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_share_linux.hpp"
42 #include "osContainer_linux.hpp"
43 #include "prims/jniFastGetField.hpp"
44 #include "prims/jvm_misc.hpp"
45 #include "runtime/arguments.hpp"
46 #include "runtime/atomic.hpp"
47 #include "runtime/extendedPC.hpp"
48 #include "runtime/globals.hpp"
49 #include "runtime/interfaceSupport.inline.hpp"
50 #include "runtime/init.hpp"
51 #include "runtime/java.hpp"
52 #include "runtime/javaCalls.hpp"
53 #include "runtime/mutexLocker.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/orderAccess.hpp"
56 #include "runtime/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/threadSMR.hpp"
64 #include "runtime/timer.hpp"
65 #include "runtime/vm_version.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::_clock_gettime)(clockid_t, struct timespec *) = NULL;
148 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
149 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
150 Mutex* os::Linux::_createThread_lock = NULL;
151 pthread_t os::Linux::_main_thread;
152 int os::Linux::_page_size = -1;
153 bool os::Linux::_supports_fast_thread_cpu_time = false;
154 uint32_t os::Linux::_os_version = 0;
155 const char * os::Linux::_glibc_version = NULL;
156 const char * os::Linux::_libpthread_version = NULL;
157
158 static jlong initial_time_count=0;
159
160 static int clock_tics_per_sec = 100;
161
162 // If the VM might have been created on the primordial thread, we need to resolve the
163 // primordial thread stack bounds and check if the current thread might be the
164 // primordial thread in places. If we know that the primordial thread is never used,
165 // such as when the VM was created by one of the standard java launchers, we can
166 // avoid this
167 static bool suppress_primordial_thread_resolution = false;
168
169 // For diagnostics to print a message once. see run_periodic_checks
170 static sigset_t check_signal_done;
171 static bool check_signals = true;
172
173 // Signal number used to suspend/resume a thread
174
175 // do not use any signal number less than SIGSEGV, see 4355769
176 static int SR_signum = SIGUSR2;
177 sigset_t SR_sigset;
178
179 // utility functions
180
181 static int SR_initialize();
182
183 julong os::available_memory() {
184 return Linux::available_memory();
185 }
186
187 julong os::Linux::available_memory() {
188 // values in struct sysinfo are "unsigned long"
189 struct sysinfo si;
190 julong avail_mem;
191
192 if (OSContainer::is_containerized()) {
193 jlong mem_limit, mem_usage;
194 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
195 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
196 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
197 }
198 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
199 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
200 }
201 if (mem_limit > 0 && mem_usage > 0 ) {
202 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
203 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
204 return avail_mem;
205 }
206 }
207
208 sysinfo(&si);
209 avail_mem = (julong)si.freeram * si.mem_unit;
210 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
211 return avail_mem;
212 }
213
214 julong os::physical_memory() {
215 jlong phys_mem = 0;
216 if (OSContainer::is_containerized()) {
217 jlong mem_limit;
218 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
219 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
220 return mem_limit;
221 }
222 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
223 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
224 }
225
226 phys_mem = Linux::physical_memory();
227 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
228 return phys_mem;
229 }
230
231 static uint64_t initial_total_ticks = 0;
232 static uint64_t initial_steal_ticks = 0;
233 static bool has_initial_tick_info = false;
234
235 static void next_line(FILE *f) {
236 int c;
237 do {
238 c = fgetc(f);
239 } while (c != '\n' && c != EOF);
240 }
241
242 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
243 FILE* fh;
244 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
245 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
246 // irq: time servicing interrupts; softirq: time servicing softirqs
247 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
248 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
249 uint64_t stealTicks = 0;
250 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
251 // control of the Linux kernel
252 uint64_t guestNiceTicks = 0;
253 int logical_cpu = -1;
254 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
255 int n;
256
257 memset(pticks, 0, sizeof(CPUPerfTicks));
258
259 if ((fh = fopen("/proc/stat", "r")) == NULL) {
260 return false;
261 }
262
263 if (which_logical_cpu == -1) {
264 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
265 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
266 UINT64_FORMAT " " UINT64_FORMAT " ",
267 &userTicks, &niceTicks, &systemTicks, &idleTicks,
268 &iowTicks, &irqTicks, &sirqTicks,
269 &stealTicks, &guestNiceTicks);
270 } else {
271 // Move to next line
272 next_line(fh);
273
274 // find the line for requested cpu faster to just iterate linefeeds?
275 for (int i = 0; i < which_logical_cpu; i++) {
276 next_line(fh);
277 }
278
279 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
280 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
281 UINT64_FORMAT " " UINT64_FORMAT " ",
282 &logical_cpu, &userTicks, &niceTicks,
283 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
284 &stealTicks, &guestNiceTicks);
285 }
286
287 fclose(fh);
288 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
289 return false;
290 }
291 pticks->used = userTicks + niceTicks;
292 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
293 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
294 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
295
296 if (n > required_tickinfo_count + 3) {
297 pticks->steal = stealTicks;
298 pticks->has_steal_ticks = true;
299 } else {
300 pticks->steal = 0;
301 pticks->has_steal_ticks = false;
302 }
303
304 return true;
305 }
306
307 // Return true if user is running as root.
308
309 bool os::have_special_privileges() {
310 static bool init = false;
311 static bool privileges = false;
312 if (!init) {
313 privileges = (getuid() != geteuid()) || (getgid() != getegid());
314 init = true;
315 }
316 return privileges;
317 }
318
319
320 #ifndef SYS_gettid
321 // i386: 224, ia64: 1105, amd64: 186, sparc 143
322 #ifdef __ia64__
323 #define SYS_gettid 1105
324 #else
325 #ifdef __i386__
326 #define SYS_gettid 224
327 #else
328 #ifdef __amd64__
329 #define SYS_gettid 186
330 #else
331 #ifdef __sparc__
332 #define SYS_gettid 143
333 #else
334 #error define gettid for the arch
335 #endif
336 #endif
337 #endif
338 #endif
339 #endif
340
341
342 // pid_t gettid()
343 //
344 // Returns the kernel thread id of the currently running thread. Kernel
345 // thread id is used to access /proc.
346 pid_t os::Linux::gettid() {
347 int rslt = syscall(SYS_gettid);
348 assert(rslt != -1, "must be."); // old linuxthreads implementation?
349 return (pid_t)rslt;
350 }
351
352 // Most versions of linux have a bug where the number of processors are
353 // determined by looking at the /proc file system. In a chroot environment,
354 // the system call returns 1. This causes the VM to act as if it is
355 // a single processor and elide locking (see is_MP() call).
356 static bool unsafe_chroot_detected = false;
357 static const char *unstable_chroot_error = "/proc file system not found.\n"
358 "Java may be unstable running multithreaded in a chroot "
359 "environment on Linux when /proc filesystem is not mounted.";
360
361 void os::Linux::initialize_system_info() {
362 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
363 if (processor_count() == 1) {
364 pid_t pid = os::Linux::gettid();
365 char fname[32];
366 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
367 FILE *fp = fopen(fname, "r");
368 if (fp == NULL) {
369 unsafe_chroot_detected = true;
370 } else {
371 fclose(fp);
372 }
373 }
374 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
375 assert(processor_count() > 0, "linux error");
376 }
377
378 void os::init_system_properties_values() {
379 // The next steps are taken in the product version:
380 //
381 // Obtain the JAVA_HOME value from the location of libjvm.so.
382 // This library should be located at:
383 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
384 //
385 // If "/jre/lib/" appears at the right place in the path, then we
386 // assume libjvm.so is installed in a JDK and we use this path.
387 //
388 // Otherwise exit with message: "Could not create the Java virtual machine."
389 //
390 // The following extra steps are taken in the debugging version:
391 //
392 // If "/jre/lib/" does NOT appear at the right place in the path
393 // instead of exit check for $JAVA_HOME environment variable.
394 //
395 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
396 // then we append a fake suffix "hotspot/libjvm.so" to this path so
397 // it looks like libjvm.so is installed there
398 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
399 //
400 // Otherwise exit.
401 //
402 // Important note: if the location of libjvm.so changes this
403 // code needs to be changed accordingly.
404
405 // See ld(1):
406 // The linker uses the following search paths to locate required
407 // shared libraries:
408 // 1: ...
409 // ...
410 // 7: The default directories, normally /lib and /usr/lib.
411 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
412 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
413 #else
414 #define DEFAULT_LIBPATH "/lib:/usr/lib"
415 #endif
416
417 // Base path of extensions installed on the system.
418 #define SYS_EXT_DIR "/usr/java/packages"
419 #define EXTENSIONS_DIR "/lib/ext"
420
421 // Buffer that fits several sprintfs.
422 // Note that the space for the colon and the trailing null are provided
423 // by the nulls included by the sizeof operator.
424 const size_t bufsize =
425 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
426 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
427 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
428
429 // sysclasspath, java_home, dll_dir
430 {
431 char *pslash;
432 os::jvm_path(buf, bufsize);
433
434 // Found the full path to libjvm.so.
435 // Now cut the path to <java_home>/jre if we can.
436 pslash = strrchr(buf, '/');
437 if (pslash != NULL) {
438 *pslash = '\0'; // Get rid of /libjvm.so.
439 }
440 pslash = strrchr(buf, '/');
441 if (pslash != NULL) {
442 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
443 }
444 Arguments::set_dll_dir(buf);
445
446 if (pslash != NULL) {
447 pslash = strrchr(buf, '/');
448 if (pslash != NULL) {
449 *pslash = '\0'; // Get rid of /lib.
450 }
451 }
452 Arguments::set_java_home(buf);
453 set_boot_path('/', ':');
454 }
455
456 // Where to look for native libraries.
457 //
458 // Note: Due to a legacy implementation, most of the library path
459 // is set in the launcher. This was to accomodate linking restrictions
460 // on legacy Linux implementations (which are no longer supported).
461 // Eventually, all the library path setting will be done here.
462 //
463 // However, to prevent the proliferation of improperly built native
464 // libraries, the new path component /usr/java/packages is added here.
465 // Eventually, all the library path setting will be done here.
466 {
467 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
468 // should always exist (until the legacy problem cited above is
469 // addressed).
470 const char *v = ::getenv("LD_LIBRARY_PATH");
471 const char *v_colon = ":";
472 if (v == NULL) { v = ""; v_colon = ""; }
473 // That's +1 for the colon and +1 for the trailing '\0'.
474 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
475 strlen(v) + 1 +
476 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
477 mtInternal);
478 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
479 Arguments::set_library_path(ld_library_path);
480 FREE_C_HEAP_ARRAY(char, ld_library_path);
481 }
482
483 // Extensions directories.
484 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
485 Arguments::set_ext_dirs(buf);
486
487 FREE_C_HEAP_ARRAY(char, buf);
488
489 #undef DEFAULT_LIBPATH
490 #undef SYS_EXT_DIR
491 #undef EXTENSIONS_DIR
492 }
493
494 ////////////////////////////////////////////////////////////////////////////////
495 // breakpoint support
496
497 void os::breakpoint() {
498 BREAKPOINT;
499 }
500
501 extern "C" void breakpoint() {
502 // use debugger to set breakpoint here
503 }
504
505 ////////////////////////////////////////////////////////////////////////////////
506 // signal support
507
508 debug_only(static bool signal_sets_initialized = false);
509 static sigset_t unblocked_sigs, vm_sigs;
510
511 void os::Linux::signal_sets_init() {
512 // Should also have an assertion stating we are still single-threaded.
513 assert(!signal_sets_initialized, "Already initialized");
514 // Fill in signals that are necessarily unblocked for all threads in
515 // the VM. Currently, we unblock the following signals:
516 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
517 // by -Xrs (=ReduceSignalUsage));
518 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
519 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
520 // the dispositions or masks wrt these signals.
521 // Programs embedding the VM that want to use the above signals for their
522 // own purposes must, at this time, use the "-Xrs" option to prevent
523 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
524 // (See bug 4345157, and other related bugs).
525 // In reality, though, unblocking these signals is really a nop, since
526 // these signals are not blocked by default.
527 sigemptyset(&unblocked_sigs);
528 sigaddset(&unblocked_sigs, SIGILL);
529 sigaddset(&unblocked_sigs, SIGSEGV);
530 sigaddset(&unblocked_sigs, SIGBUS);
531 sigaddset(&unblocked_sigs, SIGFPE);
532 #if defined(PPC64)
533 sigaddset(&unblocked_sigs, SIGTRAP);
534 #endif
535 sigaddset(&unblocked_sigs, SR_signum);
536
537 if (!ReduceSignalUsage) {
538 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
539 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
540 }
541 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
542 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
543 }
544 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
545 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
546 }
547 }
548 // Fill in signals that are blocked by all but the VM thread.
549 sigemptyset(&vm_sigs);
550 if (!ReduceSignalUsage) {
551 sigaddset(&vm_sigs, BREAK_SIGNAL);
552 }
553 debug_only(signal_sets_initialized = true);
554
555 }
556
557 // These are signals that are unblocked while a thread is running Java.
558 // (For some reason, they get blocked by default.)
559 sigset_t* os::Linux::unblocked_signals() {
560 assert(signal_sets_initialized, "Not initialized");
561 return &unblocked_sigs;
562 }
563
564 // These are the signals that are blocked while a (non-VM) thread is
565 // running Java. Only the VM thread handles these signals.
566 sigset_t* os::Linux::vm_signals() {
567 assert(signal_sets_initialized, "Not initialized");
568 return &vm_sigs;
569 }
570
571 void os::Linux::hotspot_sigmask(Thread* thread) {
572
573 //Save caller's signal mask before setting VM signal mask
574 sigset_t caller_sigmask;
575 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
576
577 OSThread* osthread = thread->osthread();
578 osthread->set_caller_sigmask(caller_sigmask);
579
580 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
581
582 if (!ReduceSignalUsage) {
583 if (thread->is_VM_thread()) {
584 // Only the VM thread handles BREAK_SIGNAL ...
585 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
586 } else {
587 // ... all other threads block BREAK_SIGNAL
588 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
589 }
590 }
591 }
592
593 //////////////////////////////////////////////////////////////////////////////
594 // detecting pthread library
595
596 void os::Linux::libpthread_init() {
597 // Save glibc and pthread version strings.
598 #if !defined(_CS_GNU_LIBC_VERSION) || \
599 !defined(_CS_GNU_LIBPTHREAD_VERSION)
600 #error "glibc too old (< 2.3.2)"
601 #endif
602
603 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
604 assert(n > 0, "cannot retrieve glibc version");
605 char *str = (char *)malloc(n, mtInternal);
606 confstr(_CS_GNU_LIBC_VERSION, str, n);
607 os::Linux::set_glibc_version(str);
608
609 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
610 assert(n > 0, "cannot retrieve pthread version");
611 str = (char *)malloc(n, mtInternal);
612 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
613 os::Linux::set_libpthread_version(str);
614 }
615
616 /////////////////////////////////////////////////////////////////////////////
617 // thread stack expansion
618
619 // os::Linux::manually_expand_stack() takes care of expanding the thread
620 // stack. Note that this is normally not needed: pthread stacks allocate
621 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
622 // committed. Therefore it is not necessary to expand the stack manually.
623 //
624 // Manually expanding the stack was historically needed on LinuxThreads
625 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
626 // it is kept to deal with very rare corner cases:
627 //
628 // For one, user may run the VM on an own implementation of threads
629 // whose stacks are - like the old LinuxThreads - implemented using
630 // mmap(MAP_GROWSDOWN).
631 //
632 // Also, this coding may be needed if the VM is running on the primordial
633 // thread. Normally we avoid running on the primordial thread; however,
634 // user may still invoke the VM on the primordial thread.
635 //
636 // The following historical comment describes the details about running
637 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
638
639
640 // Force Linux kernel to expand current thread stack. If "bottom" is close
641 // to the stack guard, caller should block all signals.
642 //
643 // MAP_GROWSDOWN:
644 // A special mmap() flag that is used to implement thread stacks. It tells
645 // kernel that the memory region should extend downwards when needed. This
646 // allows early versions of LinuxThreads to only mmap the first few pages
647 // when creating a new thread. Linux kernel will automatically expand thread
648 // stack as needed (on page faults).
649 //
650 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
651 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
652 // region, it's hard to tell if the fault is due to a legitimate stack
653 // access or because of reading/writing non-exist memory (e.g. buffer
654 // overrun). As a rule, if the fault happens below current stack pointer,
655 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
656 // application (see Linux kernel fault.c).
657 //
658 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
659 // stack overflow detection.
660 //
661 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
662 // not use MAP_GROWSDOWN.
663 //
664 // To get around the problem and allow stack banging on Linux, we need to
665 // manually expand thread stack after receiving the SIGSEGV.
666 //
667 // There are two ways to expand thread stack to address "bottom", we used
668 // both of them in JVM before 1.5:
669 // 1. adjust stack pointer first so that it is below "bottom", and then
670 // touch "bottom"
671 // 2. mmap() the page in question
672 //
673 // Now alternate signal stack is gone, it's harder to use 2. For instance,
674 // if current sp is already near the lower end of page 101, and we need to
675 // call mmap() to map page 100, it is possible that part of the mmap() frame
676 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
677 // That will destroy the mmap() frame and cause VM to crash.
678 //
679 // The following code works by adjusting sp first, then accessing the "bottom"
680 // page to force a page fault. Linux kernel will then automatically expand the
681 // stack mapping.
682 //
683 // _expand_stack_to() assumes its frame size is less than page size, which
684 // should always be true if the function is not inlined.
685
686 static void NOINLINE _expand_stack_to(address bottom) {
687 address sp;
688 size_t size;
689 volatile char *p;
690
691 // Adjust bottom to point to the largest address within the same page, it
692 // gives us a one-page buffer if alloca() allocates slightly more memory.
693 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
694 bottom += os::Linux::page_size() - 1;
695
696 // sp might be slightly above current stack pointer; if that's the case, we
697 // will alloca() a little more space than necessary, which is OK. Don't use
698 // os::current_stack_pointer(), as its result can be slightly below current
699 // stack pointer, causing us to not alloca enough to reach "bottom".
700 sp = (address)&sp;
701
702 if (sp > bottom) {
703 size = sp - bottom;
704 p = (volatile char *)alloca(size);
705 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
706 p[0] = '\0';
707 }
708 }
709
710 void os::Linux::expand_stack_to(address bottom) {
711 _expand_stack_to(bottom);
712 }
713
714 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
715 assert(t!=NULL, "just checking");
716 assert(t->osthread()->expanding_stack(), "expand should be set");
717 assert(t->stack_base() != NULL, "stack_base was not initialized");
718
719 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
720 sigset_t mask_all, old_sigset;
721 sigfillset(&mask_all);
722 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
723 _expand_stack_to(addr);
724 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
725 return true;
726 }
727 return false;
728 }
729
730 //////////////////////////////////////////////////////////////////////////////
731 // create new thread
732
733 // Thread start routine for all newly created threads
734 static void *thread_native_entry(Thread *thread) {
735
736 thread->record_stack_base_and_size();
737
738 // Try to randomize the cache line index of hot stack frames.
739 // This helps when threads of the same stack traces evict each other's
740 // cache lines. The threads can be either from the same JVM instance, or
741 // from different JVM instances. The benefit is especially true for
742 // processors with hyperthreading technology.
743 static int counter = 0;
744 int pid = os::current_process_id();
745 alloca(((pid ^ counter++) & 7) * 128);
746
747 thread->initialize_thread_current();
748
749 OSThread* osthread = thread->osthread();
750 Monitor* sync = osthread->startThread_lock();
751
752 osthread->set_thread_id(os::current_thread_id());
753
754 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
755 os::current_thread_id(), (uintx) pthread_self());
756
757 if (UseNUMA) {
758 int lgrp_id = os::numa_get_group_id();
759 if (lgrp_id != -1) {
760 thread->set_lgrp_id(lgrp_id);
761 }
762 }
763 // initialize signal mask for this thread
764 os::Linux::hotspot_sigmask(thread);
765
766 // initialize floating point control register
767 os::Linux::init_thread_fpu_state();
768
769 // handshaking with parent thread
770 {
771 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
772
773 // notify parent thread
774 osthread->set_state(INITIALIZED);
775 sync->notify_all();
776
777 // wait until os::start_thread()
778 while (osthread->get_state() == INITIALIZED) {
779 sync->wait(Mutex::_no_safepoint_check_flag);
780 }
781 }
782
783 // call one more level start routine
784 thread->call_run();
785
786 // Note: at this point the thread object may already have deleted itself.
787 // Prevent dereferencing it from here on out.
788 thread = NULL;
789
790 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
791 os::current_thread_id(), (uintx) pthread_self());
792
793 return 0;
794 }
795
796 bool os::create_thread(Thread* thread, ThreadType thr_type,
797 size_t req_stack_size) {
798 assert(thread->osthread() == NULL, "caller responsible");
799
800 // Allocate the OSThread object
801 OSThread* osthread = new OSThread(NULL, NULL);
802 if (osthread == NULL) {
803 return false;
804 }
805
806 // set the correct thread state
807 osthread->set_thread_type(thr_type);
808
809 // Initial state is ALLOCATED but not INITIALIZED
810 osthread->set_state(ALLOCATED);
811
812 thread->set_osthread(osthread);
813
814 // init thread attributes
815 pthread_attr_t attr;
816 pthread_attr_init(&attr);
817 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
818
819 // Calculate stack size if it's not specified by caller.
820 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
821 // In the Linux NPTL pthread implementation the guard size mechanism
822 // is not implemented properly. The posix standard requires adding
823 // the size of the guard pages to the stack size, instead Linux
824 // takes the space out of 'stacksize'. Thus we adapt the requested
825 // stack_size by the size of the guard pages to mimick proper
826 // behaviour. However, be careful not to end up with a size
827 // of zero due to overflow. Don't add the guard page in that case.
828 size_t guard_size = os::Linux::default_guard_size(thr_type);
829 if (stack_size <= SIZE_MAX - guard_size) {
830 stack_size += guard_size;
831 }
832 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
833
834 int status = pthread_attr_setstacksize(&attr, stack_size);
835 assert_status(status == 0, status, "pthread_attr_setstacksize");
836
837 // Configure glibc guard page.
838 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
839
840 ThreadState state;
841
842 {
843 pthread_t tid;
844 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
845
846 char buf[64];
847 if (ret == 0) {
848 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
849 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
850 } else {
851 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
852 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
853 // Log some OS information which might explain why creating the thread failed.
854 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
855 LogStream st(Log(os, thread)::info());
856 os::Posix::print_rlimit_info(&st);
857 os::print_memory_info(&st);
858 os::Linux::print_proc_sys_info(&st);
859 os::Linux::print_container_info(&st);
860 }
861
862 pthread_attr_destroy(&attr);
863
864 if (ret != 0) {
865 // Need to clean up stuff we've allocated so far
866 thread->set_osthread(NULL);
867 delete osthread;
868 return false;
869 }
870
871 // Store pthread info into the OSThread
872 osthread->set_pthread_id(tid);
873
874 // Wait until child thread is either initialized or aborted
875 {
876 Monitor* sync_with_child = osthread->startThread_lock();
877 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
878 while ((state = osthread->get_state()) == ALLOCATED) {
879 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
880 }
881 }
882 }
883
884 // Aborted due to thread limit being reached
885 if (state == ZOMBIE) {
886 thread->set_osthread(NULL);
887 delete osthread;
888 return false;
889 }
890
891 // The thread is returned suspended (in state INITIALIZED),
892 // and is started higher up in the call chain
893 assert(state == INITIALIZED, "race condition");
894 return true;
895 }
896
897 /////////////////////////////////////////////////////////////////////////////
898 // attach existing thread
899
900 // bootstrap the main thread
901 bool os::create_main_thread(JavaThread* thread) {
902 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
903 return create_attached_thread(thread);
904 }
905
906 bool os::create_attached_thread(JavaThread* thread) {
907 #ifdef ASSERT
908 thread->verify_not_published();
909 #endif
910
911 // Allocate the OSThread object
912 OSThread* osthread = new OSThread(NULL, NULL);
913
914 if (osthread == NULL) {
915 return false;
916 }
917
918 // Store pthread info into the OSThread
919 osthread->set_thread_id(os::Linux::gettid());
920 osthread->set_pthread_id(::pthread_self());
921
922 // initialize floating point control register
923 os::Linux::init_thread_fpu_state();
924
925 // Initial thread state is RUNNABLE
926 osthread->set_state(RUNNABLE);
927
928 thread->set_osthread(osthread);
929
930 if (UseNUMA) {
931 int lgrp_id = os::numa_get_group_id();
932 if (lgrp_id != -1) {
933 thread->set_lgrp_id(lgrp_id);
934 }
935 }
936
937 if (os::is_primordial_thread()) {
938 // If current thread is primordial thread, its stack is mapped on demand,
939 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
940 // the entire stack region to avoid SEGV in stack banging.
941 // It is also useful to get around the heap-stack-gap problem on SuSE
942 // kernel (see 4821821 for details). We first expand stack to the top
943 // of yellow zone, then enable stack yellow zone (order is significant,
944 // enabling yellow zone first will crash JVM on SuSE Linux), so there
945 // is no gap between the last two virtual memory regions.
946
947 JavaThread *jt = (JavaThread *)thread;
948 address addr = jt->stack_reserved_zone_base();
949 assert(addr != NULL, "initialization problem?");
950 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
951
952 osthread->set_expanding_stack();
953 os::Linux::manually_expand_stack(jt, addr);
954 osthread->clear_expanding_stack();
955 }
956
957 // initialize signal mask for this thread
958 // and save the caller's signal mask
959 os::Linux::hotspot_sigmask(thread);
960
961 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
962 os::current_thread_id(), (uintx) pthread_self());
963
964 return true;
965 }
966
967 void os::pd_start_thread(Thread* thread) {
968 OSThread * osthread = thread->osthread();
969 assert(osthread->get_state() != INITIALIZED, "just checking");
970 Monitor* sync_with_child = osthread->startThread_lock();
971 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
972 sync_with_child->notify();
973 }
974
975 // Free Linux resources related to the OSThread
976 void os::free_thread(OSThread* osthread) {
977 assert(osthread != NULL, "osthread not set");
978
979 // We are told to free resources of the argument thread,
980 // but we can only really operate on the current thread.
981 assert(Thread::current()->osthread() == osthread,
982 "os::free_thread but not current thread");
983
984 #ifdef ASSERT
985 sigset_t current;
986 sigemptyset(¤t);
987 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
988 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
989 #endif
990
991 // Restore caller's signal mask
992 sigset_t sigmask = osthread->caller_sigmask();
993 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
994
995 delete osthread;
996 }
997
998 //////////////////////////////////////////////////////////////////////////////
999 // primordial thread
1000
1001 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1002 bool os::is_primordial_thread(void) {
1003 if (suppress_primordial_thread_resolution) {
1004 return false;
1005 }
1006 char dummy;
1007 // If called before init complete, thread stack bottom will be null.
1008 // Can be called if fatal error occurs before initialization.
1009 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1010 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1011 os::Linux::initial_thread_stack_size() != 0,
1012 "os::init did not locate primordial thread's stack region");
1013 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1014 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1015 os::Linux::initial_thread_stack_size()) {
1016 return true;
1017 } else {
1018 return false;
1019 }
1020 }
1021
1022 // Find the virtual memory area that contains addr
1023 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1024 FILE *fp = fopen("/proc/self/maps", "r");
1025 if (fp) {
1026 address low, high;
1027 while (!feof(fp)) {
1028 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1029 if (low <= addr && addr < high) {
1030 if (vma_low) *vma_low = low;
1031 if (vma_high) *vma_high = high;
1032 fclose(fp);
1033 return true;
1034 }
1035 }
1036 for (;;) {
1037 int ch = fgetc(fp);
1038 if (ch == EOF || ch == (int)'\n') break;
1039 }
1040 }
1041 fclose(fp);
1042 }
1043 return false;
1044 }
1045
1046 // Locate primordial thread stack. This special handling of primordial thread stack
1047 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1048 // bogus value for the primordial process thread. While the launcher has created
1049 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1050 // JNI invocation API from a primordial thread.
1051 void os::Linux::capture_initial_stack(size_t max_size) {
1052
1053 // max_size is either 0 (which means accept OS default for thread stacks) or
1054 // a user-specified value known to be at least the minimum needed. If we
1055 // are actually on the primordial thread we can make it appear that we have a
1056 // smaller max_size stack by inserting the guard pages at that location. But we
1057 // cannot do anything to emulate a larger stack than what has been provided by
1058 // the OS or threading library. In fact if we try to use a stack greater than
1059 // what is set by rlimit then we will crash the hosting process.
1060
1061 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1062 // If this is "unlimited" then it will be a huge value.
1063 struct rlimit rlim;
1064 getrlimit(RLIMIT_STACK, &rlim);
1065 size_t stack_size = rlim.rlim_cur;
1066
1067 // 6308388: a bug in ld.so will relocate its own .data section to the
1068 // lower end of primordial stack; reduce ulimit -s value a little bit
1069 // so we won't install guard page on ld.so's data section.
1070 // But ensure we don't underflow the stack size - allow 1 page spare
1071 if (stack_size >= (size_t)(3 * page_size())) {
1072 stack_size -= 2 * page_size();
1073 }
1074
1075 // Try to figure out where the stack base (top) is. This is harder.
1076 //
1077 // When an application is started, glibc saves the initial stack pointer in
1078 // a global variable "__libc_stack_end", which is then used by system
1079 // libraries. __libc_stack_end should be pretty close to stack top. The
1080 // variable is available since the very early days. However, because it is
1081 // a private interface, it could disappear in the future.
1082 //
1083 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1084 // to __libc_stack_end, it is very close to stack top, but isn't the real
1085 // stack top. Note that /proc may not exist if VM is running as a chroot
1086 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1087 // /proc/<pid>/stat could change in the future (though unlikely).
1088 //
1089 // We try __libc_stack_end first. If that doesn't work, look for
1090 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1091 // as a hint, which should work well in most cases.
1092
1093 uintptr_t stack_start;
1094
1095 // try __libc_stack_end first
1096 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1097 if (p && *p) {
1098 stack_start = *p;
1099 } else {
1100 // see if we can get the start_stack field from /proc/self/stat
1101 FILE *fp;
1102 int pid;
1103 char state;
1104 int ppid;
1105 int pgrp;
1106 int session;
1107 int nr;
1108 int tpgrp;
1109 unsigned long flags;
1110 unsigned long minflt;
1111 unsigned long cminflt;
1112 unsigned long majflt;
1113 unsigned long cmajflt;
1114 unsigned long utime;
1115 unsigned long stime;
1116 long cutime;
1117 long cstime;
1118 long prio;
1119 long nice;
1120 long junk;
1121 long it_real;
1122 uintptr_t start;
1123 uintptr_t vsize;
1124 intptr_t rss;
1125 uintptr_t rsslim;
1126 uintptr_t scodes;
1127 uintptr_t ecode;
1128 int i;
1129
1130 // Figure what the primordial thread stack base is. Code is inspired
1131 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1132 // followed by command name surrounded by parentheses, state, etc.
1133 char stat[2048];
1134 int statlen;
1135
1136 fp = fopen("/proc/self/stat", "r");
1137 if (fp) {
1138 statlen = fread(stat, 1, 2047, fp);
1139 stat[statlen] = '\0';
1140 fclose(fp);
1141
1142 // Skip pid and the command string. Note that we could be dealing with
1143 // weird command names, e.g. user could decide to rename java launcher
1144 // to "java 1.4.2 :)", then the stat file would look like
1145 // 1234 (java 1.4.2 :)) R ... ...
1146 // We don't really need to know the command string, just find the last
1147 // occurrence of ")" and then start parsing from there. See bug 4726580.
1148 char * s = strrchr(stat, ')');
1149
1150 i = 0;
1151 if (s) {
1152 // Skip blank chars
1153 do { s++; } while (s && isspace(*s));
1154
1155 #define _UFM UINTX_FORMAT
1156 #define _DFM INTX_FORMAT
1157
1158 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1159 // 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
1160 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,
1161 &state, // 3 %c
1162 &ppid, // 4 %d
1163 &pgrp, // 5 %d
1164 &session, // 6 %d
1165 &nr, // 7 %d
1166 &tpgrp, // 8 %d
1167 &flags, // 9 %lu
1168 &minflt, // 10 %lu
1169 &cminflt, // 11 %lu
1170 &majflt, // 12 %lu
1171 &cmajflt, // 13 %lu
1172 &utime, // 14 %lu
1173 &stime, // 15 %lu
1174 &cutime, // 16 %ld
1175 &cstime, // 17 %ld
1176 &prio, // 18 %ld
1177 &nice, // 19 %ld
1178 &junk, // 20 %ld
1179 &it_real, // 21 %ld
1180 &start, // 22 UINTX_FORMAT
1181 &vsize, // 23 UINTX_FORMAT
1182 &rss, // 24 INTX_FORMAT
1183 &rsslim, // 25 UINTX_FORMAT
1184 &scodes, // 26 UINTX_FORMAT
1185 &ecode, // 27 UINTX_FORMAT
1186 &stack_start); // 28 UINTX_FORMAT
1187 }
1188
1189 #undef _UFM
1190 #undef _DFM
1191
1192 if (i != 28 - 2) {
1193 assert(false, "Bad conversion from /proc/self/stat");
1194 // product mode - assume we are the primordial thread, good luck in the
1195 // embedded case.
1196 warning("Can't detect primordial thread stack location - bad conversion");
1197 stack_start = (uintptr_t) &rlim;
1198 }
1199 } else {
1200 // For some reason we can't open /proc/self/stat (for example, running on
1201 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1202 // most cases, so don't abort:
1203 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1204 stack_start = (uintptr_t) &rlim;
1205 }
1206 }
1207
1208 // Now we have a pointer (stack_start) very close to the stack top, the
1209 // next thing to do is to figure out the exact location of stack top. We
1210 // can find out the virtual memory area that contains stack_start by
1211 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1212 // and its upper limit is the real stack top. (again, this would fail if
1213 // running inside chroot, because /proc may not exist.)
1214
1215 uintptr_t stack_top;
1216 address low, high;
1217 if (find_vma((address)stack_start, &low, &high)) {
1218 // success, "high" is the true stack top. (ignore "low", because initial
1219 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1220 stack_top = (uintptr_t)high;
1221 } else {
1222 // failed, likely because /proc/self/maps does not exist
1223 warning("Can't detect primordial thread stack location - find_vma failed");
1224 // best effort: stack_start is normally within a few pages below the real
1225 // stack top, use it as stack top, and reduce stack size so we won't put
1226 // guard page outside stack.
1227 stack_top = stack_start;
1228 stack_size -= 16 * page_size();
1229 }
1230
1231 // stack_top could be partially down the page so align it
1232 stack_top = align_up(stack_top, page_size());
1233
1234 // Allowed stack value is minimum of max_size and what we derived from rlimit
1235 if (max_size > 0) {
1236 _initial_thread_stack_size = MIN2(max_size, stack_size);
1237 } else {
1238 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1239 // clamp it at 8MB as we do on Solaris
1240 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1241 }
1242 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1243 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1244
1245 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1246
1247 if (log_is_enabled(Info, os, thread)) {
1248 // See if we seem to be on primordial process thread
1249 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1250 uintptr_t(&rlim) < stack_top;
1251
1252 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1253 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1254 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1255 stack_top, intptr_t(_initial_thread_stack_bottom));
1256 }
1257 }
1258
1259 ////////////////////////////////////////////////////////////////////////////////
1260 // time support
1261
1262 // Time since start-up in seconds to a fine granularity.
1263 // Used by VMSelfDestructTimer and the MemProfiler.
1264 double os::elapsedTime() {
1265
1266 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1267 }
1268
1269 jlong os::elapsed_counter() {
1270 return javaTimeNanos() - initial_time_count;
1271 }
1272
1273 jlong os::elapsed_frequency() {
1274 return NANOSECS_PER_SEC; // nanosecond resolution
1275 }
1276
1277 bool os::supports_vtime() { return true; }
1278 bool os::enable_vtime() { return false; }
1279 bool os::vtime_enabled() { return false; }
1280
1281 double os::elapsedVTime() {
1282 struct rusage usage;
1283 int retval = getrusage(RUSAGE_THREAD, &usage);
1284 if (retval == 0) {
1285 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);
1286 } else {
1287 // better than nothing, but not much
1288 return elapsedTime();
1289 }
1290 }
1291
1292 jlong os::javaTimeMillis() {
1293 timeval time;
1294 int status = gettimeofday(&time, NULL);
1295 assert(status != -1, "linux error");
1296 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1297 }
1298
1299 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1300 timeval time;
1301 int status = gettimeofday(&time, NULL);
1302 assert(status != -1, "linux error");
1303 seconds = jlong(time.tv_sec);
1304 nanos = jlong(time.tv_usec) * 1000;
1305 }
1306
1307
1308 #ifndef CLOCK_MONOTONIC
1309 #define CLOCK_MONOTONIC (1)
1310 #endif
1311
1312 void os::Linux::clock_init() {
1313 // we do dlopen's in this particular order due to bug in linux
1314 // dynamical loader (see 6348968) leading to crash on exit
1315 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1316 if (handle == NULL) {
1317 handle = dlopen("librt.so", RTLD_LAZY);
1318 }
1319
1320 if (handle) {
1321 int (*clock_getres_func)(clockid_t, struct timespec*) =
1322 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1323 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1324 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1325 if (clock_getres_func && clock_gettime_func) {
1326 // See if monotonic clock is supported by the kernel. Note that some
1327 // early implementations simply return kernel jiffies (updated every
1328 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1329 // for nano time (though the monotonic property is still nice to have).
1330 // It's fixed in newer kernels, however clock_getres() still returns
1331 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1332 // resolution for now. Hopefully as people move to new kernels, this
1333 // won't be a problem.
1334 struct timespec res;
1335 struct timespec tp;
1336 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1337 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1338 // yes, monotonic clock is supported
1339 _clock_gettime = clock_gettime_func;
1340 return;
1341 } else {
1342 // close librt if there is no monotonic clock
1343 dlclose(handle);
1344 }
1345 }
1346 }
1347 warning("No monotonic clock was available - timed services may " \
1348 "be adversely affected if the time-of-day clock changes");
1349 }
1350
1351 #ifndef SYS_clock_getres
1352 #if defined(X86) || defined(PPC64) || defined(S390)
1353 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1354 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1355 #else
1356 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1357 #define sys_clock_getres(x,y) -1
1358 #endif
1359 #else
1360 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1361 #endif
1362
1363 void os::Linux::fast_thread_clock_init() {
1364 if (!UseLinuxPosixThreadCPUClocks) {
1365 return;
1366 }
1367 clockid_t clockid;
1368 struct timespec tp;
1369 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1370 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1371
1372 // Switch to using fast clocks for thread cpu time if
1373 // the sys_clock_getres() returns 0 error code.
1374 // Note, that some kernels may support the current thread
1375 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1376 // returned by the pthread_getcpuclockid().
1377 // If the fast Posix clocks are supported then the sys_clock_getres()
1378 // must return at least tp.tv_sec == 0 which means a resolution
1379 // better than 1 sec. This is extra check for reliability.
1380
1381 if (pthread_getcpuclockid_func &&
1382 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1383 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1384 _supports_fast_thread_cpu_time = true;
1385 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1386 }
1387 }
1388
1389 jlong os::javaTimeNanos() {
1390 if (os::supports_monotonic_clock()) {
1391 struct timespec tp;
1392 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1393 assert(status == 0, "gettime error");
1394 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1395 return result;
1396 } else {
1397 timeval time;
1398 int status = gettimeofday(&time, NULL);
1399 assert(status != -1, "linux error");
1400 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1401 return 1000 * usecs;
1402 }
1403 }
1404
1405 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1406 if (os::supports_monotonic_clock()) {
1407 info_ptr->max_value = ALL_64_BITS;
1408
1409 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1410 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1411 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1412 } else {
1413 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1414 info_ptr->max_value = ALL_64_BITS;
1415
1416 // gettimeofday is a real time clock so it skips
1417 info_ptr->may_skip_backward = true;
1418 info_ptr->may_skip_forward = true;
1419 }
1420
1421 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1422 }
1423
1424 // Return the real, user, and system times in seconds from an
1425 // arbitrary fixed point in the past.
1426 bool os::getTimesSecs(double* process_real_time,
1427 double* process_user_time,
1428 double* process_system_time) {
1429 struct tms ticks;
1430 clock_t real_ticks = times(&ticks);
1431
1432 if (real_ticks == (clock_t) (-1)) {
1433 return false;
1434 } else {
1435 double ticks_per_second = (double) clock_tics_per_sec;
1436 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1437 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1438 *process_real_time = ((double) real_ticks) / ticks_per_second;
1439
1440 return true;
1441 }
1442 }
1443
1444
1445 char * os::local_time_string(char *buf, size_t buflen) {
1446 struct tm t;
1447 time_t long_time;
1448 time(&long_time);
1449 localtime_r(&long_time, &t);
1450 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1451 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1452 t.tm_hour, t.tm_min, t.tm_sec);
1453 return buf;
1454 }
1455
1456 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1457 return localtime_r(clock, res);
1458 }
1459
1460 ////////////////////////////////////////////////////////////////////////////////
1461 // runtime exit support
1462
1463 // Note: os::shutdown() might be called very early during initialization, or
1464 // called from signal handler. Before adding something to os::shutdown(), make
1465 // sure it is async-safe and can handle partially initialized VM.
1466 void os::shutdown() {
1467
1468 // allow PerfMemory to attempt cleanup of any persistent resources
1469 perfMemory_exit();
1470
1471 // needs to remove object in file system
1472 AttachListener::abort();
1473
1474 // flush buffered output, finish log files
1475 ostream_abort();
1476
1477 // Check for abort hook
1478 abort_hook_t abort_hook = Arguments::abort_hook();
1479 if (abort_hook != NULL) {
1480 abort_hook();
1481 }
1482
1483 }
1484
1485 // Note: os::abort() might be called very early during initialization, or
1486 // called from signal handler. Before adding something to os::abort(), make
1487 // sure it is async-safe and can handle partially initialized VM.
1488 void os::abort(bool dump_core, void* siginfo, const void* context) {
1489 os::shutdown();
1490 if (dump_core) {
1491 if (DumpPrivateMappingsInCore) {
1492 ClassLoader::close_jrt_image();
1493 }
1494 #ifndef PRODUCT
1495 fdStream out(defaultStream::output_fd());
1496 out.print_raw("Current thread is ");
1497 char buf[16];
1498 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1499 out.print_raw_cr(buf);
1500 out.print_raw_cr("Dumping core ...");
1501 #endif
1502 ::abort(); // dump core
1503 }
1504
1505 ::exit(1);
1506 }
1507
1508 // Die immediately, no exit hook, no abort hook, no cleanup.
1509 void os::die() {
1510 ::abort();
1511 }
1512
1513
1514 // This method is a copy of JDK's sysGetLastErrorString
1515 // from src/solaris/hpi/src/system_md.c
1516
1517 size_t os::lasterror(char *buf, size_t len) {
1518 if (errno == 0) return 0;
1519
1520 const char *s = os::strerror(errno);
1521 size_t n = ::strlen(s);
1522 if (n >= len) {
1523 n = len - 1;
1524 }
1525 ::strncpy(buf, s, n);
1526 buf[n] = '\0';
1527 return n;
1528 }
1529
1530 // thread_id is kernel thread id (similar to Solaris LWP id)
1531 intx os::current_thread_id() { return os::Linux::gettid(); }
1532 int os::current_process_id() {
1533 return ::getpid();
1534 }
1535
1536 // DLL functions
1537
1538 const char* os::dll_file_extension() { return ".so"; }
1539
1540 // This must be hard coded because it's the system's temporary
1541 // directory not the java application's temp directory, ala java.io.tmpdir.
1542 const char* os::get_temp_directory() { return "/tmp"; }
1543
1544 static bool file_exists(const char* filename) {
1545 struct stat statbuf;
1546 if (filename == NULL || strlen(filename) == 0) {
1547 return false;
1548 }
1549 return os::stat(filename, &statbuf) == 0;
1550 }
1551
1552 // check if addr is inside libjvm.so
1553 bool os::address_is_in_vm(address addr) {
1554 static address libjvm_base_addr;
1555 Dl_info dlinfo;
1556
1557 if (libjvm_base_addr == NULL) {
1558 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1559 libjvm_base_addr = (address)dlinfo.dli_fbase;
1560 }
1561 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1562 }
1563
1564 if (dladdr((void *)addr, &dlinfo) != 0) {
1565 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1566 }
1567
1568 return false;
1569 }
1570
1571 bool os::dll_address_to_function_name(address addr, char *buf,
1572 int buflen, int *offset,
1573 bool demangle) {
1574 // buf is not optional, but offset is optional
1575 assert(buf != NULL, "sanity check");
1576
1577 Dl_info dlinfo;
1578
1579 if (dladdr((void*)addr, &dlinfo) != 0) {
1580 // see if we have a matching symbol
1581 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1582 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1583 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1584 }
1585 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1586 return true;
1587 }
1588 // no matching symbol so try for just file info
1589 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1590 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1591 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1592 return true;
1593 }
1594 }
1595 }
1596
1597 buf[0] = '\0';
1598 if (offset != NULL) *offset = -1;
1599 return false;
1600 }
1601
1602 struct _address_to_library_name {
1603 address addr; // input : memory address
1604 size_t buflen; // size of fname
1605 char* fname; // output: library name
1606 address base; // library base addr
1607 };
1608
1609 static int address_to_library_name_callback(struct dl_phdr_info *info,
1610 size_t size, void *data) {
1611 int i;
1612 bool found = false;
1613 address libbase = NULL;
1614 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1615
1616 // iterate through all loadable segments
1617 for (i = 0; i < info->dlpi_phnum; i++) {
1618 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1619 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1620 // base address of a library is the lowest address of its loaded
1621 // segments.
1622 if (libbase == NULL || libbase > segbase) {
1623 libbase = segbase;
1624 }
1625 // see if 'addr' is within current segment
1626 if (segbase <= d->addr &&
1627 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1628 found = true;
1629 }
1630 }
1631 }
1632
1633 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1634 // so dll_address_to_library_name() can fall through to use dladdr() which
1635 // can figure out executable name from argv[0].
1636 if (found && info->dlpi_name && info->dlpi_name[0]) {
1637 d->base = libbase;
1638 if (d->fname) {
1639 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1640 }
1641 return 1;
1642 }
1643 return 0;
1644 }
1645
1646 bool os::dll_address_to_library_name(address addr, char* buf,
1647 int buflen, int* offset) {
1648 // buf is not optional, but offset is optional
1649 assert(buf != NULL, "sanity check");
1650
1651 Dl_info dlinfo;
1652 struct _address_to_library_name data;
1653
1654 // There is a bug in old glibc dladdr() implementation that it could resolve
1655 // to wrong library name if the .so file has a base address != NULL. Here
1656 // we iterate through the program headers of all loaded libraries to find
1657 // out which library 'addr' really belongs to. This workaround can be
1658 // removed once the minimum requirement for glibc is moved to 2.3.x.
1659 data.addr = addr;
1660 data.fname = buf;
1661 data.buflen = buflen;
1662 data.base = NULL;
1663 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1664
1665 if (rslt) {
1666 // buf already contains library name
1667 if (offset) *offset = addr - data.base;
1668 return true;
1669 }
1670 if (dladdr((void*)addr, &dlinfo) != 0) {
1671 if (dlinfo.dli_fname != NULL) {
1672 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1673 }
1674 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1675 *offset = addr - (address)dlinfo.dli_fbase;
1676 }
1677 return true;
1678 }
1679
1680 buf[0] = '\0';
1681 if (offset) *offset = -1;
1682 return false;
1683 }
1684
1685 // Loads .dll/.so and
1686 // in case of error it checks if .dll/.so was built for the
1687 // same architecture as Hotspot is running on
1688
1689
1690 // Remember the stack's state. The Linux dynamic linker will change
1691 // the stack to 'executable' at most once, so we must safepoint only once.
1692 bool os::Linux::_stack_is_executable = false;
1693
1694 // VM operation that loads a library. This is necessary if stack protection
1695 // of the Java stacks can be lost during loading the library. If we
1696 // do not stop the Java threads, they can stack overflow before the stacks
1697 // are protected again.
1698 class VM_LinuxDllLoad: public VM_Operation {
1699 private:
1700 const char *_filename;
1701 char *_ebuf;
1702 int _ebuflen;
1703 void *_lib;
1704 public:
1705 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1706 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1707 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1708 void doit() {
1709 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1710 os::Linux::_stack_is_executable = true;
1711 }
1712 void* loaded_library() { return _lib; }
1713 };
1714
1715 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1716 void * result = NULL;
1717 bool load_attempted = false;
1718
1719 // Check whether the library to load might change execution rights
1720 // of the stack. If they are changed, the protection of the stack
1721 // guard pages will be lost. We need a safepoint to fix this.
1722 //
1723 // See Linux man page execstack(8) for more info.
1724 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1725 if (!ElfFile::specifies_noexecstack(filename)) {
1726 if (!is_init_completed()) {
1727 os::Linux::_stack_is_executable = true;
1728 // This is OK - No Java threads have been created yet, and hence no
1729 // stack guard pages to fix.
1730 //
1731 // Dynamic loader will make all stacks executable after
1732 // this function returns, and will not do that again.
1733 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1734 } else {
1735 warning("You have loaded library %s which might have disabled stack guard. "
1736 "The VM will try to fix the stack guard now.\n"
1737 "It's highly recommended that you fix the library with "
1738 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1739 filename);
1740
1741 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1742 JavaThread *jt = JavaThread::current();
1743 if (jt->thread_state() != _thread_in_native) {
1744 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1745 // that requires ExecStack. Cannot enter safe point. Let's give up.
1746 warning("Unable to fix stack guard. Giving up.");
1747 } else {
1748 if (!LoadExecStackDllInVMThread) {
1749 // This is for the case where the DLL has an static
1750 // constructor function that executes JNI code. We cannot
1751 // load such DLLs in the VMThread.
1752 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1753 }
1754
1755 ThreadInVMfromNative tiv(jt);
1756 debug_only(VMNativeEntryWrapper vew;)
1757
1758 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1759 VMThread::execute(&op);
1760 if (LoadExecStackDllInVMThread) {
1761 result = op.loaded_library();
1762 }
1763 load_attempted = true;
1764 }
1765 }
1766 }
1767 }
1768
1769 if (!load_attempted) {
1770 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1771 }
1772
1773 if (result != NULL) {
1774 // Successful loading
1775 return result;
1776 }
1777
1778 Elf32_Ehdr elf_head;
1779 int diag_msg_max_length=ebuflen-strlen(ebuf);
1780 char* diag_msg_buf=ebuf+strlen(ebuf);
1781
1782 if (diag_msg_max_length==0) {
1783 // No more space in ebuf for additional diagnostics message
1784 return NULL;
1785 }
1786
1787
1788 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1789
1790 if (file_descriptor < 0) {
1791 // Can't open library, report dlerror() message
1792 return NULL;
1793 }
1794
1795 bool failed_to_read_elf_head=
1796 (sizeof(elf_head)!=
1797 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1798
1799 ::close(file_descriptor);
1800 if (failed_to_read_elf_head) {
1801 // file i/o error - report dlerror() msg
1802 return NULL;
1803 }
1804
1805 typedef struct {
1806 Elf32_Half code; // Actual value as defined in elf.h
1807 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1808 unsigned char elf_class; // 32 or 64 bit
1809 unsigned char endianess; // MSB or LSB
1810 char* name; // String representation
1811 } arch_t;
1812
1813 #ifndef EM_486
1814 #define EM_486 6 /* Intel 80486 */
1815 #endif
1816 #ifndef EM_AARCH64
1817 #define EM_AARCH64 183 /* ARM AARCH64 */
1818 #endif
1819
1820 static const arch_t arch_array[]={
1821 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1822 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1823 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1824 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1825 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1826 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1827 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1828 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1829 #if defined(VM_LITTLE_ENDIAN)
1830 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1831 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1832 #else
1833 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1834 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1835 #endif
1836 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1837 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1838 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1839 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1840 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1841 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1842 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1843 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1844 };
1845
1846 #if (defined IA32)
1847 static Elf32_Half running_arch_code=EM_386;
1848 #elif (defined AMD64)
1849 static Elf32_Half running_arch_code=EM_X86_64;
1850 #elif (defined IA64)
1851 static Elf32_Half running_arch_code=EM_IA_64;
1852 #elif (defined __sparc) && (defined _LP64)
1853 static Elf32_Half running_arch_code=EM_SPARCV9;
1854 #elif (defined __sparc) && (!defined _LP64)
1855 static Elf32_Half running_arch_code=EM_SPARC;
1856 #elif (defined __powerpc64__)
1857 static Elf32_Half running_arch_code=EM_PPC64;
1858 #elif (defined __powerpc__)
1859 static Elf32_Half running_arch_code=EM_PPC;
1860 #elif (defined AARCH64)
1861 static Elf32_Half running_arch_code=EM_AARCH64;
1862 #elif (defined ARM)
1863 static Elf32_Half running_arch_code=EM_ARM;
1864 #elif (defined S390)
1865 static Elf32_Half running_arch_code=EM_S390;
1866 #elif (defined ALPHA)
1867 static Elf32_Half running_arch_code=EM_ALPHA;
1868 #elif (defined MIPSEL)
1869 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1870 #elif (defined PARISC)
1871 static Elf32_Half running_arch_code=EM_PARISC;
1872 #elif (defined MIPS)
1873 static Elf32_Half running_arch_code=EM_MIPS;
1874 #elif (defined M68K)
1875 static Elf32_Half running_arch_code=EM_68K;
1876 #elif (defined SH)
1877 static Elf32_Half running_arch_code=EM_SH;
1878 #else
1879 #error Method os::dll_load requires that one of following is defined:\
1880 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1881 #endif
1882
1883 // Identify compatability class for VM's architecture and library's architecture
1884 // Obtain string descriptions for architectures
1885
1886 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1887 int running_arch_index=-1;
1888
1889 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1890 if (running_arch_code == arch_array[i].code) {
1891 running_arch_index = i;
1892 }
1893 if (lib_arch.code == arch_array[i].code) {
1894 lib_arch.compat_class = arch_array[i].compat_class;
1895 lib_arch.name = arch_array[i].name;
1896 }
1897 }
1898
1899 assert(running_arch_index != -1,
1900 "Didn't find running architecture code (running_arch_code) in arch_array");
1901 if (running_arch_index == -1) {
1902 // Even though running architecture detection failed
1903 // we may still continue with reporting dlerror() message
1904 return NULL;
1905 }
1906
1907 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1908 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1909 return NULL;
1910 }
1911
1912 #ifndef S390
1913 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1914 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1915 return NULL;
1916 }
1917 #endif // !S390
1918
1919 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1920 if (lib_arch.name!=NULL) {
1921 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1922 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1923 lib_arch.name, arch_array[running_arch_index].name);
1924 } else {
1925 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1926 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1927 lib_arch.code,
1928 arch_array[running_arch_index].name);
1929 }
1930 }
1931
1932 return NULL;
1933 }
1934
1935 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1936 int ebuflen) {
1937 void * result = ::dlopen(filename, RTLD_LAZY);
1938 if (result == NULL) {
1939 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1940 ebuf[ebuflen-1] = '\0';
1941 }
1942 return result;
1943 }
1944
1945 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1946 int ebuflen) {
1947 void * result = NULL;
1948 if (LoadExecStackDllInVMThread) {
1949 result = dlopen_helper(filename, ebuf, ebuflen);
1950 }
1951
1952 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1953 // library that requires an executable stack, or which does not have this
1954 // stack attribute set, dlopen changes the stack attribute to executable. The
1955 // read protection of the guard pages gets lost.
1956 //
1957 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1958 // may have been queued at the same time.
1959
1960 if (!_stack_is_executable) {
1961 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1962 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1963 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1964 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1965 warning("Attempt to reguard stack yellow zone failed.");
1966 }
1967 }
1968 }
1969 }
1970
1971 return result;
1972 }
1973
1974 void* os::dll_lookup(void* handle, const char* name) {
1975 void* res = dlsym(handle, name);
1976 return res;
1977 }
1978
1979 void* os::get_default_process_handle() {
1980 return (void*)::dlopen(NULL, RTLD_LAZY);
1981 }
1982
1983 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1984 int fd = ::open(filename, O_RDONLY);
1985 if (fd == -1) {
1986 return false;
1987 }
1988
1989 if (hdr != NULL) {
1990 st->print_cr("%s", hdr);
1991 }
1992
1993 char buf[33];
1994 int bytes;
1995 buf[32] = '\0';
1996 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1997 st->print_raw(buf, bytes);
1998 }
1999
2000 ::close(fd);
2001
2002 return true;
2003 }
2004
2005 void os::print_dll_info(outputStream *st) {
2006 st->print_cr("Dynamic libraries:");
2007
2008 char fname[32];
2009 pid_t pid = os::Linux::gettid();
2010
2011 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2012
2013 if (!_print_ascii_file(fname, st)) {
2014 st->print("Can not get library information for pid = %d\n", pid);
2015 }
2016 }
2017
2018 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2019 FILE *procmapsFile = NULL;
2020
2021 // Open the procfs maps file for the current process
2022 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2023 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2024 char line[PATH_MAX + 100];
2025
2026 // Read line by line from 'file'
2027 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2028 u8 base, top, offset, inode;
2029 char permissions[5];
2030 char device[6];
2031 char name[PATH_MAX + 1];
2032
2033 // Parse fields from line
2034 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2035 &base, &top, permissions, &offset, device, &inode, name);
2036
2037 // Filter by device id '00:00' so that we only get file system mapped files.
2038 if (strcmp(device, "00:00") != 0) {
2039
2040 // Call callback with the fields of interest
2041 if(callback(name, (address)base, (address)top, param)) {
2042 // Oops abort, callback aborted
2043 fclose(procmapsFile);
2044 return 1;
2045 }
2046 }
2047 }
2048 fclose(procmapsFile);
2049 }
2050 return 0;
2051 }
2052
2053 void os::print_os_info_brief(outputStream* st) {
2054 os::Linux::print_distro_info(st);
2055
2056 os::Posix::print_uname_info(st);
2057
2058 os::Linux::print_libversion_info(st);
2059
2060 }
2061
2062 void os::print_os_info(outputStream* st) {
2063 st->print("OS:");
2064
2065 os::Linux::print_distro_info(st);
2066
2067 os::Posix::print_uname_info(st);
2068
2069 // Print warning if unsafe chroot environment detected
2070 if (unsafe_chroot_detected) {
2071 st->print("WARNING!! ");
2072 st->print_cr("%s", unstable_chroot_error);
2073 }
2074
2075 os::Linux::print_libversion_info(st);
2076
2077 os::Posix::print_rlimit_info(st);
2078
2079 os::Posix::print_load_average(st);
2080
2081 os::Linux::print_full_memory_info(st);
2082
2083 os::Linux::print_proc_sys_info(st);
2084
2085 os::Linux::print_ld_preload_file(st);
2086
2087 os::Linux::print_container_info(st);
2088
2089 VM_Version::print_platform_virtualization_info(st);
2090
2091 os::Linux::print_steal_info(st);
2092 }
2093
2094 // Try to identify popular distros.
2095 // Most Linux distributions have a /etc/XXX-release file, which contains
2096 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2097 // file that also contains the OS version string. Some have more than one
2098 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2099 // /etc/redhat-release.), so the order is important.
2100 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2101 // their own specific XXX-release file as well as a redhat-release file.
2102 // Because of this the XXX-release file needs to be searched for before the
2103 // redhat-release file.
2104 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2105 // search for redhat-release / SuSE-release needs to be before lsb-release.
2106 // Since the lsb-release file is the new standard it needs to be searched
2107 // before the older style release files.
2108 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2109 // next to last resort. The os-release file is a new standard that contains
2110 // distribution information and the system-release file seems to be an old
2111 // standard that has been replaced by the lsb-release and os-release files.
2112 // Searching for the debian_version file is the last resort. It contains
2113 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2114 // "Debian " is printed before the contents of the debian_version file.
2115
2116 const char* distro_files[] = {
2117 "/etc/oracle-release",
2118 "/etc/mandriva-release",
2119 "/etc/mandrake-release",
2120 "/etc/sun-release",
2121 "/etc/redhat-release",
2122 "/etc/SuSE-release",
2123 "/etc/lsb-release",
2124 "/etc/turbolinux-release",
2125 "/etc/gentoo-release",
2126 "/etc/ltib-release",
2127 "/etc/angstrom-version",
2128 "/etc/system-release",
2129 "/etc/os-release",
2130 NULL };
2131
2132 void os::Linux::print_distro_info(outputStream* st) {
2133 for (int i = 0;; i++) {
2134 const char* file = distro_files[i];
2135 if (file == NULL) {
2136 break; // done
2137 }
2138 // If file prints, we found it.
2139 if (_print_ascii_file(file, st)) {
2140 return;
2141 }
2142 }
2143
2144 if (file_exists("/etc/debian_version")) {
2145 st->print("Debian ");
2146 _print_ascii_file("/etc/debian_version", st);
2147 } else {
2148 st->print("Linux");
2149 }
2150 st->cr();
2151 }
2152
2153 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2154 char buf[256];
2155 while (fgets(buf, sizeof(buf), fp)) {
2156 // Edit out extra stuff in expected format
2157 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2158 char* ptr = strstr(buf, "\""); // the name is in quotes
2159 if (ptr != NULL) {
2160 ptr++; // go beyond first quote
2161 char* nl = strchr(ptr, '\"');
2162 if (nl != NULL) *nl = '\0';
2163 strncpy(distro, ptr, length);
2164 } else {
2165 ptr = strstr(buf, "=");
2166 ptr++; // go beyond equals then
2167 char* nl = strchr(ptr, '\n');
2168 if (nl != NULL) *nl = '\0';
2169 strncpy(distro, ptr, length);
2170 }
2171 return;
2172 } else if (get_first_line) {
2173 char* nl = strchr(buf, '\n');
2174 if (nl != NULL) *nl = '\0';
2175 strncpy(distro, buf, length);
2176 return;
2177 }
2178 }
2179 // print last line and close
2180 char* nl = strchr(buf, '\n');
2181 if (nl != NULL) *nl = '\0';
2182 strncpy(distro, buf, length);
2183 }
2184
2185 static void parse_os_info(char* distro, size_t length, const char* file) {
2186 FILE* fp = fopen(file, "r");
2187 if (fp != NULL) {
2188 // if suse format, print out first line
2189 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2190 parse_os_info_helper(fp, distro, length, get_first_line);
2191 fclose(fp);
2192 }
2193 }
2194
2195 void os::get_summary_os_info(char* buf, size_t buflen) {
2196 for (int i = 0;; i++) {
2197 const char* file = distro_files[i];
2198 if (file == NULL) {
2199 break; // ran out of distro_files
2200 }
2201 if (file_exists(file)) {
2202 parse_os_info(buf, buflen, file);
2203 return;
2204 }
2205 }
2206 // special case for debian
2207 if (file_exists("/etc/debian_version")) {
2208 strncpy(buf, "Debian ", buflen);
2209 if (buflen > 7) {
2210 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2211 }
2212 } else {
2213 strncpy(buf, "Linux", buflen);
2214 }
2215 }
2216
2217 void os::Linux::print_libversion_info(outputStream* st) {
2218 // libc, pthread
2219 st->print("libc:");
2220 st->print("%s ", os::Linux::glibc_version());
2221 st->print("%s ", os::Linux::libpthread_version());
2222 st->cr();
2223 }
2224
2225 void os::Linux::print_proc_sys_info(outputStream* st) {
2226 st->cr();
2227 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2228 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2229 st->cr();
2230 st->cr();
2231
2232 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2233 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2234 st->cr();
2235 st->cr();
2236
2237 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2238 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2239 st->cr();
2240 st->cr();
2241 }
2242
2243 void os::Linux::print_full_memory_info(outputStream* st) {
2244 st->print("\n/proc/meminfo:\n");
2245 _print_ascii_file("/proc/meminfo", st);
2246 st->cr();
2247 }
2248
2249 void os::Linux::print_ld_preload_file(outputStream* st) {
2250 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2251 st->cr();
2252 }
2253
2254 void os::Linux::print_container_info(outputStream* st) {
2255 if (!OSContainer::is_containerized()) {
2256 return;
2257 }
2258
2259 st->print("container (cgroup) information:\n");
2260
2261 const char *p_ct = OSContainer::container_type();
2262 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2263
2264 char *p = OSContainer::cpu_cpuset_cpus();
2265 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2266 free(p);
2267
2268 p = OSContainer::cpu_cpuset_memory_nodes();
2269 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2270 free(p);
2271
2272 int i = OSContainer::active_processor_count();
2273 if (i > 0) {
2274 st->print("active_processor_count: %d\n", i);
2275 } else {
2276 st->print("active_processor_count: failed\n");
2277 }
2278
2279 i = OSContainer::cpu_quota();
2280 st->print("cpu_quota: %d\n", i);
2281
2282 i = OSContainer::cpu_period();
2283 st->print("cpu_period: %d\n", i);
2284
2285 i = OSContainer::cpu_shares();
2286 st->print("cpu_shares: %d\n", i);
2287
2288 jlong j = OSContainer::memory_limit_in_bytes();
2289 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2290
2291 j = OSContainer::memory_and_swap_limit_in_bytes();
2292 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2293
2294 j = OSContainer::memory_soft_limit_in_bytes();
2295 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2296
2297 j = OSContainer::OSContainer::memory_usage_in_bytes();
2298 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2299
2300 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2301 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2302 st->cr();
2303 }
2304
2305 void os::Linux::print_steal_info(outputStream* st) {
2306 if (has_initial_tick_info) {
2307 CPUPerfTicks pticks;
2308 bool res = os::Linux::get_tick_information(&pticks, -1);
2309
2310 if (res && pticks.has_steal_ticks) {
2311 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2312 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2313 double steal_ticks_perc = 0.0;
2314 if (total_ticks_difference != 0) {
2315 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2316 }
2317 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2318 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2319 }
2320 }
2321 }
2322
2323 void os::print_memory_info(outputStream* st) {
2324
2325 st->print("Memory:");
2326 st->print(" %dk page", os::vm_page_size()>>10);
2327
2328 // values in struct sysinfo are "unsigned long"
2329 struct sysinfo si;
2330 sysinfo(&si);
2331
2332 st->print(", physical " UINT64_FORMAT "k",
2333 os::physical_memory() >> 10);
2334 st->print("(" UINT64_FORMAT "k free)",
2335 os::available_memory() >> 10);
2336 st->print(", swap " UINT64_FORMAT "k",
2337 ((jlong)si.totalswap * si.mem_unit) >> 10);
2338 st->print("(" UINT64_FORMAT "k free)",
2339 ((jlong)si.freeswap * si.mem_unit) >> 10);
2340 st->cr();
2341 }
2342
2343 // Print the first "model name" line and the first "flags" line
2344 // that we find and nothing more. We assume "model name" comes
2345 // before "flags" so if we find a second "model name", then the
2346 // "flags" field is considered missing.
2347 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2348 #if defined(IA32) || defined(AMD64)
2349 // Other platforms have less repetitive cpuinfo files
2350 FILE *fp = fopen("/proc/cpuinfo", "r");
2351 if (fp) {
2352 while (!feof(fp)) {
2353 if (fgets(buf, buflen, fp)) {
2354 // Assume model name comes before flags
2355 bool model_name_printed = false;
2356 if (strstr(buf, "model name") != NULL) {
2357 if (!model_name_printed) {
2358 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2359 st->print_raw(buf);
2360 model_name_printed = true;
2361 } else {
2362 // model name printed but not flags? Odd, just return
2363 fclose(fp);
2364 return true;
2365 }
2366 }
2367 // print the flags line too
2368 if (strstr(buf, "flags") != NULL) {
2369 st->print_raw(buf);
2370 fclose(fp);
2371 return true;
2372 }
2373 }
2374 }
2375 fclose(fp);
2376 }
2377 #endif // x86 platforms
2378 return false;
2379 }
2380
2381 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2382 // Only print the model name if the platform provides this as a summary
2383 if (!print_model_name_and_flags(st, buf, buflen)) {
2384 st->print("\n/proc/cpuinfo:\n");
2385 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2386 st->print_cr(" <Not Available>");
2387 }
2388 }
2389 }
2390
2391 #if defined(AMD64) || defined(IA32) || defined(X32)
2392 const char* search_string = "model name";
2393 #elif defined(M68K)
2394 const char* search_string = "CPU";
2395 #elif defined(PPC64)
2396 const char* search_string = "cpu";
2397 #elif defined(S390)
2398 const char* search_string = "machine =";
2399 #elif defined(SPARC)
2400 const char* search_string = "cpu";
2401 #else
2402 const char* search_string = "Processor";
2403 #endif
2404
2405 // Parses the cpuinfo file for string representing the model name.
2406 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2407 FILE* fp = fopen("/proc/cpuinfo", "r");
2408 if (fp != NULL) {
2409 while (!feof(fp)) {
2410 char buf[256];
2411 if (fgets(buf, sizeof(buf), fp)) {
2412 char* start = strstr(buf, search_string);
2413 if (start != NULL) {
2414 char *ptr = start + strlen(search_string);
2415 char *end = buf + strlen(buf);
2416 while (ptr != end) {
2417 // skip whitespace and colon for the rest of the name.
2418 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2419 break;
2420 }
2421 ptr++;
2422 }
2423 if (ptr != end) {
2424 // reasonable string, get rid of newline and keep the rest
2425 char* nl = strchr(buf, '\n');
2426 if (nl != NULL) *nl = '\0';
2427 strncpy(cpuinfo, ptr, length);
2428 fclose(fp);
2429 return;
2430 }
2431 }
2432 }
2433 }
2434 fclose(fp);
2435 }
2436 // cpuinfo not found or parsing failed, just print generic string. The entire
2437 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2438 #if defined(AARCH64)
2439 strncpy(cpuinfo, "AArch64", length);
2440 #elif defined(AMD64)
2441 strncpy(cpuinfo, "x86_64", length);
2442 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2443 strncpy(cpuinfo, "ARM", length);
2444 #elif defined(IA32)
2445 strncpy(cpuinfo, "x86_32", length);
2446 #elif defined(IA64)
2447 strncpy(cpuinfo, "IA64", length);
2448 #elif defined(PPC)
2449 strncpy(cpuinfo, "PPC64", length);
2450 #elif defined(S390)
2451 strncpy(cpuinfo, "S390", length);
2452 #elif defined(SPARC)
2453 strncpy(cpuinfo, "sparcv9", length);
2454 #elif defined(ZERO_LIBARCH)
2455 strncpy(cpuinfo, ZERO_LIBARCH, length);
2456 #else
2457 strncpy(cpuinfo, "unknown", length);
2458 #endif
2459 }
2460
2461 static void print_signal_handler(outputStream* st, int sig,
2462 char* buf, size_t buflen);
2463
2464 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2465 st->print_cr("Signal Handlers:");
2466 print_signal_handler(st, SIGSEGV, buf, buflen);
2467 print_signal_handler(st, SIGBUS , buf, buflen);
2468 print_signal_handler(st, SIGFPE , buf, buflen);
2469 print_signal_handler(st, SIGPIPE, buf, buflen);
2470 print_signal_handler(st, SIGXFSZ, buf, buflen);
2471 print_signal_handler(st, SIGILL , buf, buflen);
2472 print_signal_handler(st, SR_signum, buf, buflen);
2473 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2474 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2475 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2476 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2477 #if defined(PPC64)
2478 print_signal_handler(st, SIGTRAP, buf, buflen);
2479 #endif
2480 }
2481
2482 static char saved_jvm_path[MAXPATHLEN] = {0};
2483
2484 // Find the full path to the current module, libjvm.so
2485 void os::jvm_path(char *buf, jint buflen) {
2486 // Error checking.
2487 if (buflen < MAXPATHLEN) {
2488 assert(false, "must use a large-enough buffer");
2489 buf[0] = '\0';
2490 return;
2491 }
2492 // Lazy resolve the path to current module.
2493 if (saved_jvm_path[0] != 0) {
2494 strcpy(buf, saved_jvm_path);
2495 return;
2496 }
2497
2498 char dli_fname[MAXPATHLEN];
2499 bool ret = dll_address_to_library_name(
2500 CAST_FROM_FN_PTR(address, os::jvm_path),
2501 dli_fname, sizeof(dli_fname), NULL);
2502 assert(ret, "cannot locate libjvm");
2503 char *rp = NULL;
2504 if (ret && dli_fname[0] != '\0') {
2505 rp = os::Posix::realpath(dli_fname, buf, buflen);
2506 }
2507 if (rp == NULL) {
2508 return;
2509 }
2510
2511 if (Arguments::sun_java_launcher_is_altjvm()) {
2512 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2513 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2514 // If "/jre/lib/" appears at the right place in the string, then
2515 // assume we are installed in a JDK and we're done. Otherwise, check
2516 // for a JAVA_HOME environment variable and fix up the path so it
2517 // looks like libjvm.so is installed there (append a fake suffix
2518 // hotspot/libjvm.so).
2519 const char *p = buf + strlen(buf) - 1;
2520 for (int count = 0; p > buf && count < 5; ++count) {
2521 for (--p; p > buf && *p != '/'; --p)
2522 /* empty */ ;
2523 }
2524
2525 if (strncmp(p, "/jre/lib/", 9) != 0) {
2526 // Look for JAVA_HOME in the environment.
2527 char* java_home_var = ::getenv("JAVA_HOME");
2528 if (java_home_var != NULL && java_home_var[0] != 0) {
2529 char* jrelib_p;
2530 int len;
2531
2532 // Check the current module name "libjvm.so".
2533 p = strrchr(buf, '/');
2534 if (p == NULL) {
2535 return;
2536 }
2537 assert(strstr(p, "/libjvm") == p, "invalid library name");
2538
2539 rp = os::Posix::realpath(java_home_var, buf, buflen);
2540 if (rp == NULL) {
2541 return;
2542 }
2543
2544 // determine if this is a legacy image or modules image
2545 // modules image doesn't have "jre" subdirectory
2546 len = strlen(buf);
2547 assert(len < buflen, "Ran out of buffer room");
2548 jrelib_p = buf + len;
2549 snprintf(jrelib_p, buflen-len, "/jre/lib");
2550 if (0 != access(buf, F_OK)) {
2551 snprintf(jrelib_p, buflen-len, "/lib");
2552 }
2553
2554 if (0 == access(buf, F_OK)) {
2555 // Use current module name "libjvm.so"
2556 len = strlen(buf);
2557 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2558 } else {
2559 // Go back to path of .so
2560 rp = os::Posix::realpath(dli_fname, buf, buflen);
2561 if (rp == NULL) {
2562 return;
2563 }
2564 }
2565 }
2566 }
2567 }
2568
2569 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2570 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2571 }
2572
2573 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2574 // no prefix required, not even "_"
2575 }
2576
2577 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2578 // no suffix required
2579 }
2580
2581 ////////////////////////////////////////////////////////////////////////////////
2582 // sun.misc.Signal support
2583
2584 static volatile jint sigint_count = 0;
2585
2586 static void UserHandler(int sig, void *siginfo, void *context) {
2587 // 4511530 - sem_post is serialized and handled by the manager thread. When
2588 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2589 // don't want to flood the manager thread with sem_post requests.
2590 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2591 return;
2592 }
2593
2594 // Ctrl-C is pressed during error reporting, likely because the error
2595 // handler fails to abort. Let VM die immediately.
2596 if (sig == SIGINT && VMError::is_error_reported()) {
2597 os::die();
2598 }
2599
2600 os::signal_notify(sig);
2601 }
2602
2603 void* os::user_handler() {
2604 return CAST_FROM_FN_PTR(void*, UserHandler);
2605 }
2606
2607 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2608 struct timespec ts;
2609 // Semaphore's are always associated with CLOCK_REALTIME
2610 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2611 // see os_posix.cpp for discussion on overflow checking
2612 if (sec >= MAX_SECS) {
2613 ts.tv_sec += MAX_SECS;
2614 ts.tv_nsec = 0;
2615 } else {
2616 ts.tv_sec += sec;
2617 ts.tv_nsec += nsec;
2618 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2619 ts.tv_nsec -= NANOSECS_PER_SEC;
2620 ++ts.tv_sec; // note: this must be <= max_secs
2621 }
2622 }
2623
2624 return ts;
2625 }
2626
2627 extern "C" {
2628 typedef void (*sa_handler_t)(int);
2629 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2630 }
2631
2632 void* os::signal(int signal_number, void* handler) {
2633 struct sigaction sigAct, oldSigAct;
2634
2635 sigfillset(&(sigAct.sa_mask));
2636 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2637 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2638
2639 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2640 // -1 means registration failed
2641 return (void *)-1;
2642 }
2643
2644 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2645 }
2646
2647 void os::signal_raise(int signal_number) {
2648 ::raise(signal_number);
2649 }
2650
2651 // The following code is moved from os.cpp for making this
2652 // code platform specific, which it is by its very nature.
2653
2654 // Will be modified when max signal is changed to be dynamic
2655 int os::sigexitnum_pd() {
2656 return NSIG;
2657 }
2658
2659 // a counter for each possible signal value
2660 static volatile jint pending_signals[NSIG+1] = { 0 };
2661
2662 // Linux(POSIX) specific hand shaking semaphore.
2663 static Semaphore* sig_sem = NULL;
2664 static PosixSemaphore sr_semaphore;
2665
2666 static void jdk_misc_signal_init() {
2667 // Initialize signal structures
2668 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2669
2670 // Initialize signal semaphore
2671 sig_sem = new Semaphore();
2672 }
2673
2674 void os::signal_notify(int sig) {
2675 if (sig_sem != NULL) {
2676 Atomic::inc(&pending_signals[sig]);
2677 sig_sem->signal();
2678 } else {
2679 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2680 // initialization isn't called.
2681 assert(ReduceSignalUsage, "signal semaphore should be created");
2682 }
2683 }
2684
2685 static int check_pending_signals() {
2686 Atomic::store(0, &sigint_count);
2687 for (;;) {
2688 for (int i = 0; i < NSIG + 1; i++) {
2689 jint n = pending_signals[i];
2690 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2691 return i;
2692 }
2693 }
2694 JavaThread *thread = JavaThread::current();
2695 ThreadBlockInVM tbivm(thread);
2696
2697 bool threadIsSuspended;
2698 do {
2699 thread->set_suspend_equivalent();
2700 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2701 sig_sem->wait();
2702
2703 // were we externally suspended while we were waiting?
2704 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2705 if (threadIsSuspended) {
2706 // The semaphore has been incremented, but while we were waiting
2707 // another thread suspended us. We don't want to continue running
2708 // while suspended because that would surprise the thread that
2709 // suspended us.
2710 sig_sem->signal();
2711
2712 thread->java_suspend_self();
2713 }
2714 } while (threadIsSuspended);
2715 }
2716 }
2717
2718 int os::signal_wait() {
2719 return check_pending_signals();
2720 }
2721
2722 ////////////////////////////////////////////////////////////////////////////////
2723 // Virtual Memory
2724
2725 int os::vm_page_size() {
2726 // Seems redundant as all get out
2727 assert(os::Linux::page_size() != -1, "must call os::init");
2728 return os::Linux::page_size();
2729 }
2730
2731 // Solaris allocates memory by pages.
2732 int os::vm_allocation_granularity() {
2733 assert(os::Linux::page_size() != -1, "must call os::init");
2734 return os::Linux::page_size();
2735 }
2736
2737 // Rationale behind this function:
2738 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2739 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2740 // samples for JITted code. Here we create private executable mapping over the code cache
2741 // and then we can use standard (well, almost, as mapping can change) way to provide
2742 // info for the reporting script by storing timestamp and location of symbol
2743 void linux_wrap_code(char* base, size_t size) {
2744 static volatile jint cnt = 0;
2745
2746 if (!UseOprofile) {
2747 return;
2748 }
2749
2750 char buf[PATH_MAX+1];
2751 int num = Atomic::add(1, &cnt);
2752
2753 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2754 os::get_temp_directory(), os::current_process_id(), num);
2755 unlink(buf);
2756
2757 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2758
2759 if (fd != -1) {
2760 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2761 if (rv != (off_t)-1) {
2762 if (::write(fd, "", 1) == 1) {
2763 mmap(base, size,
2764 PROT_READ|PROT_WRITE|PROT_EXEC,
2765 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2766 }
2767 }
2768 ::close(fd);
2769 unlink(buf);
2770 }
2771 }
2772
2773 static bool recoverable_mmap_error(int err) {
2774 // See if the error is one we can let the caller handle. This
2775 // list of errno values comes from JBS-6843484. I can't find a
2776 // Linux man page that documents this specific set of errno
2777 // values so while this list currently matches Solaris, it may
2778 // change as we gain experience with this failure mode.
2779 switch (err) {
2780 case EBADF:
2781 case EINVAL:
2782 case ENOTSUP:
2783 // let the caller deal with these errors
2784 return true;
2785
2786 default:
2787 // Any remaining errors on this OS can cause our reserved mapping
2788 // to be lost. That can cause confusion where different data
2789 // structures think they have the same memory mapped. The worst
2790 // scenario is if both the VM and a library think they have the
2791 // same memory mapped.
2792 return false;
2793 }
2794 }
2795
2796 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2797 int err) {
2798 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2799 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2800 os::strerror(err), err);
2801 }
2802
2803 static void warn_fail_commit_memory(char* addr, size_t size,
2804 size_t alignment_hint, bool exec,
2805 int err) {
2806 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2807 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2808 alignment_hint, exec, os::strerror(err), err);
2809 }
2810
2811 // NOTE: Linux kernel does not really reserve the pages for us.
2812 // All it does is to check if there are enough free pages
2813 // left at the time of mmap(). This could be a potential
2814 // problem.
2815 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2816 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2817 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2818 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2819 if (res != (uintptr_t) MAP_FAILED) {
2820 if (UseNUMAInterleaving) {
2821 numa_make_global(addr, size);
2822 }
2823 return 0;
2824 }
2825
2826 int err = errno; // save errno from mmap() call above
2827
2828 if (!recoverable_mmap_error(err)) {
2829 warn_fail_commit_memory(addr, size, exec, err);
2830 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2831 }
2832
2833 return err;
2834 }
2835
2836 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2837 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2838 }
2839
2840 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2841 const char* mesg) {
2842 assert(mesg != NULL, "mesg must be specified");
2843 int err = os::Linux::commit_memory_impl(addr, size, exec);
2844 if (err != 0) {
2845 // the caller wants all commit errors to exit with the specified mesg:
2846 warn_fail_commit_memory(addr, size, exec, err);
2847 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2848 }
2849 }
2850
2851 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2852 #ifndef MAP_HUGETLB
2853 #define MAP_HUGETLB 0x40000
2854 #endif
2855
2856 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2857 #ifndef MADV_HUGEPAGE
2858 #define MADV_HUGEPAGE 14
2859 #endif
2860
2861 int os::Linux::commit_memory_impl(char* addr, size_t size,
2862 size_t alignment_hint, bool exec) {
2863 int err = os::Linux::commit_memory_impl(addr, size, exec);
2864 if (err == 0) {
2865 realign_memory(addr, size, alignment_hint);
2866 }
2867 return err;
2868 }
2869
2870 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2871 bool exec) {
2872 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2873 }
2874
2875 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2876 size_t alignment_hint, bool exec,
2877 const char* mesg) {
2878 assert(mesg != NULL, "mesg must be specified");
2879 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2880 if (err != 0) {
2881 // the caller wants all commit errors to exit with the specified mesg:
2882 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2883 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2884 }
2885 }
2886
2887 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2888 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2889 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2890 // be supported or the memory may already be backed by huge pages.
2891 ::madvise(addr, bytes, MADV_HUGEPAGE);
2892 }
2893 }
2894
2895 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2896 // This method works by doing an mmap over an existing mmaping and effectively discarding
2897 // the existing pages. However it won't work for SHM-based large pages that cannot be
2898 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2899 // small pages on top of the SHM segment. This method always works for small pages, so we
2900 // allow that in any case.
2901 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2902 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2903 }
2904 }
2905
2906 void os::numa_make_global(char *addr, size_t bytes) {
2907 Linux::numa_interleave_memory(addr, bytes);
2908 }
2909
2910 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2911 // bind policy to MPOL_PREFERRED for the current thread.
2912 #define USE_MPOL_PREFERRED 0
2913
2914 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2915 // To make NUMA and large pages more robust when both enabled, we need to ease
2916 // the requirements on where the memory should be allocated. MPOL_BIND is the
2917 // default policy and it will force memory to be allocated on the specified
2918 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2919 // the specified node, but will not force it. Using this policy will prevent
2920 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2921 // free large pages.
2922 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2923 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2924 }
2925
2926 bool os::numa_topology_changed() { return false; }
2927
2928 size_t os::numa_get_groups_num() {
2929 // Return just the number of nodes in which it's possible to allocate memory
2930 // (in numa terminology, configured nodes).
2931 return Linux::numa_num_configured_nodes();
2932 }
2933
2934 int os::numa_get_group_id() {
2935 int cpu_id = Linux::sched_getcpu();
2936 if (cpu_id != -1) {
2937 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2938 if (lgrp_id != -1) {
2939 return lgrp_id;
2940 }
2941 }
2942 return 0;
2943 }
2944
2945 int os::Linux::get_existing_num_nodes() {
2946 int node;
2947 int highest_node_number = Linux::numa_max_node();
2948 int num_nodes = 0;
2949
2950 // Get the total number of nodes in the system including nodes without memory.
2951 for (node = 0; node <= highest_node_number; node++) {
2952 if (isnode_in_existing_nodes(node)) {
2953 num_nodes++;
2954 }
2955 }
2956 return num_nodes;
2957 }
2958
2959 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2960 int highest_node_number = Linux::numa_max_node();
2961 size_t i = 0;
2962
2963 // Map all node ids in which it is possible to allocate memory. Also nodes are
2964 // not always consecutively available, i.e. available from 0 to the highest
2965 // node number. If the nodes have been bound explicitly using numactl membind,
2966 // then allocate memory from those nodes only.
2967 for (int node = 0; node <= highest_node_number; node++) {
2968 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2969 ids[i++] = node;
2970 }
2971 }
2972 return i;
2973 }
2974
2975 bool os::get_page_info(char *start, page_info* info) {
2976 return false;
2977 }
2978
2979 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2980 page_info* page_found) {
2981 return end;
2982 }
2983
2984
2985 int os::Linux::sched_getcpu_syscall(void) {
2986 unsigned int cpu = 0;
2987 int retval = -1;
2988
2989 #if defined(IA32)
2990 #ifndef SYS_getcpu
2991 #define SYS_getcpu 318
2992 #endif
2993 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2994 #elif defined(AMD64)
2995 // Unfortunately we have to bring all these macros here from vsyscall.h
2996 // to be able to compile on old linuxes.
2997 #define __NR_vgetcpu 2
2998 #define VSYSCALL_START (-10UL << 20)
2999 #define VSYSCALL_SIZE 1024
3000 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3001 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3002 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3003 retval = vgetcpu(&cpu, NULL, NULL);
3004 #endif
3005
3006 return (retval == -1) ? retval : cpu;
3007 }
3008
3009 void os::Linux::sched_getcpu_init() {
3010 // sched_getcpu() should be in libc.
3011 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3012 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3013
3014 // If it's not, try a direct syscall.
3015 if (sched_getcpu() == -1) {
3016 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3017 (void*)&sched_getcpu_syscall));
3018 }
3019
3020 if (sched_getcpu() == -1) {
3021 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3022 }
3023 }
3024
3025 // Something to do with the numa-aware allocator needs these symbols
3026 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3027 extern "C" JNIEXPORT void numa_error(char *where) { }
3028
3029 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3030 // load symbol from base version instead.
3031 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3032 void *f = dlvsym(handle, name, "libnuma_1.1");
3033 if (f == NULL) {
3034 f = dlsym(handle, name);
3035 }
3036 return f;
3037 }
3038
3039 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3040 // Return NULL if the symbol is not defined in this particular version.
3041 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3042 return dlvsym(handle, name, "libnuma_1.2");
3043 }
3044
3045 bool os::Linux::libnuma_init() {
3046 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3047 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3048 if (handle != NULL) {
3049 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3050 libnuma_dlsym(handle, "numa_node_to_cpus")));
3051 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3052 libnuma_dlsym(handle, "numa_max_node")));
3053 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3054 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3055 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3056 libnuma_dlsym(handle, "numa_available")));
3057 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3058 libnuma_dlsym(handle, "numa_tonode_memory")));
3059 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3060 libnuma_dlsym(handle, "numa_interleave_memory")));
3061 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3062 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3063 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3064 libnuma_dlsym(handle, "numa_set_bind_policy")));
3065 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3066 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3067 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3068 libnuma_dlsym(handle, "numa_distance")));
3069 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3070 libnuma_v2_dlsym(handle, "numa_get_membind")));
3071
3072 if (numa_available() != -1) {
3073 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3074 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3075 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3076 // Create an index -> node mapping, since nodes are not always consecutive
3077 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3078 rebuild_nindex_to_node_map();
3079 // Create a cpu -> node mapping
3080 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3081 rebuild_cpu_to_node_map();
3082 return true;
3083 }
3084 }
3085 }
3086 return false;
3087 }
3088
3089 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3090 // Creating guard page is very expensive. Java thread has HotSpot
3091 // guard pages, only enable glibc guard page for non-Java threads.
3092 // (Remember: compiler thread is a Java thread, too!)
3093 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3094 }
3095
3096 void os::Linux::rebuild_nindex_to_node_map() {
3097 int highest_node_number = Linux::numa_max_node();
3098
3099 nindex_to_node()->clear();
3100 for (int node = 0; node <= highest_node_number; node++) {
3101 if (Linux::isnode_in_existing_nodes(node)) {
3102 nindex_to_node()->append(node);
3103 }
3104 }
3105 }
3106
3107 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3108 // The table is later used in get_node_by_cpu().
3109 void os::Linux::rebuild_cpu_to_node_map() {
3110 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3111 // in libnuma (possible values are starting from 16,
3112 // and continuing up with every other power of 2, but less
3113 // than the maximum number of CPUs supported by kernel), and
3114 // is a subject to change (in libnuma version 2 the requirements
3115 // are more reasonable) we'll just hardcode the number they use
3116 // in the library.
3117 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3118
3119 size_t cpu_num = processor_count();
3120 size_t cpu_map_size = NCPUS / BitsPerCLong;
3121 size_t cpu_map_valid_size =
3122 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3123
3124 cpu_to_node()->clear();
3125 cpu_to_node()->at_grow(cpu_num - 1);
3126
3127 size_t node_num = get_existing_num_nodes();
3128
3129 int distance = 0;
3130 int closest_distance = INT_MAX;
3131 int closest_node = 0;
3132 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3133 for (size_t i = 0; i < node_num; i++) {
3134 // Check if node is configured (not a memory-less node). If it is not, find
3135 // the closest configured node. Check also if node is bound, i.e. it's allowed
3136 // to allocate memory from the node. If it's not allowed, map cpus in that node
3137 // to the closest node from which memory allocation is allowed.
3138 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
3139 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
3140 closest_distance = INT_MAX;
3141 // Check distance from all remaining nodes in the system. Ignore distance
3142 // from itself, from another non-configured node, and from another non-bound
3143 // node.
3144 for (size_t m = 0; m < node_num; m++) {
3145 if (m != i &&
3146 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
3147 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
3148 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3149 // If a closest node is found, update. There is always at least one
3150 // configured and bound node in the system so there is always at least
3151 // one node close.
3152 if (distance != 0 && distance < closest_distance) {
3153 closest_distance = distance;
3154 closest_node = nindex_to_node()->at(m);
3155 }
3156 }
3157 }
3158 } else {
3159 // Current node is already a configured node.
3160 closest_node = nindex_to_node()->at(i);
3161 }
3162
3163 // Get cpus from the original node and map them to the closest node. If node
3164 // is a configured node (not a memory-less node), then original node and
3165 // closest node are the same.
3166 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3167 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3168 if (cpu_map[j] != 0) {
3169 for (size_t k = 0; k < BitsPerCLong; k++) {
3170 if (cpu_map[j] & (1UL << k)) {
3171 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3172 }
3173 }
3174 }
3175 }
3176 }
3177 }
3178 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3179 }
3180
3181 int os::Linux::get_node_by_cpu(int cpu_id) {
3182 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3183 return cpu_to_node()->at(cpu_id);
3184 }
3185 return -1;
3186 }
3187
3188 GrowableArray<int>* os::Linux::_cpu_to_node;
3189 GrowableArray<int>* os::Linux::_nindex_to_node;
3190 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3191 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3192 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3193 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3194 os::Linux::numa_available_func_t os::Linux::_numa_available;
3195 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3196 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3197 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3198 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3199 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3200 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3201 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3202 unsigned long* os::Linux::_numa_all_nodes;
3203 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3204 struct bitmask* os::Linux::_numa_nodes_ptr;
3205
3206 bool os::pd_uncommit_memory(char* addr, size_t size) {
3207 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3208 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3209 return res != (uintptr_t) MAP_FAILED;
3210 }
3211
3212 static address get_stack_commited_bottom(address bottom, size_t size) {
3213 address nbot = bottom;
3214 address ntop = bottom + size;
3215
3216 size_t page_sz = os::vm_page_size();
3217 unsigned pages = size / page_sz;
3218
3219 unsigned char vec[1];
3220 unsigned imin = 1, imax = pages + 1, imid;
3221 int mincore_return_value = 0;
3222
3223 assert(imin <= imax, "Unexpected page size");
3224
3225 while (imin < imax) {
3226 imid = (imax + imin) / 2;
3227 nbot = ntop - (imid * page_sz);
3228
3229 // Use a trick with mincore to check whether the page is mapped or not.
3230 // mincore sets vec to 1 if page resides in memory and to 0 if page
3231 // is swapped output but if page we are asking for is unmapped
3232 // it returns -1,ENOMEM
3233 mincore_return_value = mincore(nbot, page_sz, vec);
3234
3235 if (mincore_return_value == -1) {
3236 // Page is not mapped go up
3237 // to find first mapped page
3238 if (errno != EAGAIN) {
3239 assert(errno == ENOMEM, "Unexpected mincore errno");
3240 imax = imid;
3241 }
3242 } else {
3243 // Page is mapped go down
3244 // to find first not mapped page
3245 imin = imid + 1;
3246 }
3247 }
3248
3249 nbot = nbot + page_sz;
3250
3251 // Adjust stack bottom one page up if last checked page is not mapped
3252 if (mincore_return_value == -1) {
3253 nbot = nbot + page_sz;
3254 }
3255
3256 return nbot;
3257 }
3258
3259 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3260 int mincore_return_value;
3261 const size_t stripe = 1024; // query this many pages each time
3262 unsigned char vec[stripe + 1];
3263 // set a guard
3264 vec[stripe] = 'X';
3265
3266 const size_t page_sz = os::vm_page_size();
3267 size_t pages = size / page_sz;
3268
3269 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3270 assert(is_aligned(size, page_sz), "Size must be page aligned");
3271
3272 committed_start = NULL;
3273
3274 int loops = (pages + stripe - 1) / stripe;
3275 int committed_pages = 0;
3276 address loop_base = start;
3277 bool found_range = false;
3278
3279 for (int index = 0; index < loops && !found_range; index ++) {
3280 assert(pages > 0, "Nothing to do");
3281 int pages_to_query = (pages >= stripe) ? stripe : pages;
3282 pages -= pages_to_query;
3283
3284 // Get stable read
3285 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3286
3287 // During shutdown, some memory goes away without properly notifying NMT,
3288 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3289 // Bailout and return as not committed for now.
3290 if (mincore_return_value == -1 && errno == ENOMEM) {
3291 return false;
3292 }
3293
3294 assert(vec[stripe] == 'X', "overflow guard");
3295 assert(mincore_return_value == 0, "Range must be valid");
3296 // Process this stripe
3297 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3298 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3299 // End of current contiguous region
3300 if (committed_start != NULL) {
3301 found_range = true;
3302 break;
3303 }
3304 } else { // committed
3305 // Start of region
3306 if (committed_start == NULL) {
3307 committed_start = loop_base + page_sz * vecIdx;
3308 }
3309 committed_pages ++;
3310 }
3311 }
3312
3313 loop_base += pages_to_query * page_sz;
3314 }
3315
3316 if (committed_start != NULL) {
3317 assert(committed_pages > 0, "Must have committed region");
3318 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3319 assert(committed_start >= start && committed_start < start + size, "Out of range");
3320 committed_size = page_sz * committed_pages;
3321 return true;
3322 } else {
3323 assert(committed_pages == 0, "Should not have committed region");
3324 return false;
3325 }
3326 }
3327
3328
3329 // Linux uses a growable mapping for the stack, and if the mapping for
3330 // the stack guard pages is not removed when we detach a thread the
3331 // stack cannot grow beyond the pages where the stack guard was
3332 // mapped. If at some point later in the process the stack expands to
3333 // that point, the Linux kernel cannot expand the stack any further
3334 // because the guard pages are in the way, and a segfault occurs.
3335 //
3336 // However, it's essential not to split the stack region by unmapping
3337 // a region (leaving a hole) that's already part of the stack mapping,
3338 // so if the stack mapping has already grown beyond the guard pages at
3339 // the time we create them, we have to truncate the stack mapping.
3340 // So, we need to know the extent of the stack mapping when
3341 // create_stack_guard_pages() is called.
3342
3343 // We only need this for stacks that are growable: at the time of
3344 // writing thread stacks don't use growable mappings (i.e. those
3345 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3346 // only applies to the main thread.
3347
3348 // If the (growable) stack mapping already extends beyond the point
3349 // where we're going to put our guard pages, truncate the mapping at
3350 // that point by munmap()ping it. This ensures that when we later
3351 // munmap() the guard pages we don't leave a hole in the stack
3352 // mapping. This only affects the main/primordial thread
3353
3354 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3355 if (os::is_primordial_thread()) {
3356 // As we manually grow stack up to bottom inside create_attached_thread(),
3357 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3358 // we don't need to do anything special.
3359 // Check it first, before calling heavy function.
3360 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3361 unsigned char vec[1];
3362
3363 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3364 // Fallback to slow path on all errors, including EAGAIN
3365 stack_extent = (uintptr_t) get_stack_commited_bottom(
3366 os::Linux::initial_thread_stack_bottom(),
3367 (size_t)addr - stack_extent);
3368 }
3369
3370 if (stack_extent < (uintptr_t)addr) {
3371 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3372 }
3373 }
3374
3375 return os::commit_memory(addr, size, !ExecMem);
3376 }
3377
3378 // If this is a growable mapping, remove the guard pages entirely by
3379 // munmap()ping them. If not, just call uncommit_memory(). This only
3380 // affects the main/primordial thread, but guard against future OS changes.
3381 // It's safe to always unmap guard pages for primordial thread because we
3382 // always place it right after end of the mapped region.
3383
3384 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3385 uintptr_t stack_extent, stack_base;
3386
3387 if (os::is_primordial_thread()) {
3388 return ::munmap(addr, size) == 0;
3389 }
3390
3391 return os::uncommit_memory(addr, size);
3392 }
3393
3394 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3395 // at 'requested_addr'. If there are existing memory mappings at the same
3396 // location, however, they will be overwritten. If 'fixed' is false,
3397 // 'requested_addr' is only treated as a hint, the return value may or
3398 // may not start from the requested address. Unlike Linux mmap(), this
3399 // function returns NULL to indicate failure.
3400 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3401 char * addr;
3402 int flags;
3403
3404 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3405 if (fixed) {
3406 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3407 flags |= MAP_FIXED;
3408 }
3409
3410 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3411 // touch an uncommitted page. Otherwise, the read/write might
3412 // succeed if we have enough swap space to back the physical page.
3413 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3414 flags, -1, 0);
3415
3416 return addr == MAP_FAILED ? NULL : addr;
3417 }
3418
3419 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3420 // (req_addr != NULL) or with a given alignment.
3421 // - bytes shall be a multiple of alignment.
3422 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3423 // - alignment sets the alignment at which memory shall be allocated.
3424 // It must be a multiple of allocation granularity.
3425 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3426 // req_addr or NULL.
3427 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3428
3429 size_t extra_size = bytes;
3430 if (req_addr == NULL && alignment > 0) {
3431 extra_size += alignment;
3432 }
3433
3434 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3435 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3436 -1, 0);
3437 if (start == MAP_FAILED) {
3438 start = NULL;
3439 } else {
3440 if (req_addr != NULL) {
3441 if (start != req_addr) {
3442 ::munmap(start, extra_size);
3443 start = NULL;
3444 }
3445 } else {
3446 char* const start_aligned = align_up(start, alignment);
3447 char* const end_aligned = start_aligned + bytes;
3448 char* const end = start + extra_size;
3449 if (start_aligned > start) {
3450 ::munmap(start, start_aligned - start);
3451 }
3452 if (end_aligned < end) {
3453 ::munmap(end_aligned, end - end_aligned);
3454 }
3455 start = start_aligned;
3456 }
3457 }
3458 return start;
3459 }
3460
3461 static int anon_munmap(char * addr, size_t size) {
3462 return ::munmap(addr, size) == 0;
3463 }
3464
3465 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3466 size_t alignment_hint) {
3467 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3468 }
3469
3470 bool os::pd_release_memory(char* addr, size_t size) {
3471 return anon_munmap(addr, size);
3472 }
3473
3474 static bool linux_mprotect(char* addr, size_t size, int prot) {
3475 // Linux wants the mprotect address argument to be page aligned.
3476 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3477
3478 // According to SUSv3, mprotect() should only be used with mappings
3479 // established by mmap(), and mmap() always maps whole pages. Unaligned
3480 // 'addr' likely indicates problem in the VM (e.g. trying to change
3481 // protection of malloc'ed or statically allocated memory). Check the
3482 // caller if you hit this assert.
3483 assert(addr == bottom, "sanity check");
3484
3485 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3486 return ::mprotect(bottom, size, prot) == 0;
3487 }
3488
3489 // Set protections specified
3490 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3491 bool is_committed) {
3492 unsigned int p = 0;
3493 switch (prot) {
3494 case MEM_PROT_NONE: p = PROT_NONE; break;
3495 case MEM_PROT_READ: p = PROT_READ; break;
3496 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3497 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3498 default:
3499 ShouldNotReachHere();
3500 }
3501 // is_committed is unused.
3502 return linux_mprotect(addr, bytes, p);
3503 }
3504
3505 bool os::guard_memory(char* addr, size_t size) {
3506 return linux_mprotect(addr, size, PROT_NONE);
3507 }
3508
3509 bool os::unguard_memory(char* addr, size_t size) {
3510 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3511 }
3512
3513 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3514 size_t page_size) {
3515 bool result = false;
3516 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3517 MAP_ANONYMOUS|MAP_PRIVATE,
3518 -1, 0);
3519 if (p != MAP_FAILED) {
3520 void *aligned_p = align_up(p, page_size);
3521
3522 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3523
3524 munmap(p, page_size * 2);
3525 }
3526
3527 if (warn && !result) {
3528 warning("TransparentHugePages is not supported by the operating system.");
3529 }
3530
3531 return result;
3532 }
3533
3534 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3535 bool result = false;
3536 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3537 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3538 -1, 0);
3539
3540 if (p != MAP_FAILED) {
3541 // We don't know if this really is a huge page or not.
3542 FILE *fp = fopen("/proc/self/maps", "r");
3543 if (fp) {
3544 while (!feof(fp)) {
3545 char chars[257];
3546 long x = 0;
3547 if (fgets(chars, sizeof(chars), fp)) {
3548 if (sscanf(chars, "%lx-%*x", &x) == 1
3549 && x == (long)p) {
3550 if (strstr (chars, "hugepage")) {
3551 result = true;
3552 break;
3553 }
3554 }
3555 }
3556 }
3557 fclose(fp);
3558 }
3559 munmap(p, page_size);
3560 }
3561
3562 if (warn && !result) {
3563 warning("HugeTLBFS is not supported by the operating system.");
3564 }
3565
3566 return result;
3567 }
3568
3569 // From the coredump_filter documentation:
3570 //
3571 // - (bit 0) anonymous private memory
3572 // - (bit 1) anonymous shared memory
3573 // - (bit 2) file-backed private memory
3574 // - (bit 3) file-backed shared memory
3575 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3576 // effective only if the bit 2 is cleared)
3577 // - (bit 5) hugetlb private memory
3578 // - (bit 6) hugetlb shared memory
3579 // - (bit 7) dax private memory
3580 // - (bit 8) dax shared memory
3581 //
3582 static void set_coredump_filter(CoredumpFilterBit bit) {
3583 FILE *f;
3584 long cdm;
3585
3586 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3587 return;
3588 }
3589
3590 if (fscanf(f, "%lx", &cdm) != 1) {
3591 fclose(f);
3592 return;
3593 }
3594
3595 long saved_cdm = cdm;
3596 rewind(f);
3597 cdm |= bit;
3598
3599 if (cdm != saved_cdm) {
3600 fprintf(f, "%#lx", cdm);
3601 }
3602
3603 fclose(f);
3604 }
3605
3606 // Large page support
3607
3608 static size_t _large_page_size = 0;
3609
3610 size_t os::Linux::find_large_page_size() {
3611 size_t large_page_size = 0;
3612
3613 // large_page_size on Linux is used to round up heap size. x86 uses either
3614 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3615 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3616 // page as large as 256M.
3617 //
3618 // Here we try to figure out page size by parsing /proc/meminfo and looking
3619 // for a line with the following format:
3620 // Hugepagesize: 2048 kB
3621 //
3622 // If we can't determine the value (e.g. /proc is not mounted, or the text
3623 // format has been changed), we'll use the largest page size supported by
3624 // the processor.
3625
3626 #ifndef ZERO
3627 large_page_size =
3628 AARCH64_ONLY(2 * M)
3629 AMD64_ONLY(2 * M)
3630 ARM32_ONLY(2 * M)
3631 IA32_ONLY(4 * M)
3632 IA64_ONLY(256 * M)
3633 PPC_ONLY(4 * M)
3634 S390_ONLY(1 * M)
3635 SPARC_ONLY(4 * M);
3636 #endif // ZERO
3637
3638 FILE *fp = fopen("/proc/meminfo", "r");
3639 if (fp) {
3640 while (!feof(fp)) {
3641 int x = 0;
3642 char buf[16];
3643 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3644 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3645 large_page_size = x * K;
3646 break;
3647 }
3648 } else {
3649 // skip to next line
3650 for (;;) {
3651 int ch = fgetc(fp);
3652 if (ch == EOF || ch == (int)'\n') break;
3653 }
3654 }
3655 }
3656 fclose(fp);
3657 }
3658
3659 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3660 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3661 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3662 proper_unit_for_byte_size(large_page_size));
3663 }
3664
3665 return large_page_size;
3666 }
3667
3668 size_t os::Linux::setup_large_page_size() {
3669 _large_page_size = Linux::find_large_page_size();
3670 const size_t default_page_size = (size_t)Linux::page_size();
3671 if (_large_page_size > default_page_size) {
3672 _page_sizes[0] = _large_page_size;
3673 _page_sizes[1] = default_page_size;
3674 _page_sizes[2] = 0;
3675 }
3676
3677 return _large_page_size;
3678 }
3679
3680 bool os::Linux::setup_large_page_type(size_t page_size) {
3681 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3682 FLAG_IS_DEFAULT(UseSHM) &&
3683 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3684
3685 // The type of large pages has not been specified by the user.
3686
3687 // Try UseHugeTLBFS and then UseSHM.
3688 UseHugeTLBFS = UseSHM = true;
3689
3690 // Don't try UseTransparentHugePages since there are known
3691 // performance issues with it turned on. This might change in the future.
3692 UseTransparentHugePages = false;
3693 }
3694
3695 if (UseTransparentHugePages) {
3696 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3697 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3698 UseHugeTLBFS = false;
3699 UseSHM = false;
3700 return true;
3701 }
3702 UseTransparentHugePages = false;
3703 }
3704
3705 if (UseHugeTLBFS) {
3706 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3707 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3708 UseSHM = false;
3709 return true;
3710 }
3711 UseHugeTLBFS = false;
3712 }
3713
3714 return UseSHM;
3715 }
3716
3717 void os::large_page_init() {
3718 if (!UseLargePages &&
3719 !UseTransparentHugePages &&
3720 !UseHugeTLBFS &&
3721 !UseSHM) {
3722 // Not using large pages.
3723 return;
3724 }
3725
3726 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3727 // The user explicitly turned off large pages.
3728 // Ignore the rest of the large pages flags.
3729 UseTransparentHugePages = false;
3730 UseHugeTLBFS = false;
3731 UseSHM = false;
3732 return;
3733 }
3734
3735 size_t large_page_size = Linux::setup_large_page_size();
3736 UseLargePages = Linux::setup_large_page_type(large_page_size);
3737
3738 set_coredump_filter(LARGEPAGES_BIT);
3739 }
3740
3741 #ifndef SHM_HUGETLB
3742 #define SHM_HUGETLB 04000
3743 #endif
3744
3745 #define shm_warning_format(format, ...) \
3746 do { \
3747 if (UseLargePages && \
3748 (!FLAG_IS_DEFAULT(UseLargePages) || \
3749 !FLAG_IS_DEFAULT(UseSHM) || \
3750 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3751 warning(format, __VA_ARGS__); \
3752 } \
3753 } while (0)
3754
3755 #define shm_warning(str) shm_warning_format("%s", str)
3756
3757 #define shm_warning_with_errno(str) \
3758 do { \
3759 int err = errno; \
3760 shm_warning_format(str " (error = %d)", err); \
3761 } while (0)
3762
3763 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3764 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3765
3766 if (!is_aligned(alignment, SHMLBA)) {
3767 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3768 return NULL;
3769 }
3770
3771 // To ensure that we get 'alignment' aligned memory from shmat,
3772 // we pre-reserve aligned virtual memory and then attach to that.
3773
3774 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3775 if (pre_reserved_addr == NULL) {
3776 // Couldn't pre-reserve aligned memory.
3777 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3778 return NULL;
3779 }
3780
3781 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3782 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3783
3784 if ((intptr_t)addr == -1) {
3785 int err = errno;
3786 shm_warning_with_errno("Failed to attach shared memory.");
3787
3788 assert(err != EACCES, "Unexpected error");
3789 assert(err != EIDRM, "Unexpected error");
3790 assert(err != EINVAL, "Unexpected error");
3791
3792 // Since we don't know if the kernel unmapped the pre-reserved memory area
3793 // we can't unmap it, since that would potentially unmap memory that was
3794 // mapped from other threads.
3795 return NULL;
3796 }
3797
3798 return addr;
3799 }
3800
3801 static char* shmat_at_address(int shmid, char* req_addr) {
3802 if (!is_aligned(req_addr, SHMLBA)) {
3803 assert(false, "Requested address needs to be SHMLBA aligned");
3804 return NULL;
3805 }
3806
3807 char* addr = (char*)shmat(shmid, req_addr, 0);
3808
3809 if ((intptr_t)addr == -1) {
3810 shm_warning_with_errno("Failed to attach shared memory.");
3811 return NULL;
3812 }
3813
3814 return addr;
3815 }
3816
3817 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3818 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3819 if (req_addr != NULL) {
3820 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3821 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3822 return shmat_at_address(shmid, req_addr);
3823 }
3824
3825 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3826 // return large page size aligned memory addresses when req_addr == NULL.
3827 // However, if the alignment is larger than the large page size, we have
3828 // to manually ensure that the memory returned is 'alignment' aligned.
3829 if (alignment > os::large_page_size()) {
3830 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3831 return shmat_with_alignment(shmid, bytes, alignment);
3832 } else {
3833 return shmat_at_address(shmid, NULL);
3834 }
3835 }
3836
3837 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3838 char* req_addr, bool exec) {
3839 // "exec" is passed in but not used. Creating the shared image for
3840 // the code cache doesn't have an SHM_X executable permission to check.
3841 assert(UseLargePages && UseSHM, "only for SHM large pages");
3842 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3843 assert(is_aligned(req_addr, alignment), "Unaligned address");
3844
3845 if (!is_aligned(bytes, os::large_page_size())) {
3846 return NULL; // Fallback to small pages.
3847 }
3848
3849 // Create a large shared memory region to attach to based on size.
3850 // Currently, size is the total size of the heap.
3851 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3852 if (shmid == -1) {
3853 // Possible reasons for shmget failure:
3854 // 1. shmmax is too small for Java heap.
3855 // > check shmmax value: cat /proc/sys/kernel/shmmax
3856 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3857 // 2. not enough large page memory.
3858 // > check available large pages: cat /proc/meminfo
3859 // > increase amount of large pages:
3860 // echo new_value > /proc/sys/vm/nr_hugepages
3861 // Note 1: different Linux may use different name for this property,
3862 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3863 // Note 2: it's possible there's enough physical memory available but
3864 // they are so fragmented after a long run that they can't
3865 // coalesce into large pages. Try to reserve large pages when
3866 // the system is still "fresh".
3867 shm_warning_with_errno("Failed to reserve shared memory.");
3868 return NULL;
3869 }
3870
3871 // Attach to the region.
3872 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3873
3874 // Remove shmid. If shmat() is successful, the actual shared memory segment
3875 // will be deleted when it's detached by shmdt() or when the process
3876 // terminates. If shmat() is not successful this will remove the shared
3877 // segment immediately.
3878 shmctl(shmid, IPC_RMID, NULL);
3879
3880 return addr;
3881 }
3882
3883 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3884 int error) {
3885 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3886
3887 bool warn_on_failure = UseLargePages &&
3888 (!FLAG_IS_DEFAULT(UseLargePages) ||
3889 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3890 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3891
3892 if (warn_on_failure) {
3893 char msg[128];
3894 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3895 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3896 warning("%s", msg);
3897 }
3898 }
3899
3900 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3901 char* req_addr,
3902 bool exec) {
3903 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3904 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3905 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3906
3907 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3908 char* addr = (char*)::mmap(req_addr, bytes, prot,
3909 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3910 -1, 0);
3911
3912 if (addr == MAP_FAILED) {
3913 warn_on_large_pages_failure(req_addr, bytes, errno);
3914 return NULL;
3915 }
3916
3917 assert(is_aligned(addr, os::large_page_size()), "Must be");
3918
3919 return addr;
3920 }
3921
3922 // Reserve memory using mmap(MAP_HUGETLB).
3923 // - bytes shall be a multiple of alignment.
3924 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3925 // - alignment sets the alignment at which memory shall be allocated.
3926 // It must be a multiple of allocation granularity.
3927 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3928 // req_addr or NULL.
3929 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3930 size_t alignment,
3931 char* req_addr,
3932 bool exec) {
3933 size_t large_page_size = os::large_page_size();
3934 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3935
3936 assert(is_aligned(req_addr, alignment), "Must be");
3937 assert(is_aligned(bytes, alignment), "Must be");
3938
3939 // First reserve - but not commit - the address range in small pages.
3940 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3941
3942 if (start == NULL) {
3943 return NULL;
3944 }
3945
3946 assert(is_aligned(start, alignment), "Must be");
3947
3948 char* end = start + bytes;
3949
3950 // Find the regions of the allocated chunk that can be promoted to large pages.
3951 char* lp_start = align_up(start, large_page_size);
3952 char* lp_end = align_down(end, large_page_size);
3953
3954 size_t lp_bytes = lp_end - lp_start;
3955
3956 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3957
3958 if (lp_bytes == 0) {
3959 // The mapped region doesn't even span the start and the end of a large page.
3960 // Fall back to allocate a non-special area.
3961 ::munmap(start, end - start);
3962 return NULL;
3963 }
3964
3965 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3966
3967 void* result;
3968
3969 // Commit small-paged leading area.
3970 if (start != lp_start) {
3971 result = ::mmap(start, lp_start - start, prot,
3972 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3973 -1, 0);
3974 if (result == MAP_FAILED) {
3975 ::munmap(lp_start, end - lp_start);
3976 return NULL;
3977 }
3978 }
3979
3980 // Commit large-paged area.
3981 result = ::mmap(lp_start, lp_bytes, prot,
3982 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3983 -1, 0);
3984 if (result == MAP_FAILED) {
3985 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3986 // If the mmap above fails, the large pages region will be unmapped and we
3987 // have regions before and after with small pages. Release these regions.
3988 //
3989 // | mapped | unmapped | mapped |
3990 // ^ ^ ^ ^
3991 // start lp_start lp_end end
3992 //
3993 ::munmap(start, lp_start - start);
3994 ::munmap(lp_end, end - lp_end);
3995 return NULL;
3996 }
3997
3998 // Commit small-paged trailing area.
3999 if (lp_end != end) {
4000 result = ::mmap(lp_end, end - lp_end, prot,
4001 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4002 -1, 0);
4003 if (result == MAP_FAILED) {
4004 ::munmap(start, lp_end - start);
4005 return NULL;
4006 }
4007 }
4008
4009 return start;
4010 }
4011
4012 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4013 size_t alignment,
4014 char* req_addr,
4015 bool exec) {
4016 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4017 assert(is_aligned(req_addr, alignment), "Must be");
4018 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4019 assert(is_power_of_2(os::large_page_size()), "Must be");
4020 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4021
4022 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4023 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4024 } else {
4025 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4026 }
4027 }
4028
4029 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4030 char* req_addr, bool exec) {
4031 assert(UseLargePages, "only for large pages");
4032
4033 char* addr;
4034 if (UseSHM) {
4035 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4036 } else {
4037 assert(UseHugeTLBFS, "must be");
4038 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4039 }
4040
4041 if (addr != NULL) {
4042 if (UseNUMAInterleaving) {
4043 numa_make_global(addr, bytes);
4044 }
4045
4046 // The memory is committed
4047 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4048 }
4049
4050 return addr;
4051 }
4052
4053 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4054 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4055 return shmdt(base) == 0;
4056 }
4057
4058 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4059 return pd_release_memory(base, bytes);
4060 }
4061
4062 bool os::release_memory_special(char* base, size_t bytes) {
4063 bool res;
4064 if (MemTracker::tracking_level() > NMT_minimal) {
4065 Tracker tkr(Tracker::release);
4066 res = os::Linux::release_memory_special_impl(base, bytes);
4067 if (res) {
4068 tkr.record((address)base, bytes);
4069 }
4070
4071 } else {
4072 res = os::Linux::release_memory_special_impl(base, bytes);
4073 }
4074 return res;
4075 }
4076
4077 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4078 assert(UseLargePages, "only for large pages");
4079 bool res;
4080
4081 if (UseSHM) {
4082 res = os::Linux::release_memory_special_shm(base, bytes);
4083 } else {
4084 assert(UseHugeTLBFS, "must be");
4085 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4086 }
4087 return res;
4088 }
4089
4090 size_t os::large_page_size() {
4091 return _large_page_size;
4092 }
4093
4094 // With SysV SHM the entire memory region must be allocated as shared
4095 // memory.
4096 // HugeTLBFS allows application to commit large page memory on demand.
4097 // However, when committing memory with HugeTLBFS fails, the region
4098 // that was supposed to be committed will lose the old reservation
4099 // and allow other threads to steal that memory region. Because of this
4100 // behavior we can't commit HugeTLBFS memory.
4101 bool os::can_commit_large_page_memory() {
4102 return UseTransparentHugePages;
4103 }
4104
4105 bool os::can_execute_large_page_memory() {
4106 return UseTransparentHugePages || UseHugeTLBFS;
4107 }
4108
4109 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4110 assert(file_desc >= 0, "file_desc is not valid");
4111 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4112 if (result != NULL) {
4113 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4114 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4115 }
4116 }
4117 return result;
4118 }
4119
4120 // Reserve memory at an arbitrary address, only if that area is
4121 // available (and not reserved for something else).
4122
4123 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4124 const int max_tries = 10;
4125 char* base[max_tries];
4126 size_t size[max_tries];
4127 const size_t gap = 0x000000;
4128
4129 // Assert only that the size is a multiple of the page size, since
4130 // that's all that mmap requires, and since that's all we really know
4131 // about at this low abstraction level. If we need higher alignment,
4132 // we can either pass an alignment to this method or verify alignment
4133 // in one of the methods further up the call chain. See bug 5044738.
4134 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4135
4136 // Repeatedly allocate blocks until the block is allocated at the
4137 // right spot.
4138
4139 // Linux mmap allows caller to pass an address as hint; give it a try first,
4140 // if kernel honors the hint then we can return immediately.
4141 char * addr = anon_mmap(requested_addr, bytes, false);
4142 if (addr == requested_addr) {
4143 return requested_addr;
4144 }
4145
4146 if (addr != NULL) {
4147 // mmap() is successful but it fails to reserve at the requested address
4148 anon_munmap(addr, bytes);
4149 }
4150
4151 int i;
4152 for (i = 0; i < max_tries; ++i) {
4153 base[i] = reserve_memory(bytes);
4154
4155 if (base[i] != NULL) {
4156 // Is this the block we wanted?
4157 if (base[i] == requested_addr) {
4158 size[i] = bytes;
4159 break;
4160 }
4161
4162 // Does this overlap the block we wanted? Give back the overlapped
4163 // parts and try again.
4164
4165 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4166 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4167 unmap_memory(base[i], top_overlap);
4168 base[i] += top_overlap;
4169 size[i] = bytes - top_overlap;
4170 } else {
4171 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4172 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4173 unmap_memory(requested_addr, bottom_overlap);
4174 size[i] = bytes - bottom_overlap;
4175 } else {
4176 size[i] = bytes;
4177 }
4178 }
4179 }
4180 }
4181
4182 // Give back the unused reserved pieces.
4183
4184 for (int j = 0; j < i; ++j) {
4185 if (base[j] != NULL) {
4186 unmap_memory(base[j], size[j]);
4187 }
4188 }
4189
4190 if (i < max_tries) {
4191 return requested_addr;
4192 } else {
4193 return NULL;
4194 }
4195 }
4196
4197 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4198 return ::read(fd, buf, nBytes);
4199 }
4200
4201 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4202 return ::pread(fd, buf, nBytes, offset);
4203 }
4204
4205 // Short sleep, direct OS call.
4206 //
4207 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4208 // sched_yield(2) will actually give up the CPU:
4209 //
4210 // * Alone on this pariticular CPU, keeps running.
4211 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4212 // (pre 2.6.39).
4213 //
4214 // So calling this with 0 is an alternative.
4215 //
4216 void os::naked_short_sleep(jlong ms) {
4217 struct timespec req;
4218
4219 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4220 req.tv_sec = 0;
4221 if (ms > 0) {
4222 req.tv_nsec = (ms % 1000) * 1000000;
4223 } else {
4224 req.tv_nsec = 1;
4225 }
4226
4227 nanosleep(&req, NULL);
4228
4229 return;
4230 }
4231
4232 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4233 void os::infinite_sleep() {
4234 while (true) { // sleep forever ...
4235 ::sleep(100); // ... 100 seconds at a time
4236 }
4237 }
4238
4239 // Used to convert frequent JVM_Yield() to nops
4240 bool os::dont_yield() {
4241 return DontYieldALot;
4242 }
4243
4244 void os::naked_yield() {
4245 sched_yield();
4246 }
4247
4248 ////////////////////////////////////////////////////////////////////////////////
4249 // thread priority support
4250
4251 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4252 // only supports dynamic priority, static priority must be zero. For real-time
4253 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4254 // However, for large multi-threaded applications, SCHED_RR is not only slower
4255 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4256 // of 5 runs - Sep 2005).
4257 //
4258 // The following code actually changes the niceness of kernel-thread/LWP. It
4259 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4260 // not the entire user process, and user level threads are 1:1 mapped to kernel
4261 // threads. It has always been the case, but could change in the future. For
4262 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4263 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4264 // (e.g., root privilege or CAP_SYS_NICE capability).
4265
4266 int os::java_to_os_priority[CriticalPriority + 1] = {
4267 19, // 0 Entry should never be used
4268
4269 4, // 1 MinPriority
4270 3, // 2
4271 2, // 3
4272
4273 1, // 4
4274 0, // 5 NormPriority
4275 -1, // 6
4276
4277 -2, // 7
4278 -3, // 8
4279 -4, // 9 NearMaxPriority
4280
4281 -5, // 10 MaxPriority
4282
4283 -5 // 11 CriticalPriority
4284 };
4285
4286 static int prio_init() {
4287 if (ThreadPriorityPolicy == 1) {
4288 if (geteuid() != 0) {
4289 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4290 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4291 "e.g., being the root user. If the necessary permission is not " \
4292 "possessed, changes to priority will be silently ignored.");
4293 }
4294 }
4295 }
4296 if (UseCriticalJavaThreadPriority) {
4297 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4298 }
4299 return 0;
4300 }
4301
4302 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4303 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4304
4305 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4306 return (ret == 0) ? OS_OK : OS_ERR;
4307 }
4308
4309 OSReturn os::get_native_priority(const Thread* const thread,
4310 int *priority_ptr) {
4311 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4312 *priority_ptr = java_to_os_priority[NormPriority];
4313 return OS_OK;
4314 }
4315
4316 errno = 0;
4317 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4318 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4319 }
4320
4321 // Hint to the underlying OS that a task switch would not be good.
4322 // Void return because it's a hint and can fail.
4323 void os::hint_no_preempt() {}
4324
4325 ////////////////////////////////////////////////////////////////////////////////
4326 // suspend/resume support
4327
4328 // The low-level signal-based suspend/resume support is a remnant from the
4329 // old VM-suspension that used to be for java-suspension, safepoints etc,
4330 // within hotspot. Currently used by JFR's OSThreadSampler
4331 //
4332 // The remaining code is greatly simplified from the more general suspension
4333 // code that used to be used.
4334 //
4335 // The protocol is quite simple:
4336 // - suspend:
4337 // - sends a signal to the target thread
4338 // - polls the suspend state of the osthread using a yield loop
4339 // - target thread signal handler (SR_handler) sets suspend state
4340 // and blocks in sigsuspend until continued
4341 // - resume:
4342 // - sets target osthread state to continue
4343 // - sends signal to end the sigsuspend loop in the SR_handler
4344 //
4345 // Note that the SR_lock plays no role in this suspend/resume protocol,
4346 // but is checked for NULL in SR_handler as a thread termination indicator.
4347 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4348 //
4349 // Note that resume_clear_context() and suspend_save_context() are needed
4350 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4351 // which in part is used by:
4352 // - Forte Analyzer: AsyncGetCallTrace()
4353 // - StackBanging: get_frame_at_stack_banging_point()
4354
4355 static void resume_clear_context(OSThread *osthread) {
4356 osthread->set_ucontext(NULL);
4357 osthread->set_siginfo(NULL);
4358 }
4359
4360 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4361 ucontext_t* context) {
4362 osthread->set_ucontext(context);
4363 osthread->set_siginfo(siginfo);
4364 }
4365
4366 // Handler function invoked when a thread's execution is suspended or
4367 // resumed. We have to be careful that only async-safe functions are
4368 // called here (Note: most pthread functions are not async safe and
4369 // should be avoided.)
4370 //
4371 // Note: sigwait() is a more natural fit than sigsuspend() from an
4372 // interface point of view, but sigwait() prevents the signal hander
4373 // from being run. libpthread would get very confused by not having
4374 // its signal handlers run and prevents sigwait()'s use with the
4375 // mutex granting granting signal.
4376 //
4377 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4378 //
4379 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4380 // Save and restore errno to avoid confusing native code with EINTR
4381 // after sigsuspend.
4382 int old_errno = errno;
4383
4384 Thread* thread = Thread::current_or_null_safe();
4385 assert(thread != NULL, "Missing current thread in SR_handler");
4386
4387 // On some systems we have seen signal delivery get "stuck" until the signal
4388 // mask is changed as part of thread termination. Check that the current thread
4389 // has not already terminated (via SR_lock()) - else the following assertion
4390 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4391 // destructor has completed.
4392
4393 if (thread->SR_lock() == NULL) {
4394 return;
4395 }
4396
4397 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4398
4399 OSThread* osthread = thread->osthread();
4400
4401 os::SuspendResume::State current = osthread->sr.state();
4402 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4403 suspend_save_context(osthread, siginfo, context);
4404
4405 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4406 os::SuspendResume::State state = osthread->sr.suspended();
4407 if (state == os::SuspendResume::SR_SUSPENDED) {
4408 sigset_t suspend_set; // signals for sigsuspend()
4409 sigemptyset(&suspend_set);
4410 // get current set of blocked signals and unblock resume signal
4411 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4412 sigdelset(&suspend_set, SR_signum);
4413
4414 sr_semaphore.signal();
4415 // wait here until we are resumed
4416 while (1) {
4417 sigsuspend(&suspend_set);
4418
4419 os::SuspendResume::State result = osthread->sr.running();
4420 if (result == os::SuspendResume::SR_RUNNING) {
4421 sr_semaphore.signal();
4422 break;
4423 }
4424 }
4425
4426 } else if (state == os::SuspendResume::SR_RUNNING) {
4427 // request was cancelled, continue
4428 } else {
4429 ShouldNotReachHere();
4430 }
4431
4432 resume_clear_context(osthread);
4433 } else if (current == os::SuspendResume::SR_RUNNING) {
4434 // request was cancelled, continue
4435 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4436 // ignore
4437 } else {
4438 // ignore
4439 }
4440
4441 errno = old_errno;
4442 }
4443
4444 static int SR_initialize() {
4445 struct sigaction act;
4446 char *s;
4447
4448 // Get signal number to use for suspend/resume
4449 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4450 int sig = ::strtol(s, 0, 10);
4451 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4452 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4453 SR_signum = sig;
4454 } else {
4455 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4456 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4457 }
4458 }
4459
4460 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4461 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4462
4463 sigemptyset(&SR_sigset);
4464 sigaddset(&SR_sigset, SR_signum);
4465
4466 // Set up signal handler for suspend/resume
4467 act.sa_flags = SA_RESTART|SA_SIGINFO;
4468 act.sa_handler = (void (*)(int)) SR_handler;
4469
4470 // SR_signum is blocked by default.
4471 // 4528190 - We also need to block pthread restart signal (32 on all
4472 // supported Linux platforms). Note that LinuxThreads need to block
4473 // this signal for all threads to work properly. So we don't have
4474 // to use hard-coded signal number when setting up the mask.
4475 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4476
4477 if (sigaction(SR_signum, &act, 0) == -1) {
4478 return -1;
4479 }
4480
4481 // Save signal flag
4482 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4483 return 0;
4484 }
4485
4486 static int sr_notify(OSThread* osthread) {
4487 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4488 assert_status(status == 0, status, "pthread_kill");
4489 return status;
4490 }
4491
4492 // "Randomly" selected value for how long we want to spin
4493 // before bailing out on suspending a thread, also how often
4494 // we send a signal to a thread we want to resume
4495 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4496 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4497
4498 // returns true on success and false on error - really an error is fatal
4499 // but this seems the normal response to library errors
4500 static bool do_suspend(OSThread* osthread) {
4501 assert(osthread->sr.is_running(), "thread should be running");
4502 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4503
4504 // mark as suspended and send signal
4505 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4506 // failed to switch, state wasn't running?
4507 ShouldNotReachHere();
4508 return false;
4509 }
4510
4511 if (sr_notify(osthread) != 0) {
4512 ShouldNotReachHere();
4513 }
4514
4515 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4516 while (true) {
4517 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4518 break;
4519 } else {
4520 // timeout
4521 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4522 if (cancelled == os::SuspendResume::SR_RUNNING) {
4523 return false;
4524 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4525 // make sure that we consume the signal on the semaphore as well
4526 sr_semaphore.wait();
4527 break;
4528 } else {
4529 ShouldNotReachHere();
4530 return false;
4531 }
4532 }
4533 }
4534
4535 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4536 return true;
4537 }
4538
4539 static void do_resume(OSThread* osthread) {
4540 assert(osthread->sr.is_suspended(), "thread should be suspended");
4541 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4542
4543 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4544 // failed to switch to WAKEUP_REQUEST
4545 ShouldNotReachHere();
4546 return;
4547 }
4548
4549 while (true) {
4550 if (sr_notify(osthread) == 0) {
4551 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4552 if (osthread->sr.is_running()) {
4553 return;
4554 }
4555 }
4556 } else {
4557 ShouldNotReachHere();
4558 }
4559 }
4560
4561 guarantee(osthread->sr.is_running(), "Must be running!");
4562 }
4563
4564 ///////////////////////////////////////////////////////////////////////////////////
4565 // signal handling (except suspend/resume)
4566
4567 // This routine may be used by user applications as a "hook" to catch signals.
4568 // The user-defined signal handler must pass unrecognized signals to this
4569 // routine, and if it returns true (non-zero), then the signal handler must
4570 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4571 // routine will never retun false (zero), but instead will execute a VM panic
4572 // routine kill the process.
4573 //
4574 // If this routine returns false, it is OK to call it again. This allows
4575 // the user-defined signal handler to perform checks either before or after
4576 // the VM performs its own checks. Naturally, the user code would be making
4577 // a serious error if it tried to handle an exception (such as a null check
4578 // or breakpoint) that the VM was generating for its own correct operation.
4579 //
4580 // This routine may recognize any of the following kinds of signals:
4581 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4582 // It should be consulted by handlers for any of those signals.
4583 //
4584 // The caller of this routine must pass in the three arguments supplied
4585 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4586 // field of the structure passed to sigaction(). This routine assumes that
4587 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4588 //
4589 // Note that the VM will print warnings if it detects conflicting signal
4590 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4591 //
4592 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4593 siginfo_t* siginfo,
4594 void* ucontext,
4595 int abort_if_unrecognized);
4596
4597 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4598 assert(info != NULL && uc != NULL, "it must be old kernel");
4599 int orig_errno = errno; // Preserve errno value over signal handler.
4600 JVM_handle_linux_signal(sig, info, uc, true);
4601 errno = orig_errno;
4602 }
4603
4604
4605 // This boolean allows users to forward their own non-matching signals
4606 // to JVM_handle_linux_signal, harmlessly.
4607 bool os::Linux::signal_handlers_are_installed = false;
4608
4609 // For signal-chaining
4610 bool os::Linux::libjsig_is_loaded = false;
4611 typedef struct sigaction *(*get_signal_t)(int);
4612 get_signal_t os::Linux::get_signal_action = NULL;
4613
4614 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4615 struct sigaction *actp = NULL;
4616
4617 if (libjsig_is_loaded) {
4618 // Retrieve the old signal handler from libjsig
4619 actp = (*get_signal_action)(sig);
4620 }
4621 if (actp == NULL) {
4622 // Retrieve the preinstalled signal handler from jvm
4623 actp = os::Posix::get_preinstalled_handler(sig);
4624 }
4625
4626 return actp;
4627 }
4628
4629 static bool call_chained_handler(struct sigaction *actp, int sig,
4630 siginfo_t *siginfo, void *context) {
4631 // Call the old signal handler
4632 if (actp->sa_handler == SIG_DFL) {
4633 // It's more reasonable to let jvm treat it as an unexpected exception
4634 // instead of taking the default action.
4635 return false;
4636 } else if (actp->sa_handler != SIG_IGN) {
4637 if ((actp->sa_flags & SA_NODEFER) == 0) {
4638 // automaticlly block the signal
4639 sigaddset(&(actp->sa_mask), sig);
4640 }
4641
4642 sa_handler_t hand = NULL;
4643 sa_sigaction_t sa = NULL;
4644 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4645 // retrieve the chained handler
4646 if (siginfo_flag_set) {
4647 sa = actp->sa_sigaction;
4648 } else {
4649 hand = actp->sa_handler;
4650 }
4651
4652 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4653 actp->sa_handler = SIG_DFL;
4654 }
4655
4656 // try to honor the signal mask
4657 sigset_t oset;
4658 sigemptyset(&oset);
4659 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4660
4661 // call into the chained handler
4662 if (siginfo_flag_set) {
4663 (*sa)(sig, siginfo, context);
4664 } else {
4665 (*hand)(sig);
4666 }
4667
4668 // restore the signal mask
4669 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4670 }
4671 // Tell jvm's signal handler the signal is taken care of.
4672 return true;
4673 }
4674
4675 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4676 bool chained = false;
4677 // signal-chaining
4678 if (UseSignalChaining) {
4679 struct sigaction *actp = get_chained_signal_action(sig);
4680 if (actp != NULL) {
4681 chained = call_chained_handler(actp, sig, siginfo, context);
4682 }
4683 }
4684 return chained;
4685 }
4686
4687 // for diagnostic
4688 int sigflags[NSIG];
4689
4690 int os::Linux::get_our_sigflags(int sig) {
4691 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4692 return sigflags[sig];
4693 }
4694
4695 void os::Linux::set_our_sigflags(int sig, int flags) {
4696 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4697 if (sig > 0 && sig < NSIG) {
4698 sigflags[sig] = flags;
4699 }
4700 }
4701
4702 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4703 // Check for overwrite.
4704 struct sigaction oldAct;
4705 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4706
4707 void* oldhand = oldAct.sa_sigaction
4708 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4709 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4710 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4711 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4712 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4713 if (AllowUserSignalHandlers || !set_installed) {
4714 // Do not overwrite; user takes responsibility to forward to us.
4715 return;
4716 } else if (UseSignalChaining) {
4717 // save the old handler in jvm
4718 os::Posix::save_preinstalled_handler(sig, oldAct);
4719 // libjsig also interposes the sigaction() call below and saves the
4720 // old sigaction on it own.
4721 } else {
4722 fatal("Encountered unexpected pre-existing sigaction handler "
4723 "%#lx for signal %d.", (long)oldhand, sig);
4724 }
4725 }
4726
4727 struct sigaction sigAct;
4728 sigfillset(&(sigAct.sa_mask));
4729 sigAct.sa_handler = SIG_DFL;
4730 if (!set_installed) {
4731 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4732 } else {
4733 sigAct.sa_sigaction = signalHandler;
4734 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4735 }
4736 // Save flags, which are set by ours
4737 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4738 sigflags[sig] = sigAct.sa_flags;
4739
4740 int ret = sigaction(sig, &sigAct, &oldAct);
4741 assert(ret == 0, "check");
4742
4743 void* oldhand2 = oldAct.sa_sigaction
4744 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4745 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4746 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4747 }
4748
4749 // install signal handlers for signals that HotSpot needs to
4750 // handle in order to support Java-level exception handling.
4751
4752 void os::Linux::install_signal_handlers() {
4753 if (!signal_handlers_are_installed) {
4754 signal_handlers_are_installed = true;
4755
4756 // signal-chaining
4757 typedef void (*signal_setting_t)();
4758 signal_setting_t begin_signal_setting = NULL;
4759 signal_setting_t end_signal_setting = NULL;
4760 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4761 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4762 if (begin_signal_setting != NULL) {
4763 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4764 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4765 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4766 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4767 libjsig_is_loaded = true;
4768 assert(UseSignalChaining, "should enable signal-chaining");
4769 }
4770 if (libjsig_is_loaded) {
4771 // Tell libjsig jvm is setting signal handlers
4772 (*begin_signal_setting)();
4773 }
4774
4775 set_signal_handler(SIGSEGV, true);
4776 set_signal_handler(SIGPIPE, true);
4777 set_signal_handler(SIGBUS, true);
4778 set_signal_handler(SIGILL, true);
4779 set_signal_handler(SIGFPE, true);
4780 #if defined(PPC64)
4781 set_signal_handler(SIGTRAP, true);
4782 #endif
4783 set_signal_handler(SIGXFSZ, true);
4784
4785 if (libjsig_is_loaded) {
4786 // Tell libjsig jvm finishes setting signal handlers
4787 (*end_signal_setting)();
4788 }
4789
4790 // We don't activate signal checker if libjsig is in place, we trust ourselves
4791 // and if UserSignalHandler is installed all bets are off.
4792 // Log that signal checking is off only if -verbose:jni is specified.
4793 if (CheckJNICalls) {
4794 if (libjsig_is_loaded) {
4795 if (PrintJNIResolving) {
4796 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4797 }
4798 check_signals = false;
4799 }
4800 if (AllowUserSignalHandlers) {
4801 if (PrintJNIResolving) {
4802 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4803 }
4804 check_signals = false;
4805 }
4806 }
4807 }
4808 }
4809
4810 // This is the fastest way to get thread cpu time on Linux.
4811 // Returns cpu time (user+sys) for any thread, not only for current.
4812 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4813 // It might work on 2.6.10+ with a special kernel/glibc patch.
4814 // For reference, please, see IEEE Std 1003.1-2004:
4815 // http://www.unix.org/single_unix_specification
4816
4817 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4818 struct timespec tp;
4819 int rc = os::Linux::clock_gettime(clockid, &tp);
4820 assert(rc == 0, "clock_gettime is expected to return 0 code");
4821
4822 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4823 }
4824
4825 void os::Linux::initialize_os_info() {
4826 assert(_os_version == 0, "OS info already initialized");
4827
4828 struct utsname _uname;
4829
4830 uint32_t major;
4831 uint32_t minor;
4832 uint32_t fix;
4833
4834 int rc;
4835
4836 // Kernel version is unknown if
4837 // verification below fails.
4838 _os_version = 0x01000000;
4839
4840 rc = uname(&_uname);
4841 if (rc != -1) {
4842
4843 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4844 if (rc == 3) {
4845
4846 if (major < 256 && minor < 256 && fix < 256) {
4847 // Kernel version format is as expected,
4848 // set it overriding unknown state.
4849 _os_version = (major << 16) |
4850 (minor << 8 ) |
4851 (fix << 0 ) ;
4852 }
4853 }
4854 }
4855 }
4856
4857 uint32_t os::Linux::os_version() {
4858 assert(_os_version != 0, "not initialized");
4859 return _os_version & 0x00FFFFFF;
4860 }
4861
4862 bool os::Linux::os_version_is_known() {
4863 assert(_os_version != 0, "not initialized");
4864 return _os_version & 0x01000000 ? false : true;
4865 }
4866
4867 /////
4868 // glibc on Linux platform uses non-documented flag
4869 // to indicate, that some special sort of signal
4870 // trampoline is used.
4871 // We will never set this flag, and we should
4872 // ignore this flag in our diagnostic
4873 #ifdef SIGNIFICANT_SIGNAL_MASK
4874 #undef SIGNIFICANT_SIGNAL_MASK
4875 #endif
4876 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4877
4878 static const char* get_signal_handler_name(address handler,
4879 char* buf, int buflen) {
4880 int offset = 0;
4881 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4882 if (found) {
4883 // skip directory names
4884 const char *p1, *p2;
4885 p1 = buf;
4886 size_t len = strlen(os::file_separator());
4887 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4888 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4889 } else {
4890 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4891 }
4892 return buf;
4893 }
4894
4895 static void print_signal_handler(outputStream* st, int sig,
4896 char* buf, size_t buflen) {
4897 struct sigaction sa;
4898
4899 sigaction(sig, NULL, &sa);
4900
4901 // See comment for SIGNIFICANT_SIGNAL_MASK define
4902 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4903
4904 st->print("%s: ", os::exception_name(sig, buf, buflen));
4905
4906 address handler = (sa.sa_flags & SA_SIGINFO)
4907 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4908 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4909
4910 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4911 st->print("SIG_DFL");
4912 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4913 st->print("SIG_IGN");
4914 } else {
4915 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4916 }
4917
4918 st->print(", sa_mask[0]=");
4919 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4920
4921 address rh = VMError::get_resetted_sighandler(sig);
4922 // May be, handler was resetted by VMError?
4923 if (rh != NULL) {
4924 handler = rh;
4925 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4926 }
4927
4928 st->print(", sa_flags=");
4929 os::Posix::print_sa_flags(st, sa.sa_flags);
4930
4931 // Check: is it our handler?
4932 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4933 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4934 // It is our signal handler
4935 // check for flags, reset system-used one!
4936 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4937 st->print(
4938 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4939 os::Linux::get_our_sigflags(sig));
4940 }
4941 }
4942 st->cr();
4943 }
4944
4945
4946 #define DO_SIGNAL_CHECK(sig) \
4947 do { \
4948 if (!sigismember(&check_signal_done, sig)) { \
4949 os::Linux::check_signal_handler(sig); \
4950 } \
4951 } while (0)
4952
4953 // This method is a periodic task to check for misbehaving JNI applications
4954 // under CheckJNI, we can add any periodic checks here
4955
4956 void os::run_periodic_checks() {
4957 if (check_signals == false) return;
4958
4959 // SEGV and BUS if overridden could potentially prevent
4960 // generation of hs*.log in the event of a crash, debugging
4961 // such a case can be very challenging, so we absolutely
4962 // check the following for a good measure:
4963 DO_SIGNAL_CHECK(SIGSEGV);
4964 DO_SIGNAL_CHECK(SIGILL);
4965 DO_SIGNAL_CHECK(SIGFPE);
4966 DO_SIGNAL_CHECK(SIGBUS);
4967 DO_SIGNAL_CHECK(SIGPIPE);
4968 DO_SIGNAL_CHECK(SIGXFSZ);
4969 #if defined(PPC64)
4970 DO_SIGNAL_CHECK(SIGTRAP);
4971 #endif
4972
4973 // ReduceSignalUsage allows the user to override these handlers
4974 // see comments at the very top and jvm_md.h
4975 if (!ReduceSignalUsage) {
4976 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4977 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4978 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4979 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4980 }
4981
4982 DO_SIGNAL_CHECK(SR_signum);
4983 }
4984
4985 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4986
4987 static os_sigaction_t os_sigaction = NULL;
4988
4989 void os::Linux::check_signal_handler(int sig) {
4990 char buf[O_BUFLEN];
4991 address jvmHandler = NULL;
4992
4993
4994 struct sigaction act;
4995 if (os_sigaction == NULL) {
4996 // only trust the default sigaction, in case it has been interposed
4997 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4998 if (os_sigaction == NULL) return;
4999 }
5000
5001 os_sigaction(sig, (struct sigaction*)NULL, &act);
5002
5003
5004 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
5005
5006 address thisHandler = (act.sa_flags & SA_SIGINFO)
5007 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5008 : CAST_FROM_FN_PTR(address, act.sa_handler);
5009
5010
5011 switch (sig) {
5012 case SIGSEGV:
5013 case SIGBUS:
5014 case SIGFPE:
5015 case SIGPIPE:
5016 case SIGILL:
5017 case SIGXFSZ:
5018 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5019 break;
5020
5021 case SHUTDOWN1_SIGNAL:
5022 case SHUTDOWN2_SIGNAL:
5023 case SHUTDOWN3_SIGNAL:
5024 case BREAK_SIGNAL:
5025 jvmHandler = (address)user_handler();
5026 break;
5027
5028 default:
5029 if (sig == SR_signum) {
5030 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5031 } else {
5032 return;
5033 }
5034 break;
5035 }
5036
5037 if (thisHandler != jvmHandler) {
5038 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5039 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5040 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5041 // No need to check this sig any longer
5042 sigaddset(&check_signal_done, sig);
5043 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5044 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5045 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5046 exception_name(sig, buf, O_BUFLEN));
5047 }
5048 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5049 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5050 tty->print("expected:");
5051 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5052 tty->cr();
5053 tty->print(" found:");
5054 os::Posix::print_sa_flags(tty, act.sa_flags);
5055 tty->cr();
5056 // No need to check this sig any longer
5057 sigaddset(&check_signal_done, sig);
5058 }
5059
5060 // Dump all the signal
5061 if (sigismember(&check_signal_done, sig)) {
5062 print_signal_handlers(tty, buf, O_BUFLEN);
5063 }
5064 }
5065
5066 extern void report_error(char* file_name, int line_no, char* title,
5067 char* format, ...);
5068
5069 // this is called _before_ most of the global arguments have been parsed
5070 void os::init(void) {
5071 char dummy; // used to get a guess on initial stack address
5072
5073 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5074
5075 init_random(1234567);
5076
5077 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5078 if (Linux::page_size() == -1) {
5079 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5080 os::strerror(errno));
5081 }
5082 init_page_sizes((size_t) Linux::page_size());
5083
5084 Linux::initialize_system_info();
5085
5086 Linux::initialize_os_info();
5087
5088 os::Linux::CPUPerfTicks pticks;
5089 bool res = os::Linux::get_tick_information(&pticks, -1);
5090
5091 if (res && pticks.has_steal_ticks) {
5092 has_initial_tick_info = true;
5093 initial_total_ticks = pticks.total;
5094 initial_steal_ticks = pticks.steal;
5095 }
5096
5097 // _main_thread points to the thread that created/loaded the JVM.
5098 Linux::_main_thread = pthread_self();
5099
5100 Linux::clock_init();
5101 initial_time_count = javaTimeNanos();
5102
5103 // retrieve entry point for pthread_setname_np
5104 Linux::_pthread_setname_np =
5105 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5106
5107 os::Posix::init();
5108 }
5109
5110 // To install functions for atexit system call
5111 extern "C" {
5112 static void perfMemory_exit_helper() {
5113 perfMemory_exit();
5114 }
5115 }
5116
5117 void os::pd_init_container_support() {
5118 OSContainer::init();
5119 }
5120
5121 // this is called _after_ the global arguments have been parsed
5122 jint os::init_2(void) {
5123
5124 os::Posix::init_2();
5125
5126 Linux::fast_thread_clock_init();
5127
5128 // initialize suspend/resume support - must do this before signal_sets_init()
5129 if (SR_initialize() != 0) {
5130 perror("SR_initialize failed");
5131 return JNI_ERR;
5132 }
5133
5134 Linux::signal_sets_init();
5135 Linux::install_signal_handlers();
5136 // Initialize data for jdk.internal.misc.Signal
5137 if (!ReduceSignalUsage) {
5138 jdk_misc_signal_init();
5139 }
5140
5141 // Check and sets minimum stack sizes against command line options
5142 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5143 return JNI_ERR;
5144 }
5145
5146 #if defined(IA32)
5147 // Need to ensure we've determined the process's initial stack to
5148 // perform the workaround
5149 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5150 workaround_expand_exec_shield_cs_limit();
5151 #else
5152 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5153 if (!suppress_primordial_thread_resolution) {
5154 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5155 }
5156 #endif
5157
5158 Linux::libpthread_init();
5159 Linux::sched_getcpu_init();
5160 log_info(os)("HotSpot is running with %s, %s",
5161 Linux::glibc_version(), Linux::libpthread_version());
5162
5163 if (UseNUMA) {
5164 if (!Linux::libnuma_init()) {
5165 UseNUMA = false;
5166 } else {
5167 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5168 // If there's only one node (they start from 0) or if the process
5169 // is bound explicitly to a single node using membind, disable NUMA.
5170 UseNUMA = false;
5171 }
5172 }
5173
5174 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5175 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5176 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5177 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5178 // and disable adaptive resizing.
5179 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5180 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5181 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5182 UseAdaptiveSizePolicy = false;
5183 UseAdaptiveNUMAChunkSizing = false;
5184 }
5185 }
5186
5187 if (!UseNUMA && ForceNUMA) {
5188 UseNUMA = true;
5189 }
5190 }
5191
5192 if (MaxFDLimit) {
5193 // set the number of file descriptors to max. print out error
5194 // if getrlimit/setrlimit fails but continue regardless.
5195 struct rlimit nbr_files;
5196 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5197 if (status != 0) {
5198 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5199 } else {
5200 nbr_files.rlim_cur = nbr_files.rlim_max;
5201 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5202 if (status != 0) {
5203 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5204 }
5205 }
5206 }
5207
5208 // Initialize lock used to serialize thread creation (see os::create_thread)
5209 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5210
5211 // at-exit methods are called in the reverse order of their registration.
5212 // atexit functions are called on return from main or as a result of a
5213 // call to exit(3C). There can be only 32 of these functions registered
5214 // and atexit() does not set errno.
5215
5216 if (PerfAllowAtExitRegistration) {
5217 // only register atexit functions if PerfAllowAtExitRegistration is set.
5218 // atexit functions can be delayed until process exit time, which
5219 // can be problematic for embedded VM situations. Embedded VMs should
5220 // call DestroyJavaVM() to assure that VM resources are released.
5221
5222 // note: perfMemory_exit_helper atexit function may be removed in
5223 // the future if the appropriate cleanup code can be added to the
5224 // VM_Exit VMOperation's doit method.
5225 if (atexit(perfMemory_exit_helper) != 0) {
5226 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5227 }
5228 }
5229
5230 // initialize thread priority policy
5231 prio_init();
5232
5233 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5234 set_coredump_filter(DAX_SHARED_BIT);
5235 }
5236
5237 if (DumpPrivateMappingsInCore) {
5238 set_coredump_filter(FILE_BACKED_PVT_BIT);
5239 }
5240
5241 if (DumpSharedMappingsInCore) {
5242 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5243 }
5244
5245 return JNI_OK;
5246 }
5247
5248 // Mark the polling page as unreadable
5249 void os::make_polling_page_unreadable(void) {
5250 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5251 fatal("Could not disable polling page");
5252 }
5253 }
5254
5255 // Mark the polling page as readable
5256 void os::make_polling_page_readable(void) {
5257 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5258 fatal("Could not enable polling page");
5259 }
5260 }
5261
5262 // older glibc versions don't have this macro (which expands to
5263 // an optimized bit-counting function) so we have to roll our own
5264 #ifndef CPU_COUNT
5265
5266 static int _cpu_count(const cpu_set_t* cpus) {
5267 int count = 0;
5268 // only look up to the number of configured processors
5269 for (int i = 0; i < os::processor_count(); i++) {
5270 if (CPU_ISSET(i, cpus)) {
5271 count++;
5272 }
5273 }
5274 return count;
5275 }
5276
5277 #define CPU_COUNT(cpus) _cpu_count(cpus)
5278
5279 #endif // CPU_COUNT
5280
5281 // Get the current number of available processors for this process.
5282 // This value can change at any time during a process's lifetime.
5283 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5284 // If it appears there may be more than 1024 processors then we do a
5285 // dynamic check - see 6515172 for details.
5286 // If anything goes wrong we fallback to returning the number of online
5287 // processors - which can be greater than the number available to the process.
5288 int os::Linux::active_processor_count() {
5289 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5290 cpu_set_t* cpus_p = &cpus;
5291 int cpus_size = sizeof(cpu_set_t);
5292
5293 int configured_cpus = os::processor_count(); // upper bound on available cpus
5294 int cpu_count = 0;
5295
5296 // old build platforms may not support dynamic cpu sets
5297 #ifdef CPU_ALLOC
5298
5299 // To enable easy testing of the dynamic path on different platforms we
5300 // introduce a diagnostic flag: UseCpuAllocPath
5301 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5302 // kernel may use a mask bigger than cpu_set_t
5303 log_trace(os)("active_processor_count: using dynamic path %s"
5304 "- configured processors: %d",
5305 UseCpuAllocPath ? "(forced) " : "",
5306 configured_cpus);
5307 cpus_p = CPU_ALLOC(configured_cpus);
5308 if (cpus_p != NULL) {
5309 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5310 // zero it just to be safe
5311 CPU_ZERO_S(cpus_size, cpus_p);
5312 }
5313 else {
5314 // failed to allocate so fallback to online cpus
5315 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5316 log_trace(os)("active_processor_count: "
5317 "CPU_ALLOC failed (%s) - using "
5318 "online processor count: %d",
5319 os::strerror(errno), online_cpus);
5320 return online_cpus;
5321 }
5322 }
5323 else {
5324 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5325 configured_cpus);
5326 }
5327 #else // CPU_ALLOC
5328 // these stubs won't be executed
5329 #define CPU_COUNT_S(size, cpus) -1
5330 #define CPU_FREE(cpus)
5331
5332 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5333 configured_cpus);
5334 #endif // CPU_ALLOC
5335
5336 // pid 0 means the current thread - which we have to assume represents the process
5337 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5338 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5339 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5340 }
5341 else {
5342 cpu_count = CPU_COUNT(cpus_p);
5343 }
5344 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5345 }
5346 else {
5347 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5348 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5349 "which may exceed available processors", os::strerror(errno), cpu_count);
5350 }
5351
5352 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5353 CPU_FREE(cpus_p);
5354 }
5355
5356 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5357 return cpu_count;
5358 }
5359
5360 // Determine the active processor count from one of
5361 // three different sources:
5362 //
5363 // 1. User option -XX:ActiveProcessorCount
5364 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5365 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5366 //
5367 // Option 1, if specified, will always override.
5368 // If the cgroup subsystem is active and configured, we
5369 // will return the min of the cgroup and option 2 results.
5370 // This is required since tools, such as numactl, that
5371 // alter cpu affinity do not update cgroup subsystem
5372 // cpuset configuration files.
5373 int os::active_processor_count() {
5374 // User has overridden the number of active processors
5375 if (ActiveProcessorCount > 0) {
5376 log_trace(os)("active_processor_count: "
5377 "active processor count set by user : %d",
5378 ActiveProcessorCount);
5379 return ActiveProcessorCount;
5380 }
5381
5382 int active_cpus;
5383 if (OSContainer::is_containerized()) {
5384 active_cpus = OSContainer::active_processor_count();
5385 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5386 active_cpus);
5387 } else {
5388 active_cpus = os::Linux::active_processor_count();
5389 }
5390
5391 return active_cpus;
5392 }
5393
5394 uint os::processor_id() {
5395 const int id = Linux::sched_getcpu();
5396 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5397 return (uint)id;
5398 }
5399
5400 void os::set_native_thread_name(const char *name) {
5401 if (Linux::_pthread_setname_np) {
5402 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5403 snprintf(buf, sizeof(buf), "%s", name);
5404 buf[sizeof(buf) - 1] = '\0';
5405 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5406 // ERANGE should not happen; all other errors should just be ignored.
5407 assert(rc != ERANGE, "pthread_setname_np failed");
5408 }
5409 }
5410
5411 bool os::distribute_processes(uint length, uint* distribution) {
5412 // Not yet implemented.
5413 return false;
5414 }
5415
5416 bool os::bind_to_processor(uint processor_id) {
5417 // Not yet implemented.
5418 return false;
5419 }
5420
5421 ///
5422
5423 void os::SuspendedThreadTask::internal_do_task() {
5424 if (do_suspend(_thread->osthread())) {
5425 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5426 do_task(context);
5427 do_resume(_thread->osthread());
5428 }
5429 }
5430
5431 ////////////////////////////////////////////////////////////////////////////////
5432 // debug support
5433
5434 bool os::find(address addr, outputStream* st) {
5435 Dl_info dlinfo;
5436 memset(&dlinfo, 0, sizeof(dlinfo));
5437 if (dladdr(addr, &dlinfo) != 0) {
5438 st->print(PTR_FORMAT ": ", p2i(addr));
5439 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5440 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5441 p2i(addr) - p2i(dlinfo.dli_saddr));
5442 } else if (dlinfo.dli_fbase != NULL) {
5443 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5444 } else {
5445 st->print("<absolute address>");
5446 }
5447 if (dlinfo.dli_fname != NULL) {
5448 st->print(" in %s", dlinfo.dli_fname);
5449 }
5450 if (dlinfo.dli_fbase != NULL) {
5451 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5452 }
5453 st->cr();
5454
5455 if (Verbose) {
5456 // decode some bytes around the PC
5457 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5458 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5459 address lowest = (address) dlinfo.dli_sname;
5460 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5461 if (begin < lowest) begin = lowest;
5462 Dl_info dlinfo2;
5463 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5464 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5465 end = (address) dlinfo2.dli_saddr;
5466 }
5467 Disassembler::decode(begin, end, st);
5468 }
5469 return true;
5470 }
5471 return false;
5472 }
5473
5474 ////////////////////////////////////////////////////////////////////////////////
5475 // misc
5476
5477 // This does not do anything on Linux. This is basically a hook for being
5478 // able to use structured exception handling (thread-local exception filters)
5479 // on, e.g., Win32.
5480 void
5481 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5482 JavaCallArguments* args, Thread* thread) {
5483 f(value, method, args, thread);
5484 }
5485
5486 void os::print_statistics() {
5487 }
5488
5489 bool os::message_box(const char* title, const char* message) {
5490 int i;
5491 fdStream err(defaultStream::error_fd());
5492 for (i = 0; i < 78; i++) err.print_raw("=");
5493 err.cr();
5494 err.print_raw_cr(title);
5495 for (i = 0; i < 78; i++) err.print_raw("-");
5496 err.cr();
5497 err.print_raw_cr(message);
5498 for (i = 0; i < 78; i++) err.print_raw("=");
5499 err.cr();
5500
5501 char buf[16];
5502 // Prevent process from exiting upon "read error" without consuming all CPU
5503 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5504
5505 return buf[0] == 'y' || buf[0] == 'Y';
5506 }
5507
5508 // Is a (classpath) directory empty?
5509 bool os::dir_is_empty(const char* path) {
5510 DIR *dir = NULL;
5511 struct dirent *ptr;
5512
5513 dir = opendir(path);
5514 if (dir == NULL) return true;
5515
5516 // Scan the directory
5517 bool result = true;
5518 while (result && (ptr = readdir(dir)) != NULL) {
5519 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5520 result = false;
5521 }
5522 }
5523 closedir(dir);
5524 return result;
5525 }
5526
5527 // This code originates from JDK's sysOpen and open64_w
5528 // from src/solaris/hpi/src/system_md.c
5529
5530 int os::open(const char *path, int oflag, int mode) {
5531 if (strlen(path) > MAX_PATH - 1) {
5532 errno = ENAMETOOLONG;
5533 return -1;
5534 }
5535
5536 // All file descriptors that are opened in the Java process and not
5537 // specifically destined for a subprocess should have the close-on-exec
5538 // flag set. If we don't set it, then careless 3rd party native code
5539 // might fork and exec without closing all appropriate file descriptors
5540 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5541 // turn might:
5542 //
5543 // - cause end-of-file to fail to be detected on some file
5544 // descriptors, resulting in mysterious hangs, or
5545 //
5546 // - might cause an fopen in the subprocess to fail on a system
5547 // suffering from bug 1085341.
5548 //
5549 // (Yes, the default setting of the close-on-exec flag is a Unix
5550 // design flaw)
5551 //
5552 // See:
5553 // 1085341: 32-bit stdio routines should support file descriptors >255
5554 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5555 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5556 //
5557 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5558 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5559 // because it saves a system call and removes a small window where the flag
5560 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5561 // and we fall back to using FD_CLOEXEC (see below).
5562 #ifdef O_CLOEXEC
5563 oflag |= O_CLOEXEC;
5564 #endif
5565
5566 int fd = ::open64(path, oflag, mode);
5567 if (fd == -1) return -1;
5568
5569 //If the open succeeded, the file might still be a directory
5570 {
5571 struct stat64 buf64;
5572 int ret = ::fstat64(fd, &buf64);
5573 int st_mode = buf64.st_mode;
5574
5575 if (ret != -1) {
5576 if ((st_mode & S_IFMT) == S_IFDIR) {
5577 errno = EISDIR;
5578 ::close(fd);
5579 return -1;
5580 }
5581 } else {
5582 ::close(fd);
5583 return -1;
5584 }
5585 }
5586
5587 #ifdef FD_CLOEXEC
5588 // Validate that the use of the O_CLOEXEC flag on open above worked.
5589 // With recent kernels, we will perform this check exactly once.
5590 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5591 if (!O_CLOEXEC_is_known_to_work) {
5592 int flags = ::fcntl(fd, F_GETFD);
5593 if (flags != -1) {
5594 if ((flags & FD_CLOEXEC) != 0)
5595 O_CLOEXEC_is_known_to_work = 1;
5596 else
5597 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5598 }
5599 }
5600 #endif
5601
5602 return fd;
5603 }
5604
5605
5606 // create binary file, rewriting existing file if required
5607 int os::create_binary_file(const char* path, bool rewrite_existing) {
5608 int oflags = O_WRONLY | O_CREAT;
5609 if (!rewrite_existing) {
5610 oflags |= O_EXCL;
5611 }
5612 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5613 }
5614
5615 // return current position of file pointer
5616 jlong os::current_file_offset(int fd) {
5617 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5618 }
5619
5620 // move file pointer to the specified offset
5621 jlong os::seek_to_file_offset(int fd, jlong offset) {
5622 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5623 }
5624
5625 // This code originates from JDK's sysAvailable
5626 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5627
5628 int os::available(int fd, jlong *bytes) {
5629 jlong cur, end;
5630 int mode;
5631 struct stat64 buf64;
5632
5633 if (::fstat64(fd, &buf64) >= 0) {
5634 mode = buf64.st_mode;
5635 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5636 int n;
5637 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5638 *bytes = n;
5639 return 1;
5640 }
5641 }
5642 }
5643 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5644 return 0;
5645 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5646 return 0;
5647 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5648 return 0;
5649 }
5650 *bytes = end - cur;
5651 return 1;
5652 }
5653
5654 // Map a block of memory.
5655 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5656 char *addr, size_t bytes, bool read_only,
5657 bool allow_exec) {
5658 int prot;
5659 int flags = MAP_PRIVATE;
5660
5661 if (read_only) {
5662 prot = PROT_READ;
5663 } else {
5664 prot = PROT_READ | PROT_WRITE;
5665 }
5666
5667 if (allow_exec) {
5668 prot |= PROT_EXEC;
5669 }
5670
5671 if (addr != NULL) {
5672 flags |= MAP_FIXED;
5673 }
5674
5675 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5676 fd, file_offset);
5677 if (mapped_address == MAP_FAILED) {
5678 return NULL;
5679 }
5680 return mapped_address;
5681 }
5682
5683
5684 // Remap a block of memory.
5685 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5686 char *addr, size_t bytes, bool read_only,
5687 bool allow_exec) {
5688 // same as map_memory() on this OS
5689 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5690 allow_exec);
5691 }
5692
5693
5694 // Unmap a block of memory.
5695 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5696 return munmap(addr, bytes) == 0;
5697 }
5698
5699 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5700
5701 static jlong fast_cpu_time(Thread *thread) {
5702 clockid_t clockid;
5703 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5704 &clockid);
5705 if (rc == 0) {
5706 return os::Linux::fast_thread_cpu_time(clockid);
5707 } else {
5708 // It's possible to encounter a terminated native thread that failed
5709 // to detach itself from the VM - which should result in ESRCH.
5710 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5711 return -1;
5712 }
5713 }
5714
5715 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5716 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5717 // of a thread.
5718 //
5719 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5720 // the fast estimate available on the platform.
5721
5722 jlong os::current_thread_cpu_time() {
5723 if (os::Linux::supports_fast_thread_cpu_time()) {
5724 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5725 } else {
5726 // return user + sys since the cost is the same
5727 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5728 }
5729 }
5730
5731 jlong os::thread_cpu_time(Thread* thread) {
5732 // consistent with what current_thread_cpu_time() returns
5733 if (os::Linux::supports_fast_thread_cpu_time()) {
5734 return fast_cpu_time(thread);
5735 } else {
5736 return slow_thread_cpu_time(thread, true /* user + sys */);
5737 }
5738 }
5739
5740 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5741 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5742 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5743 } else {
5744 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5745 }
5746 }
5747
5748 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5749 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5750 return fast_cpu_time(thread);
5751 } else {
5752 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5753 }
5754 }
5755
5756 // -1 on error.
5757 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5758 pid_t tid = thread->osthread()->thread_id();
5759 char *s;
5760 char stat[2048];
5761 int statlen;
5762 char proc_name[64];
5763 int count;
5764 long sys_time, user_time;
5765 char cdummy;
5766 int idummy;
5767 long ldummy;
5768 FILE *fp;
5769
5770 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5771 fp = fopen(proc_name, "r");
5772 if (fp == NULL) return -1;
5773 statlen = fread(stat, 1, 2047, fp);
5774 stat[statlen] = '\0';
5775 fclose(fp);
5776
5777 // Skip pid and the command string. Note that we could be dealing with
5778 // weird command names, e.g. user could decide to rename java launcher
5779 // to "java 1.4.2 :)", then the stat file would look like
5780 // 1234 (java 1.4.2 :)) R ... ...
5781 // We don't really need to know the command string, just find the last
5782 // occurrence of ")" and then start parsing from there. See bug 4726580.
5783 s = strrchr(stat, ')');
5784 if (s == NULL) return -1;
5785
5786 // Skip blank chars
5787 do { s++; } while (s && isspace(*s));
5788
5789 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5790 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5791 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5792 &user_time, &sys_time);
5793 if (count != 13) return -1;
5794 if (user_sys_cpu_time) {
5795 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5796 } else {
5797 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5798 }
5799 }
5800
5801 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5802 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5803 info_ptr->may_skip_backward = false; // elapsed time not wall time
5804 info_ptr->may_skip_forward = false; // elapsed time not wall time
5805 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5806 }
5807
5808 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5809 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5810 info_ptr->may_skip_backward = false; // elapsed time not wall time
5811 info_ptr->may_skip_forward = false; // elapsed time not wall time
5812 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5813 }
5814
5815 bool os::is_thread_cpu_time_supported() {
5816 return true;
5817 }
5818
5819 // System loadavg support. Returns -1 if load average cannot be obtained.
5820 // Linux doesn't yet have a (official) notion of processor sets,
5821 // so just return the system wide load average.
5822 int os::loadavg(double loadavg[], int nelem) {
5823 return ::getloadavg(loadavg, nelem);
5824 }
5825
5826 void os::pause() {
5827 char filename[MAX_PATH];
5828 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5829 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5830 } else {
5831 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5832 }
5833
5834 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5835 if (fd != -1) {
5836 struct stat buf;
5837 ::close(fd);
5838 while (::stat(filename, &buf) == 0) {
5839 (void)::poll(NULL, 0, 100);
5840 }
5841 } else {
5842 jio_fprintf(stderr,
5843 "Could not open pause file '%s', continuing immediately.\n", filename);
5844 }
5845 }
5846
5847 extern char** environ;
5848
5849 // Run the specified command in a separate process. Return its exit value,
5850 // or -1 on failure (e.g. can't fork a new process).
5851 // Unlike system(), this function can be called from signal handler. It
5852 // doesn't block SIGINT et al.
5853 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5854 const char * argv[4] = {"sh", "-c", cmd, NULL};
5855
5856 pid_t pid ;
5857
5858 if (use_vfork_if_available) {
5859 pid = vfork();
5860 } else {
5861 pid = fork();
5862 }
5863
5864 if (pid < 0) {
5865 // fork failed
5866 return -1;
5867
5868 } else if (pid == 0) {
5869 // child process
5870
5871 execve("/bin/sh", (char* const*)argv, environ);
5872
5873 // execve failed
5874 _exit(-1);
5875
5876 } else {
5877 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5878 // care about the actual exit code, for now.
5879
5880 int status;
5881
5882 // Wait for the child process to exit. This returns immediately if
5883 // the child has already exited. */
5884 while (waitpid(pid, &status, 0) < 0) {
5885 switch (errno) {
5886 case ECHILD: return 0;
5887 case EINTR: break;
5888 default: return -1;
5889 }
5890 }
5891
5892 if (WIFEXITED(status)) {
5893 // The child exited normally; get its exit code.
5894 return WEXITSTATUS(status);
5895 } else if (WIFSIGNALED(status)) {
5896 // The child exited because of a signal
5897 // The best value to return is 0x80 + signal number,
5898 // because that is what all Unix shells do, and because
5899 // it allows callers to distinguish between process exit and
5900 // process death by signal.
5901 return 0x80 + WTERMSIG(status);
5902 } else {
5903 // Unknown exit code; pass it through
5904 return status;
5905 }
5906 }
5907 }
5908
5909 // Get the default path to the core file
5910 // Returns the length of the string
5911 int os::get_core_path(char* buffer, size_t bufferSize) {
5912 /*
5913 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5914 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5915 */
5916 const int core_pattern_len = 129;
5917 char core_pattern[core_pattern_len] = {0};
5918
5919 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5920 if (core_pattern_file == -1) {
5921 return -1;
5922 }
5923
5924 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5925 ::close(core_pattern_file);
5926 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5927 return -1;
5928 }
5929 if (core_pattern[ret-1] == '\n') {
5930 core_pattern[ret-1] = '\0';
5931 } else {
5932 core_pattern[ret] = '\0';
5933 }
5934
5935 char *pid_pos = strstr(core_pattern, "%p");
5936 int written;
5937
5938 if (core_pattern[0] == '/') {
5939 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5940 } else {
5941 char cwd[PATH_MAX];
5942
5943 const char* p = get_current_directory(cwd, PATH_MAX);
5944 if (p == NULL) {
5945 return -1;
5946 }
5947
5948 if (core_pattern[0] == '|') {
5949 written = jio_snprintf(buffer, bufferSize,
5950 "\"%s\" (or dumping to %s/core.%d)",
5951 &core_pattern[1], p, current_process_id());
5952 } else {
5953 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5954 }
5955 }
5956
5957 if (written < 0) {
5958 return -1;
5959 }
5960
5961 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5962 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5963
5964 if (core_uses_pid_file != -1) {
5965 char core_uses_pid = 0;
5966 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5967 ::close(core_uses_pid_file);
5968
5969 if (core_uses_pid == '1') {
5970 jio_snprintf(buffer + written, bufferSize - written,
5971 ".%d", current_process_id());
5972 }
5973 }
5974 }
5975
5976 return strlen(buffer);
5977 }
5978
5979 bool os::start_debugging(char *buf, int buflen) {
5980 int len = (int)strlen(buf);
5981 char *p = &buf[len];
5982
5983 jio_snprintf(p, buflen-len,
5984 "\n\n"
5985 "Do you want to debug the problem?\n\n"
5986 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5987 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5988 "Otherwise, press RETURN to abort...",
5989 os::current_process_id(), os::current_process_id(),
5990 os::current_thread_id(), os::current_thread_id());
5991
5992 bool yes = os::message_box("Unexpected Error", buf);
5993
5994 if (yes) {
5995 // yes, user asked VM to launch debugger
5996 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5997 os::current_process_id(), os::current_process_id());
5998
5999 os::fork_and_exec(buf);
6000 yes = false;
6001 }
6002 return yes;
6003 }
6004
6005
6006 // Java/Compiler thread:
6007 //
6008 // Low memory addresses
6009 // P0 +------------------------+
6010 // | |\ Java thread created by VM does not have glibc
6011 // | glibc guard page | - guard page, attached Java thread usually has
6012 // | |/ 1 glibc guard page.
6013 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6014 // | |\
6015 // | HotSpot Guard Pages | - red, yellow and reserved pages
6016 // | |/
6017 // +------------------------+ JavaThread::stack_reserved_zone_base()
6018 // | |\
6019 // | Normal Stack | -
6020 // | |/
6021 // P2 +------------------------+ Thread::stack_base()
6022 //
6023 // Non-Java thread:
6024 //
6025 // Low memory addresses
6026 // P0 +------------------------+
6027 // | |\
6028 // | glibc guard page | - usually 1 page
6029 // | |/
6030 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6031 // | |\
6032 // | Normal Stack | -
6033 // | |/
6034 // P2 +------------------------+ Thread::stack_base()
6035 //
6036 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6037 // returned from pthread_attr_getstack().
6038 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6039 // of the stack size given in pthread_attr. We work around this for
6040 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6041 //
6042 #ifndef ZERO
6043 static void current_stack_region(address * bottom, size_t * size) {
6044 if (os::is_primordial_thread()) {
6045 // primordial thread needs special handling because pthread_getattr_np()
6046 // may return bogus value.
6047 *bottom = os::Linux::initial_thread_stack_bottom();
6048 *size = os::Linux::initial_thread_stack_size();
6049 } else {
6050 pthread_attr_t attr;
6051
6052 int rslt = pthread_getattr_np(pthread_self(), &attr);
6053
6054 // JVM needs to know exact stack location, abort if it fails
6055 if (rslt != 0) {
6056 if (rslt == ENOMEM) {
6057 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6058 } else {
6059 fatal("pthread_getattr_np failed with error = %d", rslt);
6060 }
6061 }
6062
6063 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6064 fatal("Cannot locate current stack attributes!");
6065 }
6066
6067 // Work around NPTL stack guard error.
6068 size_t guard_size = 0;
6069 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6070 if (rslt != 0) {
6071 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6072 }
6073 *bottom += guard_size;
6074 *size -= guard_size;
6075
6076 pthread_attr_destroy(&attr);
6077
6078 }
6079 assert(os::current_stack_pointer() >= *bottom &&
6080 os::current_stack_pointer() < *bottom + *size, "just checking");
6081 }
6082
6083 address os::current_stack_base() {
6084 address bottom;
6085 size_t size;
6086 current_stack_region(&bottom, &size);
6087 return (bottom + size);
6088 }
6089
6090 size_t os::current_stack_size() {
6091 // This stack size includes the usable stack and HotSpot guard pages
6092 // (for the threads that have Hotspot guard pages).
6093 address bottom;
6094 size_t size;
6095 current_stack_region(&bottom, &size);
6096 return size;
6097 }
6098 #endif
6099
6100 static inline struct timespec get_mtime(const char* filename) {
6101 struct stat st;
6102 int ret = os::stat(filename, &st);
6103 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6104 return st.st_mtim;
6105 }
6106
6107 int os::compare_file_modified_times(const char* file1, const char* file2) {
6108 struct timespec filetime1 = get_mtime(file1);
6109 struct timespec filetime2 = get_mtime(file2);
6110 int diff = filetime1.tv_sec - filetime2.tv_sec;
6111 if (diff == 0) {
6112 return filetime1.tv_nsec - filetime2.tv_nsec;
6113 }
6114 return diff;
6115 }
6116
6117 /////////////// Unit tests ///////////////
6118
6119 #ifndef PRODUCT
6120
6121 #define test_log(...) \
6122 do { \
6123 if (VerboseInternalVMTests) { \
6124 tty->print_cr(__VA_ARGS__); \
6125 tty->flush(); \
6126 } \
6127 } while (false)
6128
6129 class TestReserveMemorySpecial : AllStatic {
6130 public:
6131 static void small_page_write(void* addr, size_t size) {
6132 size_t page_size = os::vm_page_size();
6133
6134 char* end = (char*)addr + size;
6135 for (char* p = (char*)addr; p < end; p += page_size) {
6136 *p = 1;
6137 }
6138 }
6139
6140 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6141 if (!UseHugeTLBFS) {
6142 return;
6143 }
6144
6145 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6146
6147 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6148
6149 if (addr != NULL) {
6150 small_page_write(addr, size);
6151
6152 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6153 }
6154 }
6155
6156 static void test_reserve_memory_special_huge_tlbfs_only() {
6157 if (!UseHugeTLBFS) {
6158 return;
6159 }
6160
6161 size_t lp = os::large_page_size();
6162
6163 for (size_t size = lp; size <= lp * 10; size += lp) {
6164 test_reserve_memory_special_huge_tlbfs_only(size);
6165 }
6166 }
6167
6168 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6169 size_t lp = os::large_page_size();
6170 size_t ag = os::vm_allocation_granularity();
6171
6172 // sizes to test
6173 const size_t sizes[] = {
6174 lp, lp + ag, lp + lp / 2, lp * 2,
6175 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6176 lp * 10, lp * 10 + lp / 2
6177 };
6178 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6179
6180 // For each size/alignment combination, we test three scenarios:
6181 // 1) with req_addr == NULL
6182 // 2) with a non-null req_addr at which we expect to successfully allocate
6183 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6184 // expect the allocation to either fail or to ignore req_addr
6185
6186 // Pre-allocate two areas; they shall be as large as the largest allocation
6187 // and aligned to the largest alignment we will be testing.
6188 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6189 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6190 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6191 -1, 0);
6192 assert(mapping1 != MAP_FAILED, "should work");
6193
6194 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6195 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6196 -1, 0);
6197 assert(mapping2 != MAP_FAILED, "should work");
6198
6199 // Unmap the first mapping, but leave the second mapping intact: the first
6200 // mapping will serve as a value for a "good" req_addr (case 2). The second
6201 // mapping, still intact, as "bad" req_addr (case 3).
6202 ::munmap(mapping1, mapping_size);
6203
6204 // Case 1
6205 test_log("%s, req_addr NULL:", __FUNCTION__);
6206 test_log("size align result");
6207
6208 for (int i = 0; i < num_sizes; i++) {
6209 const size_t size = sizes[i];
6210 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6211 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6212 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6213 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6214 if (p != NULL) {
6215 assert(is_aligned(p, alignment), "must be");
6216 small_page_write(p, size);
6217 os::Linux::release_memory_special_huge_tlbfs(p, size);
6218 }
6219 }
6220 }
6221
6222 // Case 2
6223 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6224 test_log("size align req_addr result");
6225
6226 for (int i = 0; i < num_sizes; i++) {
6227 const size_t size = sizes[i];
6228 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6229 char* const req_addr = align_up(mapping1, alignment);
6230 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6231 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6232 size, alignment, p2i(req_addr), p2i(p),
6233 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6234 if (p != NULL) {
6235 assert(p == req_addr, "must be");
6236 small_page_write(p, size);
6237 os::Linux::release_memory_special_huge_tlbfs(p, size);
6238 }
6239 }
6240 }
6241
6242 // Case 3
6243 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6244 test_log("size align req_addr result");
6245
6246 for (int i = 0; i < num_sizes; i++) {
6247 const size_t size = sizes[i];
6248 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6249 char* const req_addr = align_up(mapping2, alignment);
6250 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6251 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6252 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6253 // as the area around req_addr contains already existing mappings, the API should always
6254 // return NULL (as per contract, it cannot return another address)
6255 assert(p == NULL, "must be");
6256 }
6257 }
6258
6259 ::munmap(mapping2, mapping_size);
6260
6261 }
6262
6263 static void test_reserve_memory_special_huge_tlbfs() {
6264 if (!UseHugeTLBFS) {
6265 return;
6266 }
6267
6268 test_reserve_memory_special_huge_tlbfs_only();
6269 test_reserve_memory_special_huge_tlbfs_mixed();
6270 }
6271
6272 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6273 if (!UseSHM) {
6274 return;
6275 }
6276
6277 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6278
6279 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6280
6281 if (addr != NULL) {
6282 assert(is_aligned(addr, alignment), "Check");
6283 assert(is_aligned(addr, os::large_page_size()), "Check");
6284
6285 small_page_write(addr, size);
6286
6287 os::Linux::release_memory_special_shm(addr, size);
6288 }
6289 }
6290
6291 static void test_reserve_memory_special_shm() {
6292 size_t lp = os::large_page_size();
6293 size_t ag = os::vm_allocation_granularity();
6294
6295 for (size_t size = ag; size < lp * 3; size += ag) {
6296 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6297 test_reserve_memory_special_shm(size, alignment);
6298 }
6299 }
6300 }
6301
6302 static void test() {
6303 test_reserve_memory_special_huge_tlbfs();
6304 test_reserve_memory_special_shm();
6305 }
6306 };
6307
6308 void TestReserveMemorySpecial_test() {
6309 TestReserveMemorySpecial::test();
6310 }
6311
6312 #endif