::Posix::print_uname_info(st);
1952
1953 // Print warning if unsafe chroot environment detected
1954 if (unsafe_chroot_detected) {
1955 st->print("WARNING!! ");
1956 st->print_cr("%s", unstable_chroot_error);
1957 }
1958
1959 os::Linux::print_libversion_info(st);
1960
1961 os::Posix::print_rlimit_info(st);
1962
1963 os::Posix::print_load_average(st);
1964
1965 os::Linux::print_full_memory_info(st);
1966
1967 os::Linux::print_proc_sys_info(st);
1968
1969 os::Linux::print_ld_preload_file(st);
1970
1971 os::Linux::print_container_info(st);
1972
1973 os::Linux::print_virtualization_info(st);
1974 }
1975
1976 // Try to identify popular distros.
1977 // Most Linux distributions have a /etc/XXX-release file, which contains
1978 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1979 // file that also contains the OS version string. Some have more than one
1980 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1981 // /etc/redhat-release.), so the order is important.
1982 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1983 // their own specific XXX-release file as well as a redhat-release file.
1984 // Because of this the XXX-release file needs to be searched for before the
1985 // redhat-release file.
1986 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1987 // search for redhat-release / SuSE-release needs to be before lsb-release.
1988 // Since the lsb-release file is the new standard it needs to be searched
1989 // before the older style release files.
1990 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1991 // next to last resort. The os-release file is a new standard that contains
1992 // distribution information and the system-release file seems to be an old
1993 // standard that has been replaced by the lsb-release and os-release files.
1994 // Searching for the debian_version file is the last resort. It contains
1995 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1996 // "Debian " is printed before the contents of the debian_version file.
1997
1998 const char* distro_files[] = {
1999 "/etc/oracle-release",
2000 "/etc/mandriva-release",
2001 "/etc/mandrake-release",
2002 "/etc/sun-release",
2003 "/etc/redhat-release",
2004 "/etc/SuSE-release",
2005 "/etc/lsb-release",
2006 "/etc/turbolinux-release",
2007 "/etc/gentoo-release",
2008 "/etc/ltib-release",
2009 "/etc/angstrom-version",
2010 "/etc/system-release",
2011 "/etc/os-release",
2012 NULL };
2013
2014 void os::Linux::print_distro_info(outputStream* st) {
2015 for (int i = 0;; i++) {
2016 const char* file = distro_files[i];
2017 if (file == NULL) {
2018 break; // done
2019 }
2020 // If file prints, we found it.
2021 if (_print_ascii_file(file, st)) {
2022 return;
2023 }
2024 }
2025
2026 if (file_exists("/etc/debian_version")) {
2027 st->print("Debian ");
2028 _print_ascii_file("/etc/debian_version", st);
2029 } else {
2030 st->print("Linux");
2031 }
2032 st->cr();
2033 }
2034
2035 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2036 char buf[256];
2037 while (fgets(buf, sizeof(buf), fp)) {
2038 // Edit out extra stuff in expected format
2039 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2040 char* ptr = strstr(buf, "\""); // the name is in quotes
2041 if (ptr != NULL) {
2042 ptr++; // go beyond first quote
2043 char* nl = strchr(ptr, '\"');
2044 if (nl != NULL) *nl = '\0';
2045 strncpy(distro, ptr, length);
2046 } else {
2047 ptr = strstr(buf, "=");
2048 ptr++; // go beyond equals then
2049 char* nl = strchr(ptr, '\n');
2050 if (nl != NULL) *nl = '\0';
2051 strncpy(distro, ptr, length);
2052 }
2053 return;
2054 } else if (get_first_line) {
2055 char* nl = strchr(buf, '\n');
2056 if (nl != NULL) *nl = '\0';
2057 strncpy(distro, buf, length);
2058 return;
2059 }
2060 }
2061 // print last line and close
2062 char* nl = strchr(buf, '\n');
2063 if (nl != NULL) *nl = '\0';
2064 strncpy(distro, buf, length);
2065 }
2066
2067 static void parse_os_info(char* distro, size_t length, const char* file) {
2068 FILE* fp = fopen(file, "r");
2069 if (fp != NULL) {
2070 // if suse format, print out first line
2071 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2072 parse_os_info_helper(fp, distro, length, get_first_line);
2073 fclose(fp);
2074 }
2075 }
2076
2077 void os::get_summary_os_info(char* buf, size_t buflen) {
2078 for (int i = 0;; i++) {
2079 const char* file = distro_files[i];
2080 if (file == NULL) {
2081 break; // ran out of distro_files
2082 }
2083 if (file_exists(file)) {
2084 parse_os_info(buf, buflen, file);
2085 return;
2086 }
2087 }
2088 // special case for debian
2089 if (file_exists("/etc/debian_version")) {
2090 strncpy(buf, "Debian ", buflen);
2091 if (buflen > 7) {
2092 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2093 }
2094 } else {
2095 strncpy(buf, "Linux", buflen);
2096 }
2097 }
2098
2099 void os::Linux::print_libversion_info(outputStream* st) {
2100 // libc, pthread
2101 st->print("libc:");
2102 st->print("%s ", os::Linux::glibc_version());
2103 st->print("%s ", os::Linux::libpthread_version());
2104 st->cr();
2105 }
2106
2107 void os::Linux::print_proc_sys_info(outputStream* st) {
2108 st->cr();
2109 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2110 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2111 st->cr();
2112 st->cr();
2113
2114 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2115 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2116 st->cr();
2117 st->cr();
2118
2119 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2120 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2121 st->cr();
2122 st->cr();
2123 }
2124
2125 void os::Linux::print_full_memory_info(outputStream* st) {
2126 st->print("\n/proc/meminfo:\n");
2127 _print_ascii_file("/proc/meminfo", st);
2128 st->cr();
2129 }
2130
2131 void os::Linux::print_ld_preload_file(outputStream* st) {
2132 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2133 st->cr();
2134 }
2135
2136 void os::Linux::print_container_info(outputStream* st) {
2137 if (!OSContainer::is_containerized()) {
2138 return;
2139 }
2140
2141 st->print("container (cgroup) information:\n");
2142
2143 const char *p_ct = OSContainer::container_type();
2144 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2145
2146 char *p = OSContainer::cpu_cpuset_cpus();
2147 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2148 free(p);
2149
2150 p = OSContainer::cpu_cpuset_memory_nodes();
2151 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2152 free(p);
2153
2154 int i = OSContainer::active_processor_count();
2155 if (i > 0) {
2156 st->print("active_processor_count: %d\n", i);
2157 } else {
2158 st->print("active_processor_count: failed\n");
2159 }
2160
2161 i = OSContainer::cpu_quota();
2162 st->print("cpu_quota: %d\n", i);
2163
2164 i = OSContainer::cpu_period();
2165 st->print("cpu_period: %d\n", i);
2166
2167 i = OSContainer::cpu_shares();
2168 st->print("cpu_shares: %d\n", i);
2169
2170 jlong j = OSContainer::memory_limit_in_bytes();
2171 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2172
2173 j = OSContainer::memory_and_swap_limit_in_bytes();
2174 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2175
2176 j = OSContainer::memory_soft_limit_in_bytes();
2177 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2178
2179 j = OSContainer::OSContainer::memory_usage_in_bytes();
2180 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2181
2182 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2183 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2184 st->cr();
2185 }
2186
2187 void os::Linux::print_virtualization_info(outputStream* st) {
2188 #if defined(S390)
2189 // /proc/sysinfo contains interesting information about
2190 // - LPAR
2191 // - whole "Box" (CPUs )
2192 // - z/VM / KVM (VM<nn>); this is not available in an LPAR-only setup
2193 const char* kw[] = { "LPAR", "CPUs", "VM", NULL };
2194
2195 if (! print_matching_lines_from_sysinfo_file(st, kw)) {
2196 st->print_cr(" </proc/sysinfo Not Available>");
2197 }
2198 #endif
2199 }
2200
2201 void os::print_memory_info(outputStream* st) {
2202
2203 st->print("Memory:");
2204 st->print(" %dk page", os::vm_page_size()>>10);
2205
2206 // values in struct sysinfo are "unsigned long"
2207 struct sysinfo si;
2208 sysinfo(&si);
2209
2210 st->print(", physical " UINT64_FORMAT "k",
2211 os::physical_memory() >> 10);
2212 st->print("(" UINT64_FORMAT "k free)",
2213 os::available_memory() >> 10);
2214 st->print(", swap " UINT64_FORMAT "k",
2215 ((jlong)si.totalswap * si.mem_unit) >> 10);
2216 st->print("(" UINT64_FORMAT "k free)",
2217 ((jlong)si.freeswap * si.mem_unit) >> 10);
2218 st->cr();
2219 }
2220
2221 // Print the first "model name" line and the first "flags" line
2222 // that we find and nothing more. We assume "model name" comes
2223 // before "flags" so if we find a second "model name", then the
2224 // "flags" field is considered missing.
2225 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2226 #if defined(IA32) || defined(AMD64)
2227 // Other platforms have less repetitive cpuinfo files
2228 FILE *fp = fopen("/proc/cpuinfo", "r");
2229 if (fp) {
2230 while (!feof(fp)) {
2231 if (fgets(buf, buflen, fp)) {
2232 // Assume model name comes before flags
2233 bool model_name_printed = false;
2234 if (strstr(buf, "model name") != NULL) {
2235 if (!model_name_printed) {
2236 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2237 st->print_raw(buf);
2238 model_name_printed = true;
2239 } else {
2240 // model name printed but not flags? Odd, just return
2241 fclose(fp);
2242 return true;
2243 }
2244 }
2245 // print the flags line too
2246 if (strstr(buf, "flags") != NULL) {
2247 st->print_raw(buf);
2248 fclose(fp);
2249 return true;
2250 }
2251 }
2252 }
2253 fclose(fp);
2254 }
2255 #endif // x86 platforms
2256 return false;
2257 }
2258
2259 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2260 // Only print the model name if the platform provides this as a summary
2261 if (!print_model_name_and_flags(st, buf, buflen)) {
2262 st->print("\n/proc/cpuinfo:\n");
2263 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2264 st->print_cr(" <Not Available>");
2265 }
2266 }
2267 }
2268
2269 #if defined(AMD64) || defined(IA32) || defined(X32)
2270 const char* search_string = "model name";
2271 #elif defined(M68K)
2272 const char* search_string = "CPU";
2273 #elif defined(PPC64)
2274 const char* search_string = "cpu";
2275 #elif defined(S390)
2276 const char* search_string = "processor";
2277 #elif defined(SPARC)
2278 const char* search_string = "cpu";
2279 #else
2280 const char* search_string = "Processor";
2281 #endif
2282
2283 // Parses the cpuinfo file for string representing the model name.
2284 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2285 FILE* fp = fopen("/proc/cpuinfo", "r");
2286 if (fp != NULL) {
2287 while (!feof(fp)) {
2288 char buf[256];
2289 if (fgets(buf, sizeof(buf), fp)) {
2290 char* start = strstr(buf, search_string);
2291 if (start != NULL) {
2292 char *ptr = start + strlen(search_string);
2293 char *end = buf + strlen(buf);
2294 while (ptr != end) {
2295 // skip whitespace and colon for the rest of the name.
2296 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2297 break;
2298 }
2299 ptr++;
2300 }
2301 if (ptr != end) {
2302 // reasonable string, get rid of newline and keep the rest
2303 char* nl = strchr(buf, '\n');
2304 if (nl != NULL) *nl = '\0';
2305 strncpy(cpuinfo, ptr, length);
2306 fclose(fp);
2307 return;
2308 }
2309 }
2310 }
2311 }
2312 fclose(fp);
2313 }
2314 // cpuinfo not found or parsing failed, just print generic string. The entire
2315 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2316 #if defined(AARCH64)
2317 strncpy(cpuinfo, "AArch64", length);
2318 #elif defined(AMD64)
2319 strncpy(cpuinfo, "x86_64", length);
2320 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2321 strncpy(cpuinfo, "ARM", length);
2322 #elif defined(IA32)
2323 strncpy(cpuinfo, "x86_32", length);
2324 #elif defined(IA64)
2325 strncpy(cpuinfo, "IA64", length);
2326 #elif defined(PPC)
2327 strncpy(cpuinfo, "PPC64", length);
2328 #elif defined(S390)
2329 strncpy(cpuinfo, "S390", length);
2330 #elif defined(SPARC)
2331 strncpy(cpuinfo, "sparcv9", length);
2332 #elif defined(ZERO_LIBARCH)
2333 strncpy(cpuinfo, ZERO_LIBARCH, length);
2334 #else
2335 strncpy(cpuinfo, "unknown", length);
2336 #endif
2337 }
2338
2339 static void print_signal_handler(outputStream* st, int sig,
2340 char* buf, size_t buflen);
2341
2342 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2343 st->print_cr("Signal Handlers:");
2344 print_signal_handler(st, SIGSEGV, buf, buflen);
2345 print_signal_handler(st, SIGBUS , buf, buflen);
2346 print_signal_handler(st, SIGFPE , buf, buflen);
2347 print_signal_handler(st, SIGPIPE, buf, buflen);
2348 print_signal_handler(st, SIGXFSZ, buf, buflen);
2349 print_signal_handler(st, SIGILL , buf, buflen);
2350 print_signal_handler(st, SR_signum, buf, buflen);
2351 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2352 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2353 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2354 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2355 #if defined(PPC64)
2356 print_signal_handler(st, SIGTRAP, buf, buflen);
2357 #endif
2358 }
2359
2360 static char saved_jvm_path[MAXPATHLEN] = {0};
2361
2362 // Find the full path to the current module, libjvm.so
2363 void os::jvm_path(char *buf, jint buflen) {
2364 // Error checking.
2365 if (buflen < MAXPATHLEN) {
2366 assert(false, "must use a large-enough buffer");
2367 buf[0] = '\0';
2368 return;
2369 }
2370 // Lazy resolve the path to current module.
2371 if (saved_jvm_path[0] != 0) {
2372 strcpy(buf, saved_jvm_path);
2373 return;
2374 }
2375
2376 char dli_fname[MAXPATHLEN];
2377 bool ret = dll_address_to_library_name(
2378 CAST_FROM_FN_PTR(address, os::jvm_path),
2379 dli_fname, sizeof(dli_fname), NULL);
2380 assert(ret, "cannot locate libjvm");
2381 char *rp = NULL;
2382 if (ret && dli_fname[0] != '\0') {
2383 rp = os::Posix::realpath(dli_fname, buf, buflen);
2384 }
2385 if (rp == NULL) {
2386 return;
2387 }
2388
2389 if (Arguments::sun_java_launcher_is_altjvm()) {
2390 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2391 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2392 // If "/jre/lib/" appears at the right place in the string, then
2393 // assume we are installed in a JDK and we're done. Otherwise, check
2394 // for a JAVA_HOME environment variable and fix up the path so it
2395 // looks like libjvm.so is installed there (append a fake suffix
2396 // hotspot/libjvm.so).
2397 const char *p = buf + strlen(buf) - 1;
2398 for (int count = 0; p > buf && count < 5; ++count) {
2399 for (--p; p > buf && *p != '/'; --p)
2400 /* empty */ ;
2401 }
2402
2403 if (strncmp(p, "/jre/lib/", 9) != 0) {
2404 // Look for JAVA_HOME in the environment.
2405 char* java_home_var = ::getenv("JAVA_HOME");
2406 if (java_home_var != NULL && java_home_var[0] != 0) {
2407 char* jrelib_p;
2408 int len;
2409
2410 // Check the current module name "libjvm.so".
2411 p = strrchr(buf, '/');
2412 if (p == NULL) {
2413 return;
2414 }
2415 assert(strstr(p, "/libjvm") == p, "invalid library name");
2416
2417 rp = os::Posix::realpath(java_home_var, buf, buflen);
2418 if (rp == NULL) {
2419 return;
2420 }
2421
2422 // determine if this is a legacy image or modules image
2423 // modules image doesn't have "jre" subdirectory
2424 len = strlen(buf);
2425 assert(len < buflen, "Ran out of buffer room");
2426 jrelib_p = buf + len;
2427 snprintf(jrelib_p, buflen-len, "/jre/lib");
2428 if (0 != access(buf, F_OK)) {
2429 snprintf(jrelib_p, buflen-len, "/lib");
2430 }
2431
2432 if (0 == access(buf, F_OK)) {
2433 // Use current module name "libjvm.so"
2434 len = strlen(buf);
2435 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2436 } else {
2437 // Go back to path of .so
2438 rp = os::Posix::realpath(dli_fname, buf, buflen);
2439 if (rp == NULL) {
2440 return;
2441 }
2442 }
2443 }
2444 }
2445 }
2446
2447 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2448 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2449 }
2450
2451 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2452 // no prefix required, not even "_"
2453 }
2454
2455 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2456 // no suffix required
2457 }
2458
2459 ////////////////////////////////////////////////////////////////////////////////
2460 // sun.misc.Signal support
2461
2462 static volatile jint sigint_count = 0;
2463
2464 static void UserHandler(int sig, void *siginfo, void *context) {
2465 // 4511530 - sem_post is serialized and handled by the manager thread. When
2466 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2467 // don't want to flood the manager thread with sem_post requests.
2468 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2469 return;
2470 }
2471
2472 // Ctrl-C is pressed during error reporting, likely because the error
2473 // handler fails to abort. Let VM die immediately.
2474 if (sig == SIGINT && VMError::is_error_reported()) {
2475 os::die();
2476 }
2477
2478 os::signal_notify(sig);
2479 }
2480
2481 void* os::user_handler() {
2482 return CAST_FROM_FN_PTR(void*, UserHandler);
2483 }
2484
2485 extern "C" {
2486 typedef void (*sa_handler_t)(int);
2487 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2488 }
2489
2490 void* os::signal(int signal_number, void* handler) {
2491 struct sigaction sigAct, oldSigAct;
2492
2493 sigfillset(&(sigAct.sa_mask));
2494 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2495 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2496
2497 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2498 // -1 means registration failed
2499 return (void *)-1;
2500 }
2501
2502 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2503 }
2504
2505 void os::signal_raise(int signal_number) {
2506 ::raise(signal_number);
2507 }
2508
2509 // The following code is moved from os.cpp for making this
2510 // code platform specific, which it is by its very nature.
2511
2512 // Will be modified when max signal is changed to be dynamic
2513 int os::sigexitnum_pd() {
2514 return NSIG;
2515 }
2516
2517 // a counter for each possible signal value
2518 static volatile jint pending_signals[NSIG+1] = { 0 };
2519
2520 // Linux(POSIX) specific hand shaking semaphore.
2521 static Semaphore* sig_sem = NULL;
2522 static PosixSemaphore sr_semaphore;
2523
2524 static void jdk_misc_signal_init() {
2525 // Initialize signal structures
2526 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2527
2528 // Initialize signal semaphore
2529 sig_sem = new Semaphore();
2530 }
2531
2532 void os::signal_notify(int sig) {
2533 if (sig_sem != NULL) {
2534 Atomic::inc(&pending_signals[sig]);
2535 sig_sem->signal();
2536 } else {
2537 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2538 // initialization isn't called.
2539 assert(ReduceSignalUsage, "signal semaphore should be created");
2540 }
2541 }
2542
2543 static int check_pending_signals() {
2544 Atomic::store(0, &sigint_count);
2545 for (;;) {
2546 for (int i = 0; i < NSIG + 1; i++) {
2547 jint n = pending_signals[i];
2548 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2549 return i;
2550 }
2551 }
2552 JavaThread *thread = JavaThread::current();
2553 ThreadBlockInVM tbivm(thread);
2554
2555 bool threadIsSuspended;
2556 do {
2557 thread->set_suspend_equivalent();
2558 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2559 sig_sem->wait();
2560
2561 // were we externally suspended while we were waiting?
2562 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2563 if (threadIsSuspended) {
2564 // The semaphore has been incremented, but while we were waiting
2565 // another thread suspended us. We don't want to continue running
2566 // while suspended because that would surprise the thread that
2567 // suspended us.
2568 sig_sem->signal();
2569
2570 thread->java_suspend_self();
2571 }
2572 } while (threadIsSuspended);
2573 }
2574 }
2575
2576 int os::signal_wait() {
2577 return check_pending_signals();
2578 }
2579
2580 ////////////////////////////////////////////////////////////////////////////////
2581 // Virtual Memory
2582
2583 int os::vm_page_size() {
2584 // Seems redundant as all get out
2585 assert(os::Linux::page_size() != -1, "must call os::init");
2586 return os::Linux::page_size();
2587 }
2588
2589 // Solaris allocates memory by pages.
2590 int os::vm_allocation_granularity() {
2591 assert(os::Linux::page_size() != -1, "must call os::init");
2592 return os::Linux::page_size();
2593 }
2594
2595 // Rationale behind this function:
2596 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2597 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2598 // samples for JITted code. Here we create private executable mapping over the code cache
2599 // and then we can use standard (well, almost, as mapping can change) way to provide
2600 // info for the reporting script by storing timestamp and location of symbol
2601 void linux_wrap_code(char* base, size_t size) {
2602 static volatile jint cnt = 0;
2603
2604 if (!UseOprofile) {
2605 return;
2606 }
2607
2608 char buf[PATH_MAX+1];
2609 int num = Atomic::add(1, &cnt);
2610
2611 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2612 os::get_temp_directory(), os::current_process_id(), num);
2613 unlink(buf);
2614
2615 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2616
2617 if (fd != -1) {
2618 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2619 if (rv != (off_t)-1) {
2620 if (::write(fd, "", 1) == 1) {
2621 mmap(base, size,
2622 PROT_READ|PROT_WRITE|PROT_EXEC,
2623 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2624 }
2625 }
2626 ::close(fd);
2627 unlink(buf);
2628 }
2629 }
2630
2631 static bool recoverable_mmap_error(int err) {
2632 // See if the error is one we can let the caller handle. This
2633 // list of errno values comes from JBS-6843484. I can't find a
2634 // Linux man page that documents this specific set of errno
2635 // values so while this list currently matches Solaris, it may
2636 // change as we gain experience with this failure mode.
2637 switch (err) {
2638 case EBADF:
2639 case EINVAL:
2640 case ENOTSUP:
2641 // let the caller deal with these errors
2642 return true;
2643
2644 default:
2645 // Any remaining errors on this OS can cause our reserved mapping
2646 // to be lost. That can cause confusion where different data
2647 // structures think they have the same memory mapped. The worst
2648 // scenario is if both the VM and a library think they have the
2649 // same memory mapped.
2650 return false;
2651 }
2652 }
2653
2654 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2655 int err) {
2656 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2657 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2658 os::strerror(err), err);
2659 }
2660
2661 static void warn_fail_commit_memory(char* addr, size_t size,
2662 size_t alignment_hint, bool exec,
2663 int err) {
2664 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2665 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2666 alignment_hint, exec, os::strerror(err), err);
2667 }
2668
2669 // NOTE: Linux kernel does not really reserve the pages for us.
2670 // All it does is to check if there are enough free pages
2671 // left at the time of mmap(). This could be a potential
2672 // problem.
2673 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2674 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2675 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2676 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2677 if (res != (uintptr_t) MAP_FAILED) {
2678 if (UseNUMAInterleaving) {
2679 numa_make_global(addr, size);
2680 }
2681 return 0;
2682 }
2683
2684 int err = errno; // save errno from mmap() call above
2685
2686 if (!recoverable_mmap_error(err)) {
2687 warn_fail_commit_memory(addr, size, exec, err);
2688 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2689 }
2690
2691 return err;
2692 }
2693
2694 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2695 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2696 }
2697
2698 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2699 const char* mesg) {
2700 assert(mesg != NULL, "mesg must be specified");
2701 int err = os::Linux::commit_memory_impl(addr, size, exec);
2702 if (err != 0) {
2703 // the caller wants all commit errors to exit with the specified mesg:
2704 warn_fail_commit_memory(addr, size, exec, err);
2705 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2706 }
2707 }
2708
2709 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2710 #ifndef MAP_HUGETLB
2711 #define MAP_HUGETLB 0x40000
2712 #endif
2713
2714 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2715 #ifndef MADV_HUGEPAGE
2716 #define MADV_HUGEPAGE 14
2717 #endif
2718
2719 int os::Linux::commit_memory_impl(char* addr, size_t size,
2720 size_t alignment_hint, bool exec) {
2721 int err = os::Linux::commit_memory_impl(addr, size, exec);
2722 if (err == 0) {
2723 realign_memory(addr, size, alignment_hint);
2724 }
2725 return err;
2726 }
2727
2728 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2729 bool exec) {
2730 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2731 }
2732
2733 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2734 size_t alignment_hint, bool exec,
2735 const char* mesg) {
2736 assert(mesg != NULL, "mesg must be specified");
2737 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2738 if (err != 0) {
2739 // the caller wants all commit errors to exit with the specified mesg:
2740 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2741 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2742 }
2743 }
2744
2745 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2746 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2747 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2748 // be supported or the memory may already be backed by huge pages.
2749 ::madvise(addr, bytes, MADV_HUGEPAGE);
2750 }
2751 }
2752
2753 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2754 // This method works by doing an mmap over an existing mmaping and effectively discarding
2755 // the existing pages. However it won't work for SHM-based large pages that cannot be
2756 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2757 // small pages on top of the SHM segment. This method always works for small pages, so we
2758 // allow that in any case.
2759 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2760 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2761 }
2762 }
2763
2764 void os::numa_make_global(char *addr, size_t bytes) {
2765 Linux::numa_interleave_memory(addr, bytes);
2766 }
2767
2768 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2769 // bind policy to MPOL_PREFERRED for the current thread.
2770 #define USE_MPOL_PREFERRED 0
2771
2772 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2773 // To make NUMA and large pages more robust when both enabled, we need to ease
2774 // the requirements on where the memory should be allocated. MPOL_BIND is the
2775 // default policy and it will force memory to be allocated on the specified
2776 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2777 // the specified node, but will not force it. Using this policy will prevent
2778 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2779 // free large pages.
2780 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2781 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2782 }
2783
2784 bool os::numa_topology_changed() { return false; }
2785
2786 size_t os::numa_get_groups_num() {
2787 // Return just the number of nodes in which it's possible to allocate memory
2788 // (in numa terminology, configured nodes).
2789 return Linux::numa_num_configured_nodes();
2790 }
2791
2792 int os::numa_get_group_id() {
2793 int cpu_id = Linux::sched_getcpu();
2794 if (cpu_id != -1) {
2795 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2796 if (lgrp_id != -1) {
2797 return lgrp_id;
2798 }
2799 }
2800 return 0;
2801 }
2802
2803 int os::Linux::get_existing_num_nodes() {
2804 int node;
2805 int highest_node_number = Linux::numa_max_node();
2806 int num_nodes = 0;
2807
2808 // Get the total number of nodes in the system including nodes without memory.
2809 for (node = 0; node <= highest_node_number; node++) {
2810 if (is_node_in_existing_nodes(node)) {
2811 num_nodes++;
2812 }
2813 }
2814 return num_nodes;
2815 }
2816
2817 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2818 int highest_node_number = Linux::numa_max_node();
2819 size_t i = 0;
2820
2821 // Map all node ids in which it is possible to allocate memory. Also nodes are
2822 // not always consecutively available, i.e. available from 0 to the highest
2823 // node number. If the nodes have been bound explicitly using numactl membind,
2824 // then allocate memory from those nodes only.
2825 for (int node = 0; node <= highest_node_number; node++) {
2826 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
2827 ids[i++] = node;
2828 }
2829 }
2830 return i;
2831 }
2832
2833 bool os::get_page_info(char *start, page_info* info) {
2834 return false;
2835 }
2836
2837 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2838 page_info* page_found) {
2839 return end;
2840 }
2841
2842
2843 int os::Linux::sched_getcpu_syscall(void) {
2844 unsigned int cpu = 0;
2845 int retval = -1;
2846
2847 #if defined(IA32)
2848 #ifndef SYS_getcpu
2849 #define SYS_getcpu 318
2850 #endif
2851 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2852 #elif defined(AMD64)
2853 // Unfortunately we have to bring all these macros here from vsyscall.h
2854 // to be able to compile on old linuxes.
2855 #define __NR_vgetcpu 2
2856 #define VSYSCALL_START (-10UL << 20)
2857 #define VSYSCALL_SIZE 1024
2858 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2859 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2860 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2861 retval = vgetcpu(&cpu, NULL, NULL);
2862 #endif
2863
2864 return (retval == -1) ? retval : cpu;
2865 }
2866
2867 void os::Linux::sched_getcpu_init() {
2868 // sched_getcpu() should be in libc.
2869 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2870 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2871
2872 // If it's not, try a direct syscall.
2873 if (sched_getcpu() == -1) {
2874 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2875 (void*)&sched_getcpu_syscall));
2876 }
2877
2878 if (sched_getcpu() == -1) {
2879 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2880 }
2881 }
2882
2883 // Something to do with the numa-aware allocator needs these symbols
2884 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2885 extern "C" JNIEXPORT void numa_error(char *where) { }
2886
2887 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2888 // load symbol from base version instead.
2889 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2890 void *f = dlvsym(handle, name, "libnuma_1.1");
2891 if (f == NULL) {
2892 f = dlsym(handle, name);
2893 }
2894 return f;
2895 }
2896
2897 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2898 // Return NULL if the symbol is not defined in this particular version.
2899 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2900 return dlvsym(handle, name, "libnuma_1.2");
2901 }
2902
2903 bool os::Linux::libnuma_init() {
2904 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2905 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2906 if (handle != NULL) {
2907 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2908 libnuma_dlsym(handle, "numa_node_to_cpus")));
2909 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2910 libnuma_dlsym(handle, "numa_max_node")));
2911 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2912 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2913 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2914 libnuma_dlsym(handle, "numa_available")));
2915 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2916 libnuma_dlsym(handle, "numa_tonode_memory")));
2917 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2918 libnuma_dlsym(handle, "numa_interleave_memory")));
2919 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2920 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2921 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2922 libnuma_dlsym(handle, "numa_set_bind_policy")));
2923 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2924 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2925 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2926 libnuma_dlsym(handle, "numa_distance")));
2927 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2928 libnuma_v2_dlsym(handle, "numa_get_membind")));
2929 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
2930 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
2931
2932 if (numa_available() != -1) {
2933 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2934 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2935 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2936 set_numa_interleave_bitmask(_numa_get_interleave_mask());
2937 set_numa_membind_bitmask(_numa_get_membind());
2938 // Create an index -> node mapping, since nodes are not always consecutive
2939 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2940 rebuild_nindex_to_node_map();
2941 // Create a cpu -> node mapping
2942 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2943 rebuild_cpu_to_node_map();
2944 return true;
2945 }
2946 }
2947 }
2948 return false;
2949 }
2950
2951 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2952 // Creating guard page is very expensive. Java thread has HotSpot
2953 // guard pages, only enable glibc guard page for non-Java threads.
2954 // (Remember: compiler thread is a Java thread, too!)
2955 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2956 }
2957
2958 void os::Linux::rebuild_nindex_to_node_map() {
2959 int highest_node_number = Linux::numa_max_node();
2960
2961 nindex_to_node()->clear();
2962 for (int node = 0; node <= highest_node_number; node++) {
2963 if (Linux::is_node_in_existing_nodes(node)) {
2964 nindex_to_node()->append(node);
2965 }
2966 }
2967 }
2968
2969 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2970 // The table is later used in get_node_by_cpu().
2971 void os::Linux::rebuild_cpu_to_node_map() {
2972 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2973 // in libnuma (possible values are starting from 16,
2974 // and continuing up with every other power of 2, but less
2975 // than the maximum number of CPUs supported by kernel), and
2976 // is a subject to change (in libnuma version 2 the requirements
2977 // are more reasonable) we'll just hardcode the number they use
2978 // in the library.
2979 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2980
2981 size_t cpu_num = processor_count();
2982 size_t cpu_map_size = NCPUS / BitsPerCLong;
2983 size_t cpu_map_valid_size =
2984 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2985
2986 cpu_to_node()->clear();
2987 cpu_to_node()->at_grow(cpu_num - 1);
2988
2989 size_t node_num = get_existing_num_nodes();
2990
2991 int distance = 0;
2992 int closest_distance = INT_MAX;
2993 int closest_node = 0;
2994 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2995 for (size_t i = 0; i < node_num; i++) {
2996 // Check if node is configured (not a memory-less node). If it is not, find
2997 // the closest configured node. Check also if node is bound, i.e. it's allowed
2998 // to allocate memory from the node. If it's not allowed, map cpus in that node
2999 // to the closest node from which memory allocation is allowed.
3000 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3001 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3002 closest_distance = INT_MAX;
3003 // Check distance from all remaining nodes in the system. Ignore distance
3004 // from itself, from another non-configured node, and from another non-bound
3005 // node.
3006 for (size_t m = 0; m < node_num; m++) {
3007 if (m != i &&
3008 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3009 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3010 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3011 // If a closest node is found, update. There is always at least one
3012 // configured and bound node in the system so there is always at least
3013 // one node close.
3014 if (distance != 0 && distance < closest_distance) {
3015 closest_distance = distance;
3016 closest_node = nindex_to_node()->at(m);
3017 }
3018 }
3019 }
3020 } else {
3021 // Current node is already a configured node.
3022 closest_node = nindex_to_node()->at(i);
3023 }
3024
3025 // Get cpus from the original node and map them to the closest node. If node
3026 // is a configured node (not a memory-less node), then original node and
3027 // closest node are the same.
3028 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3029 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3030 if (cpu_map[j] != 0) {
3031 for (size_t k = 0; k < BitsPerCLong; k++) {
3032 if (cpu_map[j] & (1UL << k)) {
3033 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3034 }
3035 }
3036 }
3037 }
3038 }
3039 }
3040 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3041 }
3042
3043 int os::Linux::get_node_by_cpu(int cpu_id) {
3044 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3045 return cpu_to_node()->at(cpu_id);
3046 }
3047 return -1;
3048 }
3049
3050 GrowableArray<int>* os::Linux::_cpu_to_node;
3051 GrowableArray<int>* os::Linux::_nindex_to_node;
3052 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3053 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3054 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3055 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3056 os::Linux::numa_available_func_t os::Linux::_numa_available;
3057 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3058 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3059 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3060 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3061 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3062 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3063 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3064 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3065 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3066 unsigned long* os::Linux::_numa_all_nodes;
3067 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3068 struct bitmask* os::Linux::_numa_nodes_ptr;
3069 struct bitmask* os::Linux::_numa_interleave_bitmask;
3070 struct bitmask* os::Linux::_numa_membind_bitmask;
3071
3072 bool os::pd_uncommit_memory(char* addr, size_t size) {
3073 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3074 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3075 return res != (uintptr_t) MAP_FAILED;
3076 }
3077
3078 static address get_stack_commited_bottom(address bottom, size_t size) {
3079 address nbot = bottom;
3080 address ntop = bottom + size;
3081
3082 size_t page_sz = os::vm_page_size();
3083 unsigned pages = size / page_sz;
3084
3085 unsigned char vec[1];
3086 unsigned imin = 1, imax = pages + 1, imid;
3087 int mincore_return_value = 0;
3088
3089 assert(imin <= imax, "Unexpected page size");
3090
3091 while (imin < imax) {
3092 imid = (imax + imin) / 2;
3093 nbot = ntop - (imid * page_sz);
3094
3095 // Use a trick with mincore to check whether the page is mapped or not.
3096 // mincore sets vec to 1 if page resides in memory and to 0 if page
3097 // is swapped output but if page we are asking for is unmapped
3098 // it returns -1,ENOMEM
3099 mincore_return_value = mincore(nbot, page_sz, vec);
3100
3101 if (mincore_return_value == -1) {
3102 // Page is not mapped go up
3103 // to find first mapped page
3104 if (errno != EAGAIN) {
3105 assert(errno == ENOMEM, "Unexpected mincore errno");
3106 imax = imid;
3107 }
3108 } else {
3109 // Page is mapped go down
3110 // to find first not mapped page
3111 imin = imid + 1;
3112 }
3113 }
3114
3115 nbot = nbot + page_sz;
3116
3117 // Adjust stack bottom one page up if last checked page is not mapped
3118 if (mincore_return_value == -1) {
3119 nbot = nbot + page_sz;
3120 }
3121
3122 return nbot;
3123 }
3124
3125 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3126 int mincore_return_value;
3127 const size_t stripe = 1024; // query this many pages each time
3128 unsigned char vec[stripe + 1];
3129 // set a guard
3130 vec[stripe] = 'X';
3131
3132 const size_t page_sz = os::vm_page_size();
3133 size_t pages = size / page_sz;
3134
3135 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3136 assert(is_aligned(size, page_sz), "Size must be page aligned");
3137
3138 committed_start = NULL;
3139
3140 int loops = (pages + stripe - 1) / stripe;
3141 int committed_pages = 0;
3142 address loop_base = start;
3143 bool found_range = false;
3144
3145 for (int index = 0; index < loops && !found_range; index ++) {
3146 assert(pages > 0, "Nothing to do");
3147 int pages_to_query = (pages >= stripe) ? stripe : pages;
3148 pages -= pages_to_query;
3149
3150 // Get stable read
3151 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3152
3153 // During shutdown, some memory goes away without properly notifying NMT,
3154 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3155 // Bailout and return as not committed for now.
3156 if (mincore_return_value == -1 && errno == ENOMEM) {
3157 return false;
3158 }
3159
3160 assert(vec[stripe] == 'X', "overflow guard");
3161 assert(mincore_return_value == 0, "Range must be valid");
3162 // Process this stripe
3163 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3164 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3165 // End of current contiguous region
3166 if (committed_start != NULL) {
3167 found_range = true;
3168 break;
3169 }
3170 } else { // committed
3171 // Start of region
3172 if (committed_start == NULL) {
3173 committed_start = loop_base + page_sz * vecIdx;
3174 }
3175 committed_pages ++;
3176 }
3177 }
3178
3179 loop_base += pages_to_query * page_sz;
3180 }
3181
3182 if (committed_start != NULL) {
3183 assert(committed_pages > 0, "Must have committed region");
3184 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3185 assert(committed_start >= start && committed_start < start + size, "Out of range");
3186 committed_size = page_sz * committed_pages;
3187 return true;
3188 } else {
3189 assert(committed_pages == 0, "Should not have committed region");
3190 return false;
3191 }
3192 }
3193
3194
3195 // Linux uses a growable mapping for the stack, and if the mapping for
3196 // the stack guard pages is not removed when we detach a thread the
3197 // stack cannot grow beyond the pages where the stack guard was
3198 // mapped. If at some point later in the process the stack expands to
3199 // that point, the Linux kernel cannot expand the stack any further
3200 // because the guard pages are in the way, and a segfault occurs.
3201 //
3202 // However, it's essential not to split the stack region by unmapping
3203 // a region (leaving a hole) that's already part of the stack mapping,
3204 // so if the stack mapping has already grown beyond the guard pages at
3205 // the time we create them, we have to truncate the stack mapping.
3206 // So, we need to know the extent of the stack mapping when
3207 // create_stack_guard_pages() is called.
3208
3209 // We only need this for stacks that are growable: at the time of
3210 // writing thread stacks don't use growable mappings (i.e. those
3211 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3212 // only applies to the main thread.
3213
3214 // If the (growable) stack mapping already extends beyond the point
3215 // where we're going to put our guard pages, truncate the mapping at
3216 // that point by munmap()ping it. This ensures that when we later
3217 // munmap() the guard pages we don't leave a hole in the stack
3218 // mapping. This only affects the main/primordial thread
3219
3220 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3221 if (os::is_primordial_thread()) {
3222 // As we manually grow stack up to bottom inside create_attached_thread(),
3223 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3224 // we don't need to do anything special.
3225 // Check it first, before calling heavy function.
3226 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3227 unsigned char vec[1];
3228
3229 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3230 // Fallback to slow path on all errors, including EAGAIN
3231 stack_extent = (uintptr_t) get_stack_commited_bottom(
3232 os::Linux::initial_thread_stack_bottom(),
3233 (size_t)addr - stack_extent);
3234 }
3235
3236 if (stack_extent < (uintptr_t)addr) {
3237 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3238 }
3239 }
3240
3241 return os::commit_memory(addr, size, !ExecMem);
3242 }
3243
3244 // If this is a growable mapping, remove the guard pages entirely by
3245 // munmap()ping them. If not, just call uncommit_memory(). This only
3246 // affects the main/primordial thread, but guard against future OS changes.
3247 // It's safe to always unmap guard pages for primordial thread because we
3248 // always place it right after end of the mapped region.
3249
3250 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3251 uintptr_t stack_extent, stack_base;
3252
3253 if (os::is_primordial_thread()) {
3254 return ::munmap(addr, size) == 0;
3255 }
3256
3257 return os::uncommit_memory(addr, size);
3258 }
3259
3260 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3261 // at 'requested_addr'. If there are existing memory mappings at the same
3262 // location, however, they will be overwritten. If 'fixed' is false,
3263 // 'requested_addr' is only treated as a hint, the return value may or
3264 // may not start from the requested address. Unlike Linux mmap(), this
3265 // function returns NULL to indicate failure.
3266 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3267 char * addr;
3268 int flags;
3269
3270 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3271 if (fixed) {
3272 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3273 flags |= MAP_FIXED;
3274 }
3275
3276 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3277 // touch an uncommitted page. Otherwise, the read/write might
3278 // succeed if we have enough swap space to back the physical page.
3279 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3280 flags, -1, 0);
3281
3282 return addr == MAP_FAILED ? NULL : addr;
3283 }
3284
3285 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3286 // (req_addr != NULL) or with a given alignment.
3287 // - bytes shall be a multiple of alignment.
3288 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3289 // - alignment sets the alignment at which memory shall be allocated.
3290 // It must be a multiple of allocation granularity.
3291 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3292 // req_addr or NULL.
3293 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3294
3295 size_t extra_size = bytes;
3296 if (req_addr == NULL && alignment > 0) {
3297 extra_size += alignment;
3298 }
3299
3300 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3301 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3302 -1, 0);
3303 if (start == MAP_FAILED) {
3304 start = NULL;
3305 } else {
3306 if (req_addr != NULL) {
3307 if (start != req_addr) {
3308 ::munmap(start, extra_size);
3309 start = NULL;
3310 }
3311 } else {
3312 char* const start_aligned = align_up(start, alignment);
3313 char* const end_aligned = start_aligned + bytes;
3314 char* const end = start + extra_size;
3315 if (start_aligned > start) {
3316 ::munmap(start, start_aligned - start);
3317 }
3318 if (end_aligned < end) {
3319 ::munmap(end_aligned, end - end_aligned);
3320 }
3321 start = start_aligned;
3322 }
3323 }
3324 return start;
3325 }
3326
3327 static int anon_munmap(char * addr, size_t size) {
3328 return ::munmap(addr, size) == 0;
3329 }
3330
3331 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3332 size_t alignment_hint) {
3333 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3334 }
3335
3336 bool os::pd_release_memory(char* addr, size_t size) {
3337 return anon_munmap(addr, size);
3338 }
3339
3340 static bool linux_mprotect(char* addr, size_t size, int prot) {
3341 // Linux wants the mprotect address argument to be page aligned.
3342 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3343
3344 // According to SUSv3, mprotect() should only be used with mappings
3345 // established by mmap(), and mmap() always maps whole pages. Unaligned
3346 // 'addr' likely indicates problem in the VM (e.g. trying to change
3347 // protection of malloc'ed or statically allocated memory). Check the
3348 // caller if you hit this assert.
3349 assert(addr == bottom, "sanity check");
3350
3351 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3352 return ::mprotect(bottom, size, prot) == 0;
3353 }
3354
3355 // Set protections specified
3356 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3357 bool is_committed) {
3358 unsigned int p = 0;
3359 switch (prot) {
3360 case MEM_PROT_NONE: p = PROT_NONE; break;
3361 case MEM_PROT_READ: p = PROT_READ; break;
3362 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3363 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3364 default:
3365 ShouldNotReachHere();
3366 }
3367 // is_committed is unused.
3368 return linux_mprotect(addr, bytes, p);
3369 }
3370
3371 bool os::guard_memory(char* addr, size_t size) {
3372 return linux_mprotect(addr, size, PROT_NONE);
3373 }
3374
3375 bool os::unguard_memory(char* addr, size_t size) {
3376 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3377 }
3378
3379 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3380 size_t page_size) {
3381 bool result = false;
3382 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3383 MAP_ANONYMOUS|MAP_PRIVATE,
3384 -1, 0);
3385 if (p != MAP_FAILED) {
3386 void *aligned_p = align_up(p, page_size);
3387
3388 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3389
3390 munmap(p, page_size * 2);
3391 }
3392
3393 if (warn && !result) {
3394 warning("TransparentHugePages is not supported by the operating system.");
3395 }
3396
3397 return result;
3398 }
3399
3400 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3401 bool result = false;
3402 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3403 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3404 -1, 0);
3405
3406 if (p != MAP_FAILED) {
3407 // We don't know if this really is a huge page or not.
3408 FILE *fp = fopen("/proc/self/maps", "r");
3409 if (fp) {
3410 while (!feof(fp)) {
3411 char chars[257];
3412 long x = 0;
3413 if (fgets(chars, sizeof(chars), fp)) {
3414 if (sscanf(chars, "%lx-%*x", &x) == 1
3415 && x == (long)p) {
3416 if (strstr (chars, "hugepage")) {
3417 result = true;
3418 break;
3419 }
3420 }
3421 }
3422 }
3423 fclose(fp);
3424 }
3425 munmap(p, page_size);
3426 }
3427
3428 if (warn && !result) {
3429 warning("HugeTLBFS is not supported by the operating system.");
3430 }
3431
3432 return result;
3433 }
3434
3435 // From the coredump_filter documentation:
3436 //
3437 // - (bit 0) anonymous private memory
3438 // - (bit 1) anonymous shared memory
3439 // - (bit 2) file-backed private memory
3440 // - (bit 3) file-backed shared memory
3441 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3442 // effective only if the bit 2 is cleared)
3443 // - (bit 5) hugetlb private memory
3444 // - (bit 6) hugetlb shared memory
3445 // - (bit 7) dax private memory
3446 // - (bit 8) dax shared memory
3447 //
3448 static void set_coredump_filter(CoredumpFilterBit bit) {
3449 FILE *f;
3450 long cdm;
3451
3452 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3453 return;
3454 }
3455
3456 if (fscanf(f, "%lx", &cdm) != 1) {
3457 fclose(f);
3458 return;
3459 }
3460
3461 long saved_cdm = cdm;
3462 rewind(f);
3463 cdm |= bit;
3464
3465 if (cdm != saved_cdm) {
3466 fprintf(f, "%#lx", cdm);
3467 }
3468
3469 fclose(f);
3470 }
3471
3472 // Large page support
3473
3474 static size_t _large_page_size = 0;
3475
3476 size_t os::Linux::find_large_page_size() {
3477 size_t large_page_size = 0;
3478
3479 // large_page_size on Linux is used to round up heap size. x86 uses either
3480 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3481 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3482 // page as large as 256M.
3483 //
3484 // Here we try to figure out page size by parsing /proc/meminfo and looking
3485 // for a line with the following format:
3486 // Hugepagesize: 2048 kB
3487 //
3488 // If we can't determine the value (e.g. /proc is not mounted, or the text
3489 // format has been changed), we'll use the largest page size supported by
3490 // the processor.
3491
3492 #ifndef ZERO
3493 large_page_size =
3494 AARCH64_ONLY(2 * M)
3495 AMD64_ONLY(2 * M)
3496 ARM32_ONLY(2 * M)
3497 IA32_ONLY(4 * M)
3498 IA64_ONLY(256 * M)
3499 PPC_ONLY(4 * M)
3500 S390_ONLY(1 * M)
3501 SPARC_ONLY(4 * M);
3502 #endif // ZERO
3503
3504 FILE *fp = fopen("/proc/meminfo", "r");
3505 if (fp) {
3506 while (!feof(fp)) {
3507 int x = 0;
3508 char buf[16];
3509 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3510 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3511 large_page_size = x * K;
3512 break;
3513 }
3514 } else {
3515 // skip to next line
3516 for (;;) {
3517 int ch = fgetc(fp);
3518 if (ch == EOF || ch == (int)'\n') break;
3519 }
3520 }
3521 }
3522 fclose(fp);
3523 }
3524
3525 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3526 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3527 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3528 proper_unit_for_byte_size(large_page_size));
3529 }
3530
3531 return large_page_size;
3532 }
3533
3534 size_t os::Linux::setup_large_page_size() {
3535 _large_page_size = Linux::find_large_page_size();
3536 const size_t default_page_size = (size_t)Linux::page_size();
3537 if (_large_page_size > default_page_size) {
3538 _page_sizes[0] = _large_page_size;
3539 _page_sizes[1] = default_page_size;
3540 _page_sizes[2] = 0;
3541 }
3542
3543 return _large_page_size;
3544 }
3545
3546 bool os::Linux::setup_large_page_type(size_t page_size) {
3547 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3548 FLAG_IS_DEFAULT(UseSHM) &&
3549 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3550
3551 // The type of large pages has not been specified by the user.
3552
3553 // Try UseHugeTLBFS and then UseSHM.
3554 UseHugeTLBFS = UseSHM = true;
3555
3556 // Don't try UseTransparentHugePages since there are known
3557 // performance issues with it turned on. This might change in the future.
3558 UseTransparentHugePages = false;
3559 }
3560
3561 if (UseTransparentHugePages) {
3562 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3563 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3564 UseHugeTLBFS = false;
3565 UseSHM = false;
3566 return true;
3567 }
3568 UseTransparentHugePages = false;
3569 }
3570
3571 if (UseHugeTLBFS) {
3572 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3573 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3574 UseSHM = false;
3575 return true;
3576 }
3577 UseHugeTLBFS = false;
3578 }
3579
3580 return UseSHM;
3581 }
3582
3583 void os::large_page_init() {
3584 if (!UseLargePages &&
3585 !UseTransparentHugePages &&
3586 !UseHugeTLBFS &&
3587 !UseSHM) {
3588 // Not using large pages.
3589 return;
3590 }
3591
3592 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3593 // The user explicitly turned off large pages.
3594 // Ignore the rest of the large pages flags.
3595 UseTransparentHugePages = false;
3596 UseHugeTLBFS = false;
3597 UseSHM = false;
3598 return;
3599 }
3600
3601 size_t large_page_size = Linux::setup_large_page_size();
3602 UseLargePages = Linux::setup_large_page_type(large_page_size);
3603
3604 set_coredump_filter(LARGEPAGES_BIT);
3605 }
3606
3607 #ifndef SHM_HUGETLB
3608 #define SHM_HUGETLB 04000
3609 #endif
3610
3611 #define shm_warning_format(format, ...) \
3612 do { \
3613 if (UseLargePages && \
3614 (!FLAG_IS_DEFAULT(UseLargePages) || \
3615 !FLAG_IS_DEFAULT(UseSHM) || \
3616 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3617 warning(format, __VA_ARGS__); \
3618 } \
3619 } while (0)
3620
3621 #define shm_warning(str) shm_warning_format("%s", str)
3622
3623 #define shm_warning_with_errno(str) \
3624 do { \
3625 int err = errno; \
3626 shm_warning_format(str " (error = %d)", err); \
3627 } while (0)
3628
3629 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3630 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3631
3632 if (!is_aligned(alignment, SHMLBA)) {
3633 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3634 return NULL;
3635 }
3636
3637 // To ensure that we get 'alignment' aligned memory from shmat,
3638 // we pre-reserve aligned virtual memory and then attach to that.
3639
3640 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3641 if (pre_reserved_addr == NULL) {
3642 // Couldn't pre-reserve aligned memory.
3643 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3644 return NULL;
3645 }
3646
3647 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3648 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3649
3650 if ((intptr_t)addr == -1) {
3651 int err = errno;
3652 shm_warning_with_errno("Failed to attach shared memory.");
3653
3654 assert(err != EACCES, "Unexpected error");
3655 assert(err != EIDRM, "Unexpected error");
3656 assert(err != EINVAL, "Unexpected error");
3657
3658 // Since we don't know if the kernel unmapped the pre-reserved memory area
3659 // we can't unmap it, since that would potentially unmap memory that was
3660 // mapped from other threads.
3661 return NULL;
3662 }
3663
3664 return addr;
3665 }
3666
3667 static char* shmat_at_address(int shmid, char* req_addr) {
3668 if (!is_aligned(req_addr, SHMLBA)) {
3669 assert(false, "Requested address needs to be SHMLBA aligned");
3670 return NULL;
3671 }
3672
3673 char* addr = (char*)shmat(shmid, req_addr, 0);
3674
3675 if ((intptr_t)addr == -1) {
3676 shm_warning_with_errno("Failed to attach shared memory.");
3677 return NULL;
3678 }
3679
3680 return addr;
3681 }
3682
3683 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3684 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3685 if (req_addr != NULL) {
3686 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3687 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3688 return shmat_at_address(shmid, req_addr);
3689 }
3690
3691 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3692 // return large page size aligned memory addresses when req_addr == NULL.
3693 // However, if the alignment is larger than the large page size, we have
3694 // to manually ensure that the memory returned is 'alignment' aligned.
3695 if (alignment > os::large_page_size()) {
3696 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3697 return shmat_with_alignment(shmid, bytes, alignment);
3698 } else {
3699 return shmat_at_address(shmid, NULL);
3700 }
3701 }
3702
3703 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3704 char* req_addr, bool exec) {
3705 // "exec" is passed in but not used. Creating the shared image for
3706 // the code cache doesn't have an SHM_X executable permission to check.
3707 assert(UseLargePages && UseSHM, "only for SHM large pages");
3708 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3709 assert(is_aligned(req_addr, alignment), "Unaligned address");
3710
3711 if (!is_aligned(bytes, os::large_page_size())) {
3712 return NULL; // Fallback to small pages.
3713 }
3714
3715 // Create a large shared memory region to attach to based on size.
3716 // Currently, size is the total size of the heap.
3717 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3718 if (shmid == -1) {
3719 // Possible reasons for shmget failure:
3720 // 1. shmmax is too small for Java heap.
3721 // > check shmmax value: cat /proc/sys/kernel/shmmax
3722 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3723 // 2. not enough large page memory.
3724 // > check available large pages: cat /proc/meminfo
3725 // > increase amount of large pages:
3726 // echo new_value > /proc/sys/vm/nr_hugepages
3727 // Note 1: different Linux may use different name for this property,
3728 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3729 // Note 2: it's possible there's enough physical memory available but
3730 // they are so fragmented after a long run that they can't
3731 // coalesce into large pages. Try to reserve large pages when
3732 // the system is still "fresh".
3733 shm_warning_with_errno("Failed to reserve shared memory.");
3734 return NULL;
3735 }
3736
3737 // Attach to the region.
3738 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3739
3740 // Remove shmid. If shmat() is successful, the actual shared memory segment
3741 // will be deleted when it's detached by shmdt() or when the process
3742 // terminates. If shmat() is not successful this will remove the shared
3743 // segment immediately.
3744 shmctl(shmid, IPC_RMID, NULL);
3745
3746 return addr;
3747 }
3748
3749 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3750 int error) {
3751 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3752
3753 bool warn_on_failure = UseLargePages &&
3754 (!FLAG_IS_DEFAULT(UseLargePages) ||
3755 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3756 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3757
3758 if (warn_on_failure) {
3759 char msg[128];
3760 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3761 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3762 warning("%s", msg);
3763 }
3764 }
3765
3766 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3767 char* req_addr,
3768 bool exec) {
3769 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3770 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3771 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3772
3773 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3774 char* addr = (char*)::mmap(req_addr, bytes, prot,
3775 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3776 -1, 0);
3777
3778 if (addr == MAP_FAILED) {
3779 warn_on_large_pages_failure(req_addr, bytes, errno);
3780 return NULL;
3781 }
3782
3783 assert(is_aligned(addr, os::large_page_size()), "Must be");
3784
3785 return addr;
3786 }
3787
3788 // Reserve memory using mmap(MAP_HUGETLB).
3789 // - bytes shall be a multiple of alignment.
3790 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3791 // - alignment sets the alignment at which memory shall be allocated.
3792 // It must be a multiple of allocation granularity.
3793 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3794 // req_addr or NULL.
3795 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3796 size_t alignment,
3797 char* req_addr,
3798 bool exec) {
3799 size_t large_page_size = os::large_page_size();
3800 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3801
3802 assert(is_aligned(req_addr, alignment), "Must be");
3803 assert(is_aligned(bytes, alignment), "Must be");
3804
3805 // First reserve - but not commit - the address range in small pages.
3806 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3807
3808 if (start == NULL) {
3809 return NULL;
3810 }
3811
3812 assert(is_aligned(start, alignment), "Must be");
3813
3814 char* end = start + bytes;
3815
3816 // Find the regions of the allocated chunk that can be promoted to large pages.
3817 char* lp_start = align_up(start, large_page_size);
3818 char* lp_end = align_down(end, large_page_size);
3819
3820 size_t lp_bytes = lp_end - lp_start;
3821
3822 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3823
3824 if (lp_bytes == 0) {
3825 // The mapped region doesn't even span the start and the end of a large page.
3826 // Fall back to allocate a non-special area.
3827 ::munmap(start, end - start);
3828 return NULL;
3829 }
3830
3831 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3832
3833 void* result;
3834
3835 // Commit small-paged leading area.
3836 if (start != lp_start) {
3837 result = ::mmap(start, lp_start - start, prot,
3838 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3839 -1, 0);
3840 if (result == MAP_FAILED) {
3841 ::munmap(lp_start, end - lp_start);
3842 return NULL;
3843 }
3844 }
3845
3846 // Commit large-paged area.
3847 result = ::mmap(lp_start, lp_bytes, prot,
3848 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3849 -1, 0);
3850 if (result == MAP_FAILED) {
3851 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3852 // If the mmap above fails, the large pages region will be unmapped and we
3853 // have regions before and after with small pages. Release these regions.
3854 //
3855 // | mapped | unmapped | mapped |
3856 // ^ ^ ^ ^
3857 // start lp_start lp_end end
3858 //
3859 ::munmap(start, lp_start - start);
3860 ::munmap(lp_end, end - lp_end);
3861 return NULL;
3862 }
3863
3864 // Commit small-paged trailing area.
3865 if (lp_end != end) {
3866 result = ::mmap(lp_end, end - lp_end, prot,
3867 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3868 -1, 0);
3869 if (result == MAP_FAILED) {
3870 ::munmap(start, lp_end - start);
3871 return NULL;
3872 }
3873 }
3874
3875 return start;
3876 }
3877
3878 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3879 size_t alignment,
3880 char* req_addr,
3881 bool exec) {
3882 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3883 assert(is_aligned(req_addr, alignment), "Must be");
3884 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3885 assert(is_power_of_2(os::large_page_size()), "Must be");
3886 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3887
3888 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3889 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3890 } else {
3891 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3892 }
3893 }
3894
3895 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3896 char* req_addr, bool exec) {
3897 assert(UseLargePages, "only for large pages");
3898
3899 char* addr;
3900 if (UseSHM) {
3901 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3902 } else {
3903 assert(UseHugeTLBFS, "must be");
3904 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3905 }
3906
3907 if (addr != NULL) {
3908 if (UseNUMAInterleaving) {
3909 numa_make_global(addr, bytes);
3910 }
3911
3912 // The memory is committed
3913 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3914 }
3915
3916 return addr;
3917 }
3918
3919 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3920 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3921 return shmdt(base) == 0;
3922 }
3923
3924 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3925 return pd_release_memory(base, bytes);
3926 }
3927
3928 bool os::release_memory_special(char* base, size_t bytes) {
3929 bool res;
3930 if (MemTracker::tracking_level() > NMT_minimal) {
3931 Tracker tkr(Tracker::release);
3932 res = os::Linux::release_memory_special_impl(base, bytes);
3933 if (res) {
3934 tkr.record((address)base, bytes);
3935 }
3936
3937 } else {
3938 res = os::Linux::release_memory_special_impl(base, bytes);
3939 }
3940 return res;
3941 }
3942
3943 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3944 assert(UseLargePages, "only for large pages");
3945 bool res;
3946
3947 if (UseSHM) {
3948 res = os::Linux::release_memory_special_shm(base, bytes);
3949 } else {
3950 assert(UseHugeTLBFS, "must be");
3951 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3952 }
3953 return res;
3954 }
3955
3956 size_t os::large_page_size() {
3957 return _large_page_size;
3958 }
3959
3960 // With SysV SHM the entire memory region must be allocated as shared
3961 // memory.
3962 // HugeTLBFS allows application to commit large page memory on demand.
3963 // However, when committing memory with HugeTLBFS fails, the region
3964 // that was supposed to be committed will lose the old reservation
3965 // and allow other threads to steal that memory region. Because of this
3966 // behavior we can't commit HugeTLBFS memory.
3967 bool os::can_commit_large_page_memory() {
3968 return UseTransparentHugePages;
3969 }
3970
3971 bool os::can_execute_large_page_memory() {
3972 return UseTransparentHugePages || UseHugeTLBFS;
3973 }
3974
3975 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3976 assert(file_desc >= 0, "file_desc is not valid");
3977 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3978 if (result != NULL) {
3979 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3980 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3981 }
3982 }
3983 return result;
3984 }
3985
3986 // Reserve memory at an arbitrary address, only if that area is
3987 // available (and not reserved for something else).
3988
3989 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3990 const int max_tries = 10;
3991 char* base[max_tries];
3992 size_t size[max_tries];
3993 const size_t gap = 0x000000;
3994
3995 // Assert only that the size is a multiple of the page size, since
3996 // that's all that mmap requires, and since that's all we really know
3997 // about at this low abstraction level. If we need higher alignment,
3998 // we can either pass an alignment to this method or verify alignment
3999 // in one of the methods further up the call chain. See bug 5044738.
4000 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4001
4002 // Repeatedly allocate blocks until the block is allocated at the
4003 // right spot.
4004
4005 // Linux mmap allows caller to pass an address as hint; give it a try first,
4006 // if kernel honors the hint then we can return immediately.
4007 char * addr = anon_mmap(requested_addr, bytes, false);
4008 if (addr == requested_addr) {
4009 return requested_addr;
4010 }
4011
4012 if (addr != NULL) {
4013 // mmap() is successful but it fails to reserve at the requested address
4014 anon_munmap(addr, bytes);
4015 }
4016
4017 int i;
4018 for (i = 0; i < max_tries; ++i) {
4019 base[i] = reserve_memory(bytes);
4020
4021 if (base[i] != NULL) {
4022 // Is this the block we wanted?
4023 if (base[i] == requested_addr) {
4024 size[i] = bytes;
4025 break;
4026 }
4027
4028 // Does this overlap the block we wanted? Give back the overlapped
4029 // parts and try again.
4030
4031 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4032 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4033 unmap_memory(base[i], top_overlap);
4034 base[i] += top_overlap;
4035 size[i] = bytes - top_overlap;
4036 } else {
4037 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4038 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4039 unmap_memory(requested_addr, bottom_overlap);
4040 size[i] = bytes - bottom_overlap;
4041 } else {
4042 size[i] = bytes;
4043 }
4044 }
4045 }
4046 }
4047
4048 // Give back the unused reserved pieces.
4049
4050 for (int j = 0; j < i; ++j) {
4051 if (base[j] != NULL) {
4052 unmap_memory(base[j], size[j]);
4053 }
4054 }
4055
4056 if (i < max_tries) {
4057 return requested_addr;
4058 } else {
4059 return NULL;
4060 }
4061 }
4062
4063 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4064 void os::infinite_sleep() {
4065 while (true) { // sleep forever ...
4066 ::sleep(100); // ... 100 seconds at a time
4067 }
4068 }
4069
4070 // Used to convert frequent JVM_Yield() to nops
4071 bool os::dont_yield() {
4072 return DontYieldALot;
4073 }
4074
4075 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4076 // actually give up the CPU. Since skip buddy (v2.6.28):
4077 //
4078 // * Sets the yielding task as skip buddy for current CPU's run queue.
4079 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4080 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4081 //
4082 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4083 // getting re-scheduled immediately.
4084 //
4085 void os::naked_yield() {
4086 sched_yield();
4087 }
4088
4089 ////////////////////////////////////////////////////////////////////////////////
4090 // thread priority support
4091
4092 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4093 // only supports dynamic priority, static priority must be zero. For real-time
4094 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4095 // However, for large multi-threaded applications, SCHED_RR is not only slower
4096 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4097 // of 5 runs - Sep 2005).
4098 //
4099 // The following code actually changes the niceness of kernel-thread/LWP. It
4100 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4101 // not the entire user process, and user level threads are 1:1 mapped to kernel
4102 // threads. It has always been the case, but could change in the future. For
4103 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4104 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4105 // (e.g., root privilege or CAP_SYS_NICE capability).
4106
4107 int os::java_to_os_priority[CriticalPriority + 1] = {
4108 19, // 0 Entry should never be used
4109
4110 4, // 1 MinPriority
4111 3, // 2
4112 2, // 3
4113
4114 1, // 4
4115 0, // 5 NormPriority
4116 -1, // 6
4117
4118 -2, // 7
4119 -3, // 8
4120 -4, // 9 NearMaxPriority
4121
4122 -5, // 10 MaxPriority
4123
4124 -5 // 11 CriticalPriority
4125 };
4126
4127 static int prio_init() {
4128 if (ThreadPriorityPolicy == 1) {
4129 if (geteuid() != 0) {
4130 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4131 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4132 "e.g., being the root user. If the necessary permission is not " \
4133 "possessed, changes to priority will be silently ignored.");
4134 }
4135 }
4136 }
4137 if (UseCriticalJavaThreadPriority) {
4138 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4139 }
4140 return 0;
4141 }
4142
4143 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4144 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4145
4146 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4147 return (ret == 0) ? OS_OK : OS_ERR;
4148 }
4149
4150 OSReturn os::get_native_priority(const Thread* const thread,
4151 int *priority_ptr) {
4152 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4153 *priority_ptr = java_to_os_priority[NormPriority];
4154 return OS_OK;
4155 }
4156
4157 errno = 0;
4158 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4159 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4160 }
4161
4162 ////////////////////////////////////////////////////////////////////////////////
4163 // suspend/resume support
4164
4165 // The low-level signal-based suspend/resume support is a remnant from the
4166 // old VM-suspension that used to be for java-suspension, safepoints etc,
4167 // within hotspot. Currently used by JFR's OSThreadSampler
4168 //
4169 // The remaining code is greatly simplified from the more general suspension
4170 // code that used to be used.
4171 //
4172 // The protocol is quite simple:
4173 // - suspend:
4174 // - sends a signal to the target thread
4175 // - polls the suspend state of the osthread using a yield loop
4176 // - target thread signal handler (SR_handler) sets suspend state
4177 // and blocks in sigsuspend until continued
4178 // - resume:
4179 // - sets target osthread state to continue
4180 // - sends signal to end the sigsuspend loop in the SR_handler
4181 //
4182 // Note that the SR_lock plays no role in this suspend/resume protocol,
4183 // but is checked for NULL in SR_handler as a thread termination indicator.
4184 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4185 //
4186 // Note that resume_clear_context() and suspend_save_context() are needed
4187 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4188 // which in part is used by:
4189 // - Forte Analyzer: AsyncGetCallTrace()
4190 // - StackBanging: get_frame_at_stack_banging_point()
4191
4192 static void resume_clear_context(OSThread *osthread) {
4193 osthread->set_ucontext(NULL);
4194 osthread->set_siginfo(NULL);
4195 }
4196
4197 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4198 ucontext_t* context) {
4199 osthread->set_ucontext(context);
4200 osthread->set_siginfo(siginfo);
4201 }
4202
4203 // Handler function invoked when a thread's execution is suspended or
4204 // resumed. We have to be careful that only async-safe functions are
4205 // called here (Note: most pthread functions are not async safe and
4206 // should be avoided.)
4207 //
4208 // Note: sigwait() is a more natural fit than sigsuspend() from an
4209 // interface point of view, but sigwait() prevents the signal hander
4210 // from being run. libpthread would get very confused by not having
4211 // its signal handlers run and prevents sigwait()'s use with the
4212 // mutex granting granting signal.
4213 //
4214 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4215 //
4216 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4217 // Save and restore errno to avoid confusing native code with EINTR
4218 // after sigsuspend.
4219 int old_errno = errno;
4220
4221 Thread* thread = Thread::current_or_null_safe();
4222 assert(thread != NULL, "Missing current thread in SR_handler");
4223
4224 // On some systems we have seen signal delivery get "stuck" until the signal
4225 // mask is changed as part of thread termination. Check that the current thread
4226 // has not already terminated (via SR_lock()) - else the following assertion
4227 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4228 // destructor has completed.
4229
4230 if (thread->SR_lock() == NULL) {
4231 return;
4232 }
4233
4234 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4235
4236 OSThread* osthread = thread->osthread();
4237
4238 os::SuspendResume::State current = osthread->sr.state();
4239 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4240 suspend_save_context(osthread, siginfo, context);
4241
4242 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4243 os::SuspendResume::State state = osthread->sr.suspended();
4244 if (state == os::SuspendResume::SR_SUSPENDED) {
4245 sigset_t suspend_set; // signals for sigsuspend()
4246 sigemptyset(&suspend_set);
4247 // get current set of blocked signals and unblock resume signal
4248 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4249 sigdelset(&suspend_set, SR_signum);
4250
4251 sr_semaphore.signal();
4252 // wait here until we are resumed
4253 while (1) {
4254 sigsuspend(&suspend_set);
4255
4256 os::SuspendResume::State result = osthread->sr.running();
4257 if (result == os::SuspendResume::SR_RUNNING) {
4258 sr_semaphore.signal();
4259 break;
4260 }
4261 }
4262
4263 } else if (state == os::SuspendResume::SR_RUNNING) {
4264 // request was cancelled, continue
4265 } else {
4266 ShouldNotReachHere();
4267 }
4268
4269 resume_clear_context(osthread);
4270 } else if (current == os::SuspendResume::SR_RUNNING) {
4271 // request was cancelled, continue
4272 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4273 // ignore
4274 } else {
4275 // ignore
4276 }
4277
4278 errno = old_errno;
4279 }
4280
4281 static int SR_initialize() {
4282 struct sigaction act;
4283 char *s;
4284
4285 // Get signal number to use for suspend/resume
4286 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4287 int sig = ::strtol(s, 0, 10);
4288 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4289 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4290 SR_signum = sig;
4291 } else {
4292 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4293 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4294 }
4295 }
4296
4297 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4298 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4299
4300 sigemptyset(&SR_sigset);
4301 sigaddset(&SR_sigset, SR_signum);
4302
4303 // Set up signal handler for suspend/resume
4304 act.sa_flags = SA_RESTART|SA_SIGINFO;
4305 act.sa_handler = (void (*)(int)) SR_handler;
4306
4307 // SR_signum is blocked by default.
4308 // 4528190 - We also need to block pthread restart signal (32 on all
4309 // supported Linux platforms). Note that LinuxThreads need to block
4310 // this signal for all threads to work properly. So we don't have
4311 // to use hard-coded signal number when setting up the mask.
4312 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4313
4314 if (sigaction(SR_signum, &act, 0) == -1) {
4315 return -1;
4316 }
4317
4318 // Save signal flag
4319 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4320 return 0;
4321 }
4322
4323 static int sr_notify(OSThread* osthread) {
4324 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4325 assert_status(status == 0, status, "pthread_kill");
4326 return status;
4327 }
4328
4329 // "Randomly" selected value for how long we want to spin
4330 // before bailing out on suspending a thread, also how often
4331 // we send a signal to a thread we want to resume
4332 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4333 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4334
4335 // returns true on success and false on error - really an error is fatal
4336 // but this seems the normal response to library errors
4337 static bool do_suspend(OSThread* osthread) {
4338 assert(osthread->sr.is_running(), "thread should be running");
4339 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4340
4341 // mark as suspended and send signal
4342 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4343 // failed to switch, state wasn't running?
4344 ShouldNotReachHere();
4345 return false;
4346 }
4347
4348 if (sr_notify(osthread) != 0) {
4349 ShouldNotReachHere();
4350 }
4351
4352 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4353 while (true) {
4354 if (sr_semaphore.timedwait(2)) {
4355 break;
4356 } else {
4357 // timeout
4358 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4359 if (cancelled == os::SuspendResume::SR_RUNNING) {
4360 return false;
4361 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4362 // make sure that we consume the signal on the semaphore as well
4363 sr_semaphore.wait();
4364 break;
4365 } else {
4366 ShouldNotReachHere();
4367 return false;
4368 }
4369 }
4370 }
4371
4372 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4373 return true;
4374 }
4375
4376 static void do_resume(OSThread* osthread) {
4377 assert(osthread->sr.is_suspended(), "thread should be suspended");
4378 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4379
4380 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4381 // failed to switch to WAKEUP_REQUEST
4382 ShouldNotReachHere();
4383 return;
4384 }
4385
4386 while (true) {
4387 if (sr_notify(osthread) == 0) {
4388 if (sr_semaphore.timedwait(2)) {
4389 if (osthread->sr.is_running()) {
4390 return;
4391 }
4392 }
4393 } else {
4394 ShouldNotReachHere();
4395 }
4396 }
4397
4398 guarantee(osthread->sr.is_running(), "Must be running!");
4399 }
4400
4401 ///////////////////////////////////////////////////////////////////////////////////
4402 // signal handling (except suspend/resume)
4403
4404 // This routine may be used by user applications as a "hook" to catch signals.
4405 // The user-defined signal handler must pass unrecognized signals to this
4406 // routine, and if it returns true (non-zero), then the signal handler must
4407 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4408 // routine will never retun false (zero), but instead will execute a VM panic
4409 // routine kill the process.
4410 //
4411 // If this routine returns false, it is OK to call it again. This allows
4412 // the user-defined signal handler to perform checks either before or after
4413 // the VM performs its own checks. Naturally, the user code would be making
4414 // a serious error if it tried to handle an exception (such as a null check
4415 // or breakpoint) that the VM was generating for its own correct operation.
4416 //
4417 // This routine may recognize any of the following kinds of signals:
4418 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4419 // It should be consulted by handlers for any of those signals.
4420 //
4421 // The caller of this routine must pass in the three arguments supplied
4422 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4423 // field of the structure passed to sigaction(). This routine assumes that
4424 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4425 //
4426 // Note that the VM will print warnings if it detects conflicting signal
4427 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4428 //
4429 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4430 siginfo_t* siginfo,
4431 void* ucontext,
4432 int abort_if_unrecognized);
4433
4434 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4435 assert(info != NULL && uc != NULL, "it must be old kernel");
4436 int orig_errno = errno; // Preserve errno value over signal handler.
4437 JVM_handle_linux_signal(sig, info, uc, true);
4438 errno = orig_errno;
4439 }
4440
4441
4442 // This boolean allows users to forward their own non-matching signals
4443 // to JVM_handle_linux_signal, harmlessly.
4444 bool os::Linux::signal_handlers_are_installed = false;
4445
4446 // For signal-chaining
4447 bool os::Linux::libjsig_is_loaded = false;
4448 typedef struct sigaction *(*get_signal_t)(int);
4449 get_signal_t os::Linux::get_signal_action = NULL;
4450
4451 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4452 struct sigaction *actp = NULL;
4453
4454 if (libjsig_is_loaded) {
4455 // Retrieve the old signal handler from libjsig
4456 actp = (*get_signal_action)(sig);
4457 }
4458 if (actp == NULL) {
4459 // Retrieve the preinstalled signal handler from jvm
4460 actp = os::Posix::get_preinstalled_handler(sig);
4461 }
4462
4463 return actp;
4464 }
4465
4466 static bool call_chained_handler(struct sigaction *actp, int sig,
4467 siginfo_t *siginfo, void *context) {
4468 // Call the old signal handler
4469 if (actp->sa_handler == SIG_DFL) {
4470 // It's more reasonable to let jvm treat it as an unexpected exception
4471 // instead of taking the default action.
4472 return false;
4473 } else if (actp->sa_handler != SIG_IGN) {
4474 if ((actp->sa_flags & SA_NODEFER) == 0) {
4475 // automaticlly block the signal
4476 sigaddset(&(actp->sa_mask), sig);
4477 }
4478
4479 sa_handler_t hand = NULL;
4480 sa_sigaction_t sa = NULL;
4481 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4482 // retrieve the chained handler
4483 if (siginfo_flag_set) {
4484 sa = actp->sa_sigaction;
4485 } else {
4486 hand = actp->sa_handler;
4487 }
4488
4489 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4490 actp->sa_handler = SIG_DFL;
4491 }
4492
4493 // try to honor the signal mask
4494 sigset_t oset;
4495 sigemptyset(&oset);
4496 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4497
4498 // call into the chained handler
4499 if (siginfo_flag_set) {
4500 (*sa)(sig, siginfo, context);
4501 } else {
4502 (*hand)(sig);
4503 }
4504
4505 // restore the signal mask
4506 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4507 }
4508 // Tell jvm's signal handler the signal is taken care of.
4509 return true;
4510 }
4511
4512 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4513 bool chained = false;
4514 // signal-chaining
4515 if (UseSignalChaining) {
4516 struct sigaction *actp = get_chained_signal_action(sig);
4517 if (actp != NULL) {
4518 chained = call_chained_handler(actp, sig, siginfo, context);
4519 }
4520 }
4521 return chained;
4522 }
4523
4524 // for diagnostic
4525 int sigflags[NSIG];
4526
4527 int os::Linux::get_our_sigflags(int sig) {
4528 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4529 return sigflags[sig];
4530 }
4531
4532 void os::Linux::set_our_sigflags(int sig, int flags) {
4533 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4534 if (sig > 0 && sig < NSIG) {
4535 sigflags[sig] = flags;
4536 }
4537 }
4538
4539 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4540 // Check for overwrite.
4541 struct sigaction oldAct;
4542 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4543
4544 void* oldhand = oldAct.sa_sigaction
4545 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4546 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4547 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4548 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4549 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4550 if (AllowUserSignalHandlers || !set_installed) {
4551 // Do not overwrite; user takes responsibility to forward to us.
4552 return;
4553 } else if (UseSignalChaining) {
4554 // save the old handler in jvm
4555 os::Posix::save_preinstalled_handler(sig, oldAct);
4556 // libjsig also interposes the sigaction() call below and saves the
4557 // old sigaction on it own.
4558 } else {
4559 fatal("Encountered unexpected pre-existing sigaction handler "
4560 "%#lx for signal %d.", (long)oldhand, sig);
4561 }
4562 }
4563
4564 struct sigaction sigAct;
4565 sigfillset(&(sigAct.sa_mask));
4566 sigAct.sa_handler = SIG_DFL;
4567 if (!set_installed) {
4568 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4569 } else {
4570 sigAct.sa_sigaction = signalHandler;
4571 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4572 }
4573 // Save flags, which are set by ours
4574 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4575 sigflags[sig] = sigAct.sa_flags;
4576
4577 int ret = sigaction(sig, &sigAct, &oldAct);
4578 assert(ret == 0, "check");
4579
4580 void* oldhand2 = oldAct.sa_sigaction
4581 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4582 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4583 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4584 }
4585
4586 // install signal handlers for signals that HotSpot needs to
4587 // handle in order to support Java-level exception handling.
4588
4589 void os::Linux::install_signal_handlers() {
4590 if (!signal_handlers_are_installed) {
4591 signal_handlers_are_installed = true;
4592
4593 // signal-chaining
4594 typedef void (*signal_setting_t)();
4595 signal_setting_t begin_signal_setting = NULL;
4596 signal_setting_t end_signal_setting = NULL;
4597 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4598 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4599 if (begin_signal_setting != NULL) {
4600 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4601 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4602 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4603 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4604 libjsig_is_loaded = true;
4605 assert(UseSignalChaining, "should enable signal-chaining");
4606 }
4607 if (libjsig_is_loaded) {
4608 // Tell libjsig jvm is setting signal handlers
4609 (*begin_signal_setting)();
4610 }
4611
4612 os::Posix::init_sigs();
4613 set_signal_handler(SIGSEGV, true);
4614 set_signal_handler(SIGPIPE, true);
4615 set_signal_handler(SIGBUS, true);
4616 set_signal_handler(SIGILL, true);
4617 set_signal_handler(SIGFPE, true);
4618 #if defined(PPC64)
4619 set_signal_handler(SIGTRAP, true);
4620 #endif
4621 set_signal_handler(SIGXFSZ, true);
4622
4623 if (libjsig_is_loaded) {
4624 // Tell libjsig jvm finishes setting signal handlers
4625 (*end_signal_setting)();
4626 }
4627
4628 // We don't activate signal checker if libjsig is in place, we trust ourselves
4629 // and if UserSignalHandler is installed all bets are off.
4630 // Log that signal checking is off only if -verbose:jni is specified.
4631 if (CheckJNICalls) {
4632 if (libjsig_is_loaded) {
4633 if (PrintJNIResolving) {
4634 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4635 }
4636 check_signals = false;
4637 }
4638 if (AllowUserSignalHandlers) {
4639 if (PrintJNIResolving) {
4640 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4641 }
4642 check_signals = false;
4643 }
4644 }
4645 }
4646 }
4647
4648 // This is the fastest way to get thread cpu time on Linux.
4649 // Returns cpu time (user+sys) for any thread, not only for current.
4650 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4651 // It might work on 2.6.10+ with a special kernel/glibc patch.
4652 // For reference, please, see IEEE Std 1003.1-2004:
4653 // http://www.unix.org/single_unix_specification
4654
4655 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4656 struct timespec tp;
4657 int rc = os::Posix::clock_gettime(clockid, &tp);
4658 assert(rc == 0, "clock_gettime is expected to return 0 code");
4659
4660 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4661 }
4662
4663 void os::Linux::initialize_os_info() {
4664 assert(_os_version == 0, "OS info already initialized");
4665
4666 struct utsname _uname;
4667
4668 uint32_t major;
4669 uint32_t minor;
4670 uint32_t fix;
4671
4672 int rc;
4673
4674 // Kernel version is unknown if
4675 // verification below fails.
4676 _os_version = 0x01000000;
4677
4678 rc = uname(&_uname);
4679 if (rc != -1) {
4680
4681 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4682 if (rc == 3) {
4683
4684 if (major < 256 && minor < 256 && fix < 256) {
4685 // Kernel version format is as expected,
4686 // set it overriding unknown state.
4687 _os_version = (major << 16) |
4688 (minor << 8 ) |
4689 (fix << 0 ) ;
4690 }
4691 }
4692 }
4693 }
4694
4695 uint32_t os::Linux::os_version() {
4696 assert(_os_version != 0, "not initialized");
4697 return _os_version & 0x00FFFFFF;
4698 }
4699
4700 bool os::Linux::os_version_is_known() {
4701 assert(_os_version != 0, "not initialized");
4702 return _os_version & 0x01000000 ? false : true;
4703 }
4704
4705 /////
4706 // glibc on Linux platform uses non-documented flag
4707 // to indicate, that some special sort of signal
4708 // trampoline is used.
4709 // We will never set this flag, and we should
4710 // ignore this flag in our diagnostic
4711 #ifdef SIGNIFICANT_SIGNAL_MASK
4712 #undef SIGNIFICANT_SIGNAL_MASK
4713 #endif
4714 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4715
4716 static const char* get_signal_handler_name(address handler,
4717 char* buf, int buflen) {
4718 int offset = 0;
4719 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4720 if (found) {
4721 // skip directory names
4722 const char *p1, *p2;
4723 p1 = buf;
4724 size_t len = strlen(os::file_separator());
4725 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4726 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4727 } else {
4728 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4729 }
4730 return buf;
4731 }
4732
4733 static void print_signal_handler(outputStream* st, int sig,
4734 char* buf, size_t buflen) {
4735 struct sigaction sa;
4736
4737 sigaction(sig, NULL, &sa);
4738
4739 // See comment for SIGNIFICANT_SIGNAL_MASK define
4740 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4741
4742 st->print("%s: ", os::exception_name(sig, buf, buflen));
4743
4744 address handler = (sa.sa_flags & SA_SIGINFO)
4745 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4746 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4747
4748 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4749 st->print("SIG_DFL");
4750 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4751 st->print("SIG_IGN");
4752 } else {
4753 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4754 }
4755
4756 st->print(", sa_mask[0]=");
4757 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4758
4759 address rh = VMError::get_resetted_sighandler(sig);
4760 // May be, handler was resetted by VMError?
4761 if (rh != NULL) {
4762 handler = rh;
4763 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4764 }
4765
4766 st->print(", sa_flags=");
4767 os::Posix::print_sa_flags(st, sa.sa_flags);
4768
4769 // Check: is it our handler?
4770 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4771 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4772 // It is our signal handler
4773 // check for flags, reset system-used one!
4774 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4775 st->print(
4776 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4777 os::Linux::get_our_sigflags(sig));
4778 }
4779 }
4780 st->cr();
4781 }
4782
4783
4784 #define DO_SIGNAL_CHECK(sig) \
4785 do { \
4786 if (!sigismember(&check_signal_done, sig)) { \
4787 os::Linux::check_signal_handler(sig); \
4788 } \
4789 } while (0)
4790
4791 // This method is a periodic task to check for misbehaving JNI applications
4792 // under CheckJNI, we can add any periodic checks here
4793
4794 void os::run_periodic_checks() {
4795 if (check_signals == false) return;
4796
4797 // SEGV and BUS if overridden could potentially prevent
4798 // generation of hs*.log in the event of a crash, debugging
4799 // such a case can be very challenging, so we absolutely
4800 // check the following for a good measure:
4801 DO_SIGNAL_CHECK(SIGSEGV);
4802 DO_SIGNAL_CHECK(SIGILL);
4803 DO_SIGNAL_CHECK(SIGFPE);
4804 DO_SIGNAL_CHECK(SIGBUS);
4805 DO_SIGNAL_CHECK(SIGPIPE);
4806 DO_SIGNAL_CHECK(SIGXFSZ);
4807 #if defined(PPC64)
4808 DO_SIGNAL_CHECK(SIGTRAP);
4809 #endif
4810
4811 // ReduceSignalUsage allows the user to override these handlers
4812 // see comments at the very top and jvm_md.h
4813 if (!ReduceSignalUsage) {
4814 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4815 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4816 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4817 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4818 }
4819
4820 DO_SIGNAL_CHECK(SR_signum);
4821 }
4822
4823 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4824
4825 static os_sigaction_t os_sigaction = NULL;
4826
4827 void os::Linux::check_signal_handler(int sig) {
4828 char buf[O_BUFLEN];
4829 address jvmHandler = NULL;
4830
4831
4832 struct sigaction act;
4833 if (os_sigaction == NULL) {
4834 // only trust the default sigaction, in case it has been interposed
4835 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4836 if (os_sigaction == NULL) return;
4837 }
4838
4839 os_sigaction(sig, (struct sigaction*)NULL, &act);
4840
4841
4842 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4843
4844 address thisHandler = (act.sa_flags & SA_SIGINFO)
4845 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4846 : CAST_FROM_FN_PTR(address, act.sa_handler);
4847
4848
4849 switch (sig) {
4850 case SIGSEGV:
4851 case SIGBUS:
4852 case SIGFPE:
4853 case SIGPIPE:
4854 case SIGILL:
4855 case SIGXFSZ:
4856 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4857 break;
4858
4859 case SHUTDOWN1_SIGNAL:
4860 case SHUTDOWN2_SIGNAL:
4861 case SHUTDOWN3_SIGNAL:
4862 case BREAK_SIGNAL:
4863 jvmHandler = (address)user_handler();
4864 break;
4865
4866 default:
4867 if (sig == SR_signum) {
4868 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4869 } else {
4870 return;
4871 }
4872 break;
4873 }
4874
4875 if (thisHandler != jvmHandler) {
4876 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4877 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4878 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4879 // No need to check this sig any longer
4880 sigaddset(&check_signal_done, sig);
4881 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4882 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4883 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4884 exception_name(sig, buf, O_BUFLEN));
4885 }
4886 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4887 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4888 tty->print("expected:");
4889 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4890 tty->cr();
4891 tty->print(" found:");
4892 os::Posix::print_sa_flags(tty, act.sa_flags);
4893 tty->cr();
4894 // No need to check this sig any longer
4895 sigaddset(&check_signal_done, sig);
4896 }
4897
4898 // Dump all the signal
4899 if (sigismember(&check_signal_done, sig)) {
4900 print_signal_handlers(tty, buf, O_BUFLEN);
4901 }
4902 }
4903
4904 extern void report_error(char* file_name, int line_no, char* title,
4905 char* format, ...);
4906
4907 // this is called _before_ most of the global arguments have been parsed
4908 void os::init(void) {
4909 char dummy; // used to get a guess on initial stack address
4910
4911 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4912
4913 init_random(1234567);
4914
4915 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4916 if (Linux::page_size() == -1) {
4917 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4918 os::strerror(errno));
4919 }
4920 init_page_sizes((size_t) Linux::page_size());
4921
4922 Linux::initialize_system_info();
4923
4924 Linux::initialize_os_info();
4925
4926 // _main_thread points to the thread that created/loaded the JVM.
4927 Linux::_main_thread = pthread_self();
4928
4929 // retrieve entry point for pthread_setname_np
4930 Linux::_pthread_setname_np =
4931 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4932
4933 os::Posix::init();
4934
4935 initial_time_count = javaTimeNanos();
4936
4937 // Always warn if no monotonic clock available
4938 if (!os::Posix::supports_monotonic_clock()) {
4939 warning("No monotonic clock was available - timed services may " \
4940 "be adversely affected if the time-of-day clock changes");
4941 }
4942 }
4943
4944 // To install functions for atexit system call
4945 extern "C" {
4946 static void perfMemory_exit_helper() {
4947 perfMemory_exit();
4948 }
4949 }
4950
4951 void os::pd_init_container_support() {
4952 OSContainer::init();
4953 }
4954
4955 void os::Linux::numa_init() {
4956
4957 // Java can be invoked as
4958 // 1. Without numactl and heap will be allocated/configured on all nodes as
4959 // per the system policy.
4960 // 2. With numactl --interleave:
4961 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
4962 // API for membind case bitmask is reset.
4963 // Interleave is only hint and Kernel can fallback to other nodes if
4964 // no memory is available on the target nodes.
4965 // 3. With numactl --membind:
4966 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
4967 // interleave case returns bitmask of all nodes.
4968 // numa_all_nodes_ptr holds bitmask of all nodes.
4969 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
4970 // bitmask when externally configured to run on all or fewer nodes.
4971
4972 if (!Linux::libnuma_init()) {
4973 UseNUMA = false;
4974 } else {
4975 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
4976 // If there's only one node (they start from 0) or if the process
4977 // is bound explicitly to a single node using membind, disable NUMA.
4978 UseNUMA = false;
4979 } else {
4980
4981 LogTarget(Info,os) log;
4982 LogStream ls(log);
4983
4984 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
4985
4986 struct bitmask* bmp = Linux::_numa_membind_bitmask;
4987 const char* numa_mode = "membind";
4988
4989 if (Linux::is_running_in_interleave_mode()) {
4990 bmp = Linux::_numa_interleave_bitmask;
4991 numa_mode = "interleave";
4992 }
4993
4994 ls.print("UseNUMA is enabled and invoked in '%s' mode."
4995 " Heap will be configured using NUMA memory nodes:", numa_mode);
4996
4997 for (int node = 0; node <= Linux::numa_max_node(); node++) {
4998 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
4999 ls.print(" %d", node);
5000 }
5001 }
5002 }
5003 }
5004
5005 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5006 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5007 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5008 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5009 // and disable adaptive resizing.
5010 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5011 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5012 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5013 UseAdaptiveSizePolicy = false;
5014 UseAdaptiveNUMAChunkSizing = false;
5015 }
5016 }
5017
5018 if (!UseNUMA && ForceNUMA) {
5019 UseNUMA = true;
5020 }
5021 }
5022
5023 // this is called _after_ the global arguments have been parsed
5024 jint os::init_2(void) {
5025
5026 // This could be set after os::Posix::init() but all platforms
5027 // have to set it the same so we have to mirror Solaris.
5028 DEBUG_ONLY(os::set_mutex_init_done();)
5029
5030 os::Posix::init_2();
5031
5032 Linux::fast_thread_clock_init();
5033
5034 // initialize suspend/resume support - must do this before signal_sets_init()
5035 if (SR_initialize() != 0) {
5036 perror("SR_initialize failed");
5037 return JNI_ERR;
5038 }
5039
5040 Linux::signal_sets_init();
5041 Linux::install_signal_handlers();
5042 // Initialize data for jdk.internal.misc.Signal
5043 if (!ReduceSignalUsage) {
5044 jdk_misc_signal_init();
5045 }
5046
5047 // Check and sets minimum stack sizes against command line options
5048 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5049 return JNI_ERR;
5050 }
5051
5052 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5053 if (!suppress_primordial_thread_resolution) {
5054 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5055 }
5056
5057 #if defined(IA32)
5058 workaround_expand_exec_shield_cs_limit();
5059 #endif
5060
5061 Linux::libpthread_init();
5062 Linux::sched_getcpu_init();
5063 log_info(os)("HotSpot is running with %s, %s",
5064 Linux::glibc_version(), Linux::libpthread_version());
5065
5066 if (UseNUMA) {
5067 Linux::numa_init();
5068 }
5069
5070 if (MaxFDLimit) {
5071 // set the number of file descriptors to max. print out error
5072 // if getrlimit/setrlimit fails but continue regardless.
5073 struct rlimit nbr_files;
5074 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5075 if (status != 0) {
5076 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5077 } else {
5078 nbr_files.rlim_cur = nbr_files.rlim_max;
5079 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5080 if (status != 0) {
5081 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5082 }
5083 }
5084 }
5085
5086 // Initialize lock used to serialize thread creation (see os::create_thread)
5087 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5088
5089 // at-exit methods are called in the reverse order of their registration.
5090 // atexit functions are called on return from main or as a result of a
5091 // call to exit(3C). There can be only 32 of these functions registered
5092 // and atexit() does not set errno.
5093
5094 if (PerfAllowAtExitRegistration) {
5095 // only register atexit functions if PerfAllowAtExitRegistration is set.
5096 // atexit functions can be delayed until process exit time, which
5097 // can be problematic for embedded VM situations. Embedded VMs should
5098 // call DestroyJavaVM() to assure that VM resources are released.
5099
5100 // note: perfMemory_exit_helper atexit function may be removed in
5101 // the future if the appropriate cleanup code can be added to the
5102 // VM_Exit VMOperation's doit method.
5103 if (atexit(perfMemory_exit_helper) != 0) {
5104 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5105 }
5106 }
5107
5108 // initialize thread priority policy
5109 prio_init();
5110
5111 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5112 set_coredump_filter(DAX_SHARED_BIT);
5113 }
5114
5115 if (DumpPrivateMappingsInCore) {
5116 set_coredump_filter(FILE_BACKED_PVT_BIT);
5117 }
5118
5119 if (DumpSharedMappingsInCore) {
5120 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5121 }
5122
5123 return JNI_OK;
5124 }
5125
5126 // Mark the polling page as unreadable
5127 void os::make_polling_page_unreadable(void) {
5128 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5129 fatal("Could not disable polling page");
5130 }
5131 }
5132
5133 // Mark the polling page as readable
5134 void os::make_polling_page_readable(void) {
5135 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5136 fatal("Could not enable polling page");
5137 }
5138 }
5139
5140 // older glibc versions don't have this macro (which expands to
5141 // an optimized bit-counting function) so we have to roll our own
5142 #ifndef CPU_COUNT
5143
5144 static int _cpu_count(const cpu_set_t* cpus) {
5145 int count = 0;
5146 // only look up to the number of configured processors
5147 for (int i = 0; i < os::processor_count(); i++) {
5148 if (CPU_ISSET(i, cpus)) {
5149 count++;
5150 }
5151 }
5152 return count;
5153 }
5154
5155 #define CPU_COUNT(cpus) _cpu_count(cpus)
5156
5157 #endif // CPU_COUNT
5158
5159 // Get the current number of available processors for this process.
5160 // This value can change at any time during a process's lifetime.
5161 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5162 // If it appears there may be more than 1024 processors then we do a
5163 // dynamic check - see 6515172 for details.
5164 // If anything goes wrong we fallback to returning the number of online
5165 // processors - which can be greater than the number available to the process.
5166 int os::Linux::active_processor_count() {
5167 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5168 cpu_set_t* cpus_p = &cpus;
5169 int cpus_size = sizeof(cpu_set_t);
5170
5171 int configured_cpus = os::processor_count(); // upper bound on available cpus
5172 int cpu_count = 0;
5173
5174 // old build platforms may not support dynamic cpu sets
5175 #ifdef CPU_ALLOC
5176
5177 // To enable easy testing of the dynamic path on different platforms we
5178 // introduce a diagnostic flag: UseCpuAllocPath
5179 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5180 // kernel may use a mask bigger than cpu_set_t
5181 log_trace(os)("active_processor_count: using dynamic path %s"
5182 "- configured processors: %d",
5183 UseCpuAllocPath ? "(forced) " : "",
5184 configured_cpus);
5185 cpus_p = CPU_ALLOC(configured_cpus);
5186 if (cpus_p != NULL) {
5187 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5188 // zero it just to be safe
5189 CPU_ZERO_S(cpus_size, cpus_p);
5190 }
5191 else {
5192 // failed to allocate so fallback to online cpus
5193 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5194 log_trace(os)("active_processor_count: "
5195 "CPU_ALLOC failed (%s) - using "
5196 "online processor count: %d",
5197 os::strerror(errno), online_cpus);
5198 return online_cpus;
5199 }
5200 }
5201 else {
5202 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5203 configured_cpus);
5204 }
5205 #else // CPU_ALLOC
5206 // these stubs won't be executed
5207 #define CPU_COUNT_S(size, cpus) -1
5208 #define CPU_FREE(cpus)
5209
5210 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5211 configured_cpus);
5212 #endif // CPU_ALLOC
5213
5214 // pid 0 means the current thread - which we have to assume represents the process
5215 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5216 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5217 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5218 }
5219 else {
5220 cpu_count = CPU_COUNT(cpus_p);
5221 }
5222 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5223 }
5224 else {
5225 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5226 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5227 "which may exceed available processors", os::strerror(errno), cpu_count);
5228 }
5229
5230 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5231 CPU_FREE(cpus_p);
5232 }
5233
5234 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5235 return cpu_count;
5236 }
5237
5238 // Determine the active processor count from one of
5239 // three different sources:
5240 //
5241 // 1. User option -XX:ActiveProcessorCount
5242 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5243 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5244 //
5245 // Option 1, if specified, will always override.
5246 // If the cgroup subsystem is active and configured, we
5247 // will return the min of the cgroup and option 2 results.
5248 // This is required since tools, such as numactl, that
5249 // alter cpu affinity do not update cgroup subsystem
5250 // cpuset configuration files.
5251 int os::active_processor_count() {
5252 // User has overridden the number of active processors
5253 if (ActiveProcessorCount > 0) {
5254 log_trace(os)("active_processor_count: "
5255 "active processor count set by user : %d",
5256 ActiveProcessorCount);
5257 return ActiveProcessorCount;
5258 }
5259
5260 int active_cpus;
5261 if (OSContainer::is_containerized()) {
5262 active_cpus = OSContainer::active_processor_count();
5263 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5264 active_cpus);
5265 } else {
5266 active_cpus = os::Linux::active_processor_count();
5267 }
5268
5269 return active_cpus;
5270 }
5271
5272 uint os::processor_id() {
5273 const int id = Linux::sched_getcpu();
5274 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5275 return (uint)id;
5276 }
5277
5278 void os::set_native_thread_name(const char *name) {
5279 if (Linux::_pthread_setname_np) {
5280 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5281 snprintf(buf, sizeof(buf), "%s", name);
5282 buf[sizeof(buf) - 1] = '\0';
5283 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5284 // ERANGE should not happen; all other errors should just be ignored.
5285 assert(rc != ERANGE, "pthread_setname_np failed");
5286 }
5287 }
5288
5289 bool os::distribute_processes(uint length, uint* distribution) {
5290 // Not yet implemented.
5291 return false;
5292 }
5293
5294 bool os::bind_to_processor(uint processor_id) {
5295 // Not yet implemented.
5296 return false;
5297 }
5298
5299 ///
5300
5301 void os::SuspendedThreadTask::internal_do_task() {
5302 if (do_suspend(_thread->osthread())) {
5303 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5304 do_task(context);
5305 do_resume(_thread->osthread());
5306 }
5307 }
5308
5309 ////////////////////////////////////////////////////////////////////////////////
5310 // debug support
5311
5312 bool os::find(address addr, outputStream* st) {
5313 Dl_info dlinfo;
5314 memset(&dlinfo, 0, sizeof(dlinfo));
5315 if (dladdr(addr, &dlinfo) != 0) {
5316 st->print(PTR_FORMAT ": ", p2i(addr));
5317 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5318 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5319 p2i(addr) - p2i(dlinfo.dli_saddr));
5320 } else if (dlinfo.dli_fbase != NULL) {
5321 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5322 } else {
5323 st->print("<absolute address>");
5324 }
5325 if (dlinfo.dli_fname != NULL) {
5326 st->print(" in %s", dlinfo.dli_fname);
5327 }
5328 if (dlinfo.dli_fbase != NULL) {
5329 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5330 }
5331 st->cr();
5332
5333 if (Verbose) {
5334 // decode some bytes around the PC
5335 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5336 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5337 address lowest = (address) dlinfo.dli_sname;
5338 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5339 if (begin < lowest) begin = lowest;
5340 Dl_info dlinfo2;
5341 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5342 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5343 end = (address) dlinfo2.dli_saddr;
5344 }
5345 Disassembler::decode(begin, end, st);
5346 }
5347 return true;
5348 }
5349 return false;
5350 }
5351
5352 ////////////////////////////////////////////////////////////////////////////////
5353 // misc
5354
5355 // This does not do anything on Linux. This is basically a hook for being
5356 // able to use structured exception handling (thread-local exception filters)
5357 // on, e.g., Win32.
5358 void
5359 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5360 JavaCallArguments* args, Thread* thread) {
5361 f(value, method, args, thread);
5362 }
5363
5364 void os::print_statistics() {
5365 }
5366
5367 bool os::message_box(const char* title, const char* message) {
5368 int i;
5369 fdStream err(defaultStream::error_fd());
5370 for (i = 0; i < 78; i++) err.print_raw("=");
5371 err.cr();
5372 err.print_raw_cr(title);
5373 for (i = 0; i < 78; i++) err.print_raw("-");
5374 err.cr();
5375 err.print_raw_cr(message);
5376 for (i = 0; i < 78; i++) err.print_raw("=");
5377 err.cr();
5378
5379 char buf[16];
5380 // Prevent process from exiting upon "read error" without consuming all CPU
5381 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5382
5383 return buf[0] == 'y' || buf[0] == 'Y';
5384 }
5385
5386 // Is a (classpath) directory empty?
5387 bool os::dir_is_empty(const char* path) {
5388 DIR *dir = NULL;
5389 struct dirent *ptr;
5390
5391 dir = opendir(path);
5392 if (dir == NULL) return true;
5393
5394 // Scan the directory
5395 bool result = true;
5396 while (result && (ptr = readdir(dir)) != NULL) {
5397 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5398 result = false;
5399 }
5400 }
5401 closedir(dir);
5402 return result;
5403 }
5404
5405 // This code originates from JDK's sysOpen and open64_w
5406 // from src/solaris/hpi/src/system_md.c
5407
5408 int os::open(const char *path, int oflag, int mode) {
5409 if (strlen(path) > MAX_PATH - 1) {
5410 errno = ENAMETOOLONG;
5411 return -1;
5412 }
5413
5414 // All file descriptors that are opened in the Java process and not
5415 // specifically destined for a subprocess should have the close-on-exec
5416 // flag set. If we don't set it, then careless 3rd party native code
5417 // might fork and exec without closing all appropriate file descriptors
5418 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5419 // turn might:
5420 //
5421 // - cause end-of-file to fail to be detected on some file
5422 // descriptors, resulting in mysterious hangs, or
5423 //
5424 // - might cause an fopen in the subprocess to fail on a system
5425 // suffering from bug 1085341.
5426 //
5427 // (Yes, the default setting of the close-on-exec flag is a Unix
5428 // design flaw)
5429 //
5430 // See:
5431 // 1085341: 32-bit stdio routines should support file descriptors >255
5432 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5433 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5434 //
5435 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5436 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5437 // because it saves a system call and removes a small window where the flag
5438 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5439 // and we fall back to using FD_CLOEXEC (see below).
5440 #ifdef O_CLOEXEC
5441 oflag |= O_CLOEXEC;
5442 #endif
5443
5444 int fd = ::open64(path, oflag, mode);
5445 if (fd == -1) return -1;
5446
5447 //If the open succeeded, the file might still be a directory
5448 {
5449 struct stat64 buf64;
5450 int ret = ::fstat64(fd, &buf64);
5451 int st_mode = buf64.st_mode;
5452
5453 if (ret != -1) {
5454 if ((st_mode & S_IFMT) == S_IFDIR) {
5455 errno = EISDIR;
5456 ::close(fd);
5457 return -1;
5458 }
5459 } else {
5460 ::close(fd);
5461 return -1;
5462 }
5463 }
5464
5465 #ifdef FD_CLOEXEC
5466 // Validate that the use of the O_CLOEXEC flag on open above worked.
5467 // With recent kernels, we will perform this check exactly once.
5468 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5469 if (!O_CLOEXEC_is_known_to_work) {
5470 int flags = ::fcntl(fd, F_GETFD);
5471 if (flags != -1) {
5472 if ((flags & FD_CLOEXEC) != 0)
5473 O_CLOEXEC_is_known_to_work = 1;
5474 else
5475 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5476 }
5477 }
5478 #endif
5479
5480 return fd;
5481 }
5482
5483
5484 // create binary file, rewriting existing file if required
5485 int os::create_binary_file(const char* path, bool rewrite_existing) {
5486 int oflags = O_WRONLY | O_CREAT;
5487 if (!rewrite_existing) {
5488 oflags |= O_EXCL;
5489 }
5490 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5491 }
5492
5493 // return current position of file pointer
5494 jlong os::current_file_offset(int fd) {
5495 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5496 }
5497
5498 // move file pointer to the specified offset
5499 jlong os::seek_to_file_offset(int fd, jlong offset) {
5500 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5501 }
5502
5503 // This code originates from JDK's sysAvailable
5504 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5505
5506 int os::available(int fd, jlong *bytes) {
5507 jlong cur, end;
5508 int mode;
5509 struct stat64 buf64;
5510
5511 if (::fstat64(fd, &buf64) >= 0) {
5512 mode = buf64.st_mode;
5513 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5514 int n;
5515 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5516 *bytes = n;
5517 return 1;
5518 }
5519 }
5520 }
5521 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5522 return 0;
5523 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5524 return 0;
5525 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5526 return 0;
5527 }
5528 *bytes = end - cur;
5529 return 1;
5530 }
5531
5532 // Map a block of memory.
5533 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5534 char *addr, size_t bytes, bool read_only,
5535 bool allow_exec) {
5536 int prot;
5537 int flags = MAP_PRIVATE;
5538
5539 if (read_only) {
5540 prot = PROT_READ;
5541 } else {
5542 prot = PROT_READ | PROT_WRITE;
5543 }
5544
5545 if (allow_exec) {
5546 prot |= PROT_EXEC;
5547 }
5548
5549 if (addr != NULL) {
5550 flags |= MAP_FIXED;
5551 }
5552
5553 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5554 fd, file_offset);
5555 if (mapped_address == MAP_FAILED) {
5556 return NULL;
5557 }
5558 return mapped_address;
5559 }
5560
5561
5562 // Remap a block of memory.
5563 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5564 char *addr, size_t bytes, bool read_only,
5565 bool allow_exec) {
5566 // same as map_memory() on this OS
5567 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5568 allow_exec);
5569 }
5570
5571
5572 // Unmap a block of memory.
5573 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5574 return munmap(addr, bytes) == 0;
5575 }
5576
5577 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5578
5579 static jlong fast_cpu_time(Thread *thread) {
5580 clockid_t clockid;
5581 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5582 &clockid);
5583 if (rc == 0) {
5584 return os::Linux::fast_thread_cpu_time(clockid);
5585 } else {
5586 // It's possible to encounter a terminated native thread that failed
5587 // to detach itself from the VM - which should result in ESRCH.
5588 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5589 return -1;
5590 }
5591 }
5592
5593 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5594 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5595 // of a thread.
5596 //
5597 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5598 // the fast estimate available on the platform.
5599
5600 jlong os::current_thread_cpu_time() {
5601 if (os::Linux::supports_fast_thread_cpu_time()) {
5602 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5603 } else {
5604 // return user + sys since the cost is the same
5605 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5606 }
5607 }
5608
5609 jlong os::thread_cpu_time(Thread* thread) {
5610 // consistent with what current_thread_cpu_time() returns
5611 if (os::Linux::supports_fast_thread_cpu_time()) {
5612 return fast_cpu_time(thread);
5613 } else {
5614 return slow_thread_cpu_time(thread, true /* user + sys */);
5615 }
5616 }
5617
5618 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5619 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5620 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5621 } else {
5622 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5623 }
5624 }
5625
5626 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5627 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5628 return fast_cpu_time(thread);
5629 } else {
5630 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5631 }
5632 }
5633
5634 // -1 on error.
5635 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5636 pid_t tid = thread->osthread()->thread_id();
5637 char *s;
5638 char stat[2048];
5639 int statlen;
5640 char proc_name[64];
5641 int count;
5642 long sys_time, user_time;
5643 char cdummy;
5644 int idummy;
5645 long ldummy;
5646 FILE *fp;
5647
5648 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5649 fp = fopen(proc_name, "r");
5650 if (fp == NULL) return -1;
5651 statlen = fread(stat, 1, 2047, fp);
5652 stat[statlen] = '\0';
5653 fclose(fp);
5654
5655 // Skip pid and the command string. Note that we could be dealing with
5656 // weird command names, e.g. user could decide to rename java launcher
5657 // to "java 1.4.2 :)", then the stat file would look like
5658 // 1234 (java 1.4.2 :)) R ... ...
5659 // We don't really need to know the command string, just find the last
5660 // occurrence of ")" and then start parsing from there. See bug 4726580.
5661 s = strrchr(stat, ')');
5662 if (s == NULL) return -1;
5663
5664 // Skip blank chars
5665 do { s++; } while (s && isspace(*s));
5666
5667 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5668 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5669 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5670 &user_time, &sys_time);
5671 if (count != 13) return -1;
5672 if (user_sys_cpu_time) {
5673 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5674 } else {
5675 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5676 }
5677 }
5678
5679 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5680 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5681 info_ptr->may_skip_backward = false; // elapsed time not wall time
5682 info_ptr->may_skip_forward = false; // elapsed time not wall time
5683 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5684 }
5685
5686 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5687 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5688 info_ptr->may_skip_backward = false; // elapsed time not wall time
5689 info_ptr->may_skip_forward = false; // elapsed time not wall time
5690 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5691 }
5692
5693 bool os::is_thread_cpu_time_supported() {
5694 return true;
5695 }
5696
5697 // System loadavg support. Returns -1 if load average cannot be obtained.
5698 // Linux doesn't yet have a (official) notion of processor sets,
5699 // so just return the system wide load average.
5700 int os::loadavg(double loadavg[], int nelem) {
5701 return ::getloadavg(loadavg, nelem);
5702 }
5703
5704 void os::pause() {
5705 char filename[MAX_PATH];
5706 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5707 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5708 } else {
5709 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5710 }
5711
5712 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5713 if (fd != -1) {
5714 struct stat buf;
5715 ::close(fd);
5716 while (::stat(filename, &buf) == 0) {
5717 (void)::poll(NULL, 0, 100);
5718 }
5719 } else {
5720 jio_fprintf(stderr,
5721 "Could not open pause file '%s', continuing immediately.\n", filename);
5722 }
5723 }
5724
5725 extern char** environ;
5726
5727 // Run the specified command in a separate process. Return its exit value,
5728 // or -1 on failure (e.g. can't fork a new process).
5729 // Unlike system(), this function can be called from signal handler. It
5730 // doesn't block SIGINT et al.
5731 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5732 const char * argv[4] = {"sh", "-c", cmd, NULL};
5733
5734 pid_t pid ;
5735
5736 if (use_vfork_if_available) {
5737 pid = vfork();
5738 } else {
5739 pid = fork();
5740 }
5741
5742 if (pid < 0) {
5743 // fork failed
5744 return -1;
5745
5746 } else if (pid == 0) {
5747 // child process
5748
5749 execve("/bin/sh", (char* const*)argv, environ);
5750
5751 // execve failed
5752 _exit(-1);
5753
5754 } else {
5755 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5756 // care about the actual exit code, for now.
5757
5758 int status;
5759
5760 // Wait for the child process to exit. This returns immediately if
5761 // the child has already exited. */
5762 while (waitpid(pid, &status, 0) < 0) {
5763 switch (errno) {
5764 case ECHILD: return 0;
5765 case EINTR: break;
5766 default: return -1;
5767 }
5768 }
5769
5770 if (WIFEXITED(status)) {
5771 // The child exited normally; get its exit code.
5772 return WEXITSTATUS(status);
5773 } else if (WIFSIGNALED(status)) {
5774 // The child exited because of a signal
5775 // The best value to return is 0x80 + signal number,
5776 // because that is what all Unix shells do, and because
5777 // it allows callers to distinguish between process exit and
5778 // process death by signal.
5779 return 0x80 + WTERMSIG(status);
5780 } else {
5781 // Unknown exit code; pass it through
5782 return status;
5783 }
5784 }
5785 }
5786
5787 // Get the default path to the core file
5788 // Returns the length of the string
5789 int os::get_core_path(char* buffer, size_t bufferSize) {
5790 /*
5791 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5792 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5793 */
5794 const int core_pattern_len = 129;
5795 char core_pattern[core_pattern_len] = {0};
5796
5797 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5798 if (core_pattern_file == -1) {
5799 return -1;
5800 }
5801
5802 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5803 ::close(core_pattern_file);
5804 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5805 return -1;
5806 }
5807 if (core_pattern[ret-1] == '\n') {
5808 core_pattern[ret-1] = '\0';
5809 } else {
5810 core_pattern[ret] = '\0';
5811 }
5812
5813 // Replace the %p in the core pattern with the process id. NOTE: we do this
5814 // only if the pattern doesn't start with "|", and we support only one %p in
5815 // the pattern.
5816 char *pid_pos = strstr(core_pattern, "%p");
5817 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5818 int written;
5819
5820 if (core_pattern[0] == '/') {
5821 if (pid_pos != NULL) {
5822 *pid_pos = '\0';
5823 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5824 current_process_id(), tail);
5825 } else {
5826 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5827 }
5828 } else {
5829 char cwd[PATH_MAX];
5830
5831 const char* p = get_current_directory(cwd, PATH_MAX);
5832 if (p == NULL) {
5833 return -1;
5834 }
5835
5836 if (core_pattern[0] == '|') {
5837 written = jio_snprintf(buffer, bufferSize,
5838 "\"%s\" (or dumping to %s/core.%d)",
5839 &core_pattern[1], p, current_process_id());
5840 } else if (pid_pos != NULL) {
5841 *pid_pos = '\0';
5842 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5843 current_process_id(), tail);
5844 } else {
5845 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5846 }
5847 }
5848
5849 if (written < 0) {
5850 return -1;
5851 }
5852
5853 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5854 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5855
5856 if (core_uses_pid_file != -1) {
5857 char core_uses_pid = 0;
5858 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5859 ::close(core_uses_pid_file);
5860
5861 if (core_uses_pid == '1') {
5862 jio_snprintf(buffer + written, bufferSize - written,
5863 ".%d", current_process_id());
5864 }
5865 }
5866 }
5867
5868 return strlen(buffer);
5869 }
5870
5871 bool os::start_debugging(char *buf, int buflen) {
5872 int len = (int)strlen(buf);
5873 char *p = &buf[len];
5874
5875 jio_snprintf(p, buflen-len,
5876 "\n\n"
5877 "Do you want to debug the problem?\n\n"
5878 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5879 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5880 "Otherwise, press RETURN to abort...",
5881 os::current_process_id(), os::current_process_id(),
5882 os::current_thread_id(), os::current_thread_id());
5883
5884 bool yes = os::message_box("Unexpected Error", buf);
5885
5886 if (yes) {
5887 // yes, user asked VM to launch debugger
5888 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5889 os::current_process_id(), os::current_process_id());
5890
5891 os::fork_and_exec(buf);
5892 yes = false;
5893 }
5894 return yes;
5895 }
5896
5897
5898 // Java/Compiler thread:
5899 //
5900 // Low memory addresses
5901 // P0 +------------------------+
5902 // | |\ Java thread created by VM does not have glibc
5903 // | glibc guard page | - guard page, attached Java thread usually has
5904 // | |/ 1 glibc guard page.
5905 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5906 // | |\
5907 // | HotSpot Guard Pages | - red, yellow and reserved pages
5908 // | |/
5909 // +------------------------+ JavaThread::stack_reserved_zone_base()
5910 // | |\
5911 // | Normal Stack | -
5912 // | |/
5913 // P2 +------------------------+ Thread::stack_base()
5914 //
5915 // Non-Java thread:
5916 //
5917 // Low memory addresses
5918 // P0 +------------------------+
5919 // | |\
5920 // | glibc guard page | - usually 1 page
5921 // | |/
5922 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5923 // | |\
5924 // | Normal Stack | -
5925 // | |/
5926 // P2 +------------------------+ Thread::stack_base()
5927 //
5928 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5929 // returned from pthread_attr_getstack().
5930 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5931 // of the stack size given in pthread_attr. We work around this for
5932 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5933 //
5934 #ifndef ZERO
5935 static void current_stack_region(address * bottom, size_t * size) {
5936 if (os::is_primordial_thread()) {
5937 // primordial thread needs special handling because pthread_getattr_np()
5938 // may return bogus value.
5939 *bottom = os::Linux::initial_thread_stack_bottom();
5940 *size = os::Linux::initial_thread_stack_size();
5941 } else {
5942 pthread_attr_t attr;
5943
5944 int rslt = pthread_getattr_np(pthread_self(), &attr);
5945
5946 // JVM needs to know exact stack location, abort if it fails
5947 if (rslt != 0) {
5948 if (rslt == ENOMEM) {
5949 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5950 } else {
5951 fatal("pthread_getattr_np failed with error = %d", rslt);
5952 }
5953 }
5954
5955 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5956 fatal("Cannot locate current stack attributes!");
5957 }
5958
5959 // Work around NPTL stack guard error.
5960 size_t guard_size = 0;
5961 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5962 if (rslt != 0) {
5963 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5964 }
5965 *bottom += guard_size;
5966 *size -= guard_size;
5967
5968 pthread_attr_destroy(&attr);
5969
5970 }
5971 assert(os::current_stack_pointer() >= *bottom &&
5972 os::current_stack_pointer() < *bottom + *size, "just checking");
5973 }
5974
5975 address os::current_stack_base() {
5976 address bottom;
5977 size_t size;
5978 current_stack_region(&bottom, &size);
5979 return (bottom + size);
5980 }
5981
5982 size_t os::current_stack_size() {
5983 // This stack size includes the usable stack and HotSpot guard pages
5984 // (for the threads that have Hotspot guard pages).
5985 address bottom;
5986 size_t size;
5987 current_stack_region(&bottom, &size);
5988 return size;
5989 }
5990 #endif
5991
5992 static inline struct timespec get_mtime(const char* filename) {
5993 struct stat st;
5994 int ret = os::stat(filename, &st);
5995 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
5996 return st.st_mtim;
5997 }
5998
5999 int os::compare_file_modified_times(const char* file1, const char* file2) {
6000 struct timespec filetime1 = get_mtime(file1);
6001 struct timespec filetime2 = get_mtime(file2);
6002 int diff = filetime1.tv_sec - filetime2.tv_sec;
6003 if (diff == 0) {
6004 return filetime1.tv_nsec - filetime2.tv_nsec;
6005 }
6006 return diff;
6007 }
6008
6009 /////////////// Unit tests ///////////////
6010
6011 #ifndef PRODUCT
6012
6013 class TestReserveMemorySpecial : AllStatic {
6014 public:
6015 static void small_page_write(void* addr, size_t size) {
6016 size_t page_size = os::vm_page_size();
6017
6018 char* end = (char*)addr + size;
6019 for (char* p = (char*)addr; p < end; p += page_size) {
6020 *p = 1;
6021 }
6022 }
6023
6024 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6025 if (!UseHugeTLBFS) {
6026 return;
6027 }
6028
6029 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6030
6031 if (addr != NULL) {
6032 small_page_write(addr, size);
6033
6034 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6035 }
6036 }
6037
6038 static void test_reserve_memory_special_huge_tlbfs_only() {
6039 if (!UseHugeTLBFS) {
6040 return;
6041 }
6042
6043 size_t lp = os::large_page_size();
6044
6045 for (size_t size = lp; size <= lp * 10; size += lp) {
6046 test_reserve_memory_special_huge_tlbfs_only(size);
6047 }
6048 }
6049
6050 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6051 size_t lp = os::large_page_size();
6052 size_t ag = os::vm_allocation_granularity();
6053
6054 // sizes to test
6055 const size_t sizes[] = {
6056 lp, lp + ag, lp + lp / 2, lp * 2,
6057 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6058 lp * 10, lp * 10 + lp / 2
6059 };
6060 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6061
6062 // For each size/alignment combination, we test three scenarios:
6063 // 1) with req_addr == NULL
6064 // 2) with a non-null req_addr at which we expect to successfully allocate
6065 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6066 // expect the allocation to either fail or to ignore req_addr
6067
6068 // Pre-allocate two areas; they shall be as large as the largest allocation
6069 // and aligned to the largest alignment we will be testing.
6070 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6071 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6072 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6073 -1, 0);
6074 assert(mapping1 != MAP_FAILED, "should work");
6075
6076 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6077 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6078 -1, 0);
6079 assert(mapping2 != MAP_FAILED, "should work");
6080
6081 // Unmap the first mapping, but leave the second mapping intact: the first
6082 // mapping will serve as a value for a "good" req_addr (case 2). The second
6083 // mapping, still intact, as "bad" req_addr (case 3).
6084 ::munmap(mapping1, mapping_size);
6085
6086 // Case 1
6087 for (int i = 0; i < num_sizes; i++) {
6088 const size_t size = sizes[i];
6089 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6090 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6091 if (p != NULL) {
6092 assert(is_aligned(p, alignment), "must be");
6093 small_page_write(p, size);
6094 os::Linux::release_memory_special_huge_tlbfs(p, size);
6095 }
6096 }
6097 }
6098
6099 // Case 2
6100 for (int i = 0; i < num_sizes; i++) {
6101 const size_t size = sizes[i];
6102 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6103 char* const req_addr = align_up(mapping1, alignment);
6104 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6105 if (p != NULL) {
6106 assert(p == req_addr, "must be");
6107 small_page_write(p, size);
6108 os::Linux::release_memory_special_huge_tlbfs(p, size);
6109 }
6110 }
6111 }
6112
6113 // Case 3
6114 for (int i = 0; i < num_sizes; i++) {
6115 const size_t size = sizes[i];
6116 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6117 char* const req_addr = align_up(mapping2, alignment);
6118 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6119 // as the area around req_addr contains already existing mappings, the API should always
6120 // return NULL (as per contract, it cannot return another address)
6121 assert(p == NULL, "must be");
6122 }
6123 }
6124
6125 ::munmap(mapping2, mapping_size);
6126
6127 }
6128
6129 static void test_reserve_memory_special_huge_tlbfs() {
6130 if (!UseHugeTLBFS) {
6131 return;
6132 }
6133
6134 test_reserve_memory_special_huge_tlbfs_only();
6135 test_reserve_memory_special_huge_tlbfs_mixed();
6136 }
6137
6138 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6139 if (!UseSHM) {
6140 return;
6141 }
6142
6143 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6144
6145 if (addr != NULL) {
6146 assert(is_aligned(addr, alignment), "Check");
6147 assert(is_aligned(addr, os::large_page_size()), "Check");
6148
6149 small_page_write(addr, size);
6150
6151 os::Linux::release_memory_special_shm(addr, size);
6152 }
6153 }
6154
6155 static void test_reserve_memory_special_shm() {
6156 size_t lp = os::large_page_size();
6157 size_t ag = os::vm_allocation_granularity();
6158
6159 for (size_t size = ag; size < lp * 3; size += ag) {
6160 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6161 test_reserve_memory_special_shm(size, alignment);
6162 }
6163 }
6164 }
6165
6166 static void test() {
6167 test_reserve_memory_special_huge_tlbfs();
6168 test_reserve_memory_special_shm();
6169 }
6170 };
6171
6172 void TestReserveMemorySpecial_test() {
6173 TestReserveMemorySpecial::test();
6174 }
6175
6176 #endif