1 /*
2 * Copyright (c) 1999, 2020, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_posix.inline.hpp"
42 #include "os_share_linux.hpp"
43 #include "osContainer_linux.hpp"
44 #include "prims/jniFastGetField.hpp"
45 #include "prims/jvm_misc.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/atomic.hpp"
48 #include "runtime/extendedPC.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/interfaceSupport.inline.hpp"
51 #include "runtime/init.hpp"
52 #include "runtime/java.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/mutexLocker.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/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/powerOfTwo.hpp"
78 #include "utilities/vmError.hpp"
79
80 // put OS-includes here
81 # include <sys/types.h>
82 # include <sys/mman.h>
83 # include <sys/stat.h>
84 # include <sys/select.h>
85 # include <pthread.h>
86 # include <signal.h>
87 # include <endian.h>
88 # include <errno.h>
89 # include <dlfcn.h>
90 # include <stdio.h>
91 # include <unistd.h>
92 # include <sys/resource.h>
93 # include <pthread.h>
94 # include <sys/stat.h>
95 # include <sys/time.h>
96 # include <sys/times.h>
97 # include <sys/utsname.h>
98 # include <sys/socket.h>
99 # include <sys/wait.h>
100 # include <pwd.h>
101 # include <poll.h>
102 # include <fcntl.h>
103 # include <string.h>
104 # include <syscall.h>
105 # include <sys/sysinfo.h>
106 # include <gnu/libc-version.h>
107 # include <sys/ipc.h>
108 # include <sys/shm.h>
109 # include <link.h>
110 # include <stdint.h>
111 # include <inttypes.h>
112 # include <sys/ioctl.h>
113
114 #ifndef _GNU_SOURCE
115 #define _GNU_SOURCE
116 #include <sched.h>
117 #undef _GNU_SOURCE
118 #else
119 #include <sched.h>
120 #endif
121
122 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
123 // getrusage() is prepared to handle the associated failure.
124 #ifndef RUSAGE_THREAD
125 #define RUSAGE_THREAD (1) /* only the calling thread */
126 #endif
127
128 #define MAX_PATH (2 * K)
129
130 #define MAX_SECS 100000000
131
132 // for timer info max values which include all bits
133 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
134
135 enum CoredumpFilterBit {
136 FILE_BACKED_PVT_BIT = 1 << 2,
137 FILE_BACKED_SHARED_BIT = 1 << 3,
138 LARGEPAGES_BIT = 1 << 6,
139 DAX_SHARED_BIT = 1 << 8
140 };
141
142 ////////////////////////////////////////////////////////////////////////////////
143 // global variables
144 julong os::Linux::_physical_memory = 0;
145
146 address os::Linux::_initial_thread_stack_bottom = NULL;
147 uintptr_t os::Linux::_initial_thread_stack_size = 0;
148
149 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
150 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = 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 const char * os::Linux::_glibc_version = NULL;
155 const char * os::Linux::_libpthread_version = NULL;
156
157 static jlong initial_time_count=0;
158
159 static int clock_tics_per_sec = 100;
160
161 // If the VM might have been created on the primordial thread, we need to resolve the
162 // primordial thread stack bounds and check if the current thread might be the
163 // primordial thread in places. If we know that the primordial thread is never used,
164 // such as when the VM was created by one of the standard java launchers, we can
165 // avoid this
166 static bool suppress_primordial_thread_resolution = false;
167
168 // For diagnostics to print a message once. see run_periodic_checks
169 static sigset_t check_signal_done;
170 static bool check_signals = true;
171
172 // Signal number used to suspend/resume a thread
173
174 // do not use any signal number less than SIGSEGV, see 4355769
175 static int SR_signum = SIGUSR2;
176 sigset_t SR_sigset;
177
178 // utility functions
179
180 static int SR_initialize();
181
182 julong os::available_memory() {
183 return Linux::available_memory();
184 }
185
186 julong os::Linux::available_memory() {
187 // values in struct sysinfo are "unsigned long"
188 struct sysinfo si;
189 julong avail_mem;
190
191 if (OSContainer::is_containerized()) {
192 jlong mem_limit, mem_usage;
193 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
194 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
195 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
196 }
197 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
198 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
199 }
200 if (mem_limit > 0 && mem_usage > 0 ) {
201 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
202 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
203 return avail_mem;
204 }
205 }
206
207 sysinfo(&si);
208 avail_mem = (julong)si.freeram * si.mem_unit;
209 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
210 return avail_mem;
211 }
212
213 julong os::physical_memory() {
214 jlong phys_mem = 0;
215 if (OSContainer::is_containerized()) {
216 jlong mem_limit;
217 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
218 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
219 return mem_limit;
220 }
221 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
222 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
223 }
224
225 phys_mem = Linux::physical_memory();
226 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
227 return phys_mem;
228 }
229
230 static uint64_t initial_total_ticks = 0;
231 static uint64_t initial_steal_ticks = 0;
232 static bool has_initial_tick_info = false;
233
234 static void next_line(FILE *f) {
235 int c;
236 do {
237 c = fgetc(f);
238 } while (c != '\n' && c != EOF);
239 }
240
241 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
242 FILE* fh;
243 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
244 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
245 // irq: time servicing interrupts; softirq: time servicing softirqs
246 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
247 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
248 uint64_t stealTicks = 0;
249 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
250 // control of the Linux kernel
251 uint64_t guestNiceTicks = 0;
252 int logical_cpu = -1;
253 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
254 int n;
255
256 memset(pticks, 0, sizeof(CPUPerfTicks));
257
258 if ((fh = fopen("/proc/stat", "r")) == NULL) {
259 return false;
260 }
261
262 if (which_logical_cpu == -1) {
263 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
264 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
265 UINT64_FORMAT " " UINT64_FORMAT " ",
266 &userTicks, &niceTicks, &systemTicks, &idleTicks,
267 &iowTicks, &irqTicks, &sirqTicks,
268 &stealTicks, &guestNiceTicks);
269 } else {
270 // Move to next line
271 next_line(fh);
272
273 // find the line for requested cpu faster to just iterate linefeeds?
274 for (int i = 0; i < which_logical_cpu; i++) {
275 next_line(fh);
276 }
277
278 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
279 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
280 UINT64_FORMAT " " UINT64_FORMAT " ",
281 &logical_cpu, &userTicks, &niceTicks,
282 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
283 &stealTicks, &guestNiceTicks);
284 }
285
286 fclose(fh);
287 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
288 return false;
289 }
290 pticks->used = userTicks + niceTicks;
291 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
292 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
293 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
294
295 if (n > required_tickinfo_count + 3) {
296 pticks->steal = stealTicks;
297 pticks->has_steal_ticks = true;
298 } else {
299 pticks->steal = 0;
300 pticks->has_steal_ticks = false;
301 }
302
303 return true;
304 }
305
306 // Return true if user is running as root.
307
308 bool os::have_special_privileges() {
309 static bool init = false;
310 static bool privileges = false;
311 if (!init) {
312 privileges = (getuid() != geteuid()) || (getgid() != getegid());
313 init = true;
314 }
315 return privileges;
316 }
317
318
319 #ifndef SYS_gettid
320 // i386: 224, ia64: 1105, amd64: 186, sparc 143
321 #ifdef __ia64__
322 #define SYS_gettid 1105
323 #else
324 #ifdef __i386__
325 #define SYS_gettid 224
326 #else
327 #ifdef __amd64__
328 #define SYS_gettid 186
329 #else
330 #ifdef __sparc__
331 #define SYS_gettid 143
332 #else
333 #error define gettid for the arch
334 #endif
335 #endif
336 #endif
337 #endif
338 #endif
339
340
341 // pid_t gettid()
342 //
343 // Returns the kernel thread id of the currently running thread. Kernel
344 // thread id is used to access /proc.
345 pid_t os::Linux::gettid() {
346 int rslt = syscall(SYS_gettid);
347 assert(rslt != -1, "must be."); // old linuxthreads implementation?
348 return (pid_t)rslt;
349 }
350
351 // Most versions of linux have a bug where the number of processors are
352 // determined by looking at the /proc file system. In a chroot environment,
353 // the system call returns 1.
354 static bool unsafe_chroot_detected = false;
355 static const char *unstable_chroot_error = "/proc file system not found.\n"
356 "Java may be unstable running multithreaded in a chroot "
357 "environment on Linux when /proc filesystem is not mounted.";
358
359 void os::Linux::initialize_system_info() {
360 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
361 if (processor_count() == 1) {
362 pid_t pid = os::Linux::gettid();
363 char fname[32];
364 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
365 FILE *fp = fopen(fname, "r");
366 if (fp == NULL) {
367 unsafe_chroot_detected = true;
368 } else {
369 fclose(fp);
370 }
371 }
372 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
373 assert(processor_count() > 0, "linux error");
374 }
375
376 void os::init_system_properties_values() {
377 // The next steps are taken in the product version:
378 //
379 // Obtain the JAVA_HOME value from the location of libjvm.so.
380 // This library should be located at:
381 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
382 //
383 // If "/jre/lib/" appears at the right place in the path, then we
384 // assume libjvm.so is installed in a JDK and we use this path.
385 //
386 // Otherwise exit with message: "Could not create the Java virtual machine."
387 //
388 // The following extra steps are taken in the debugging version:
389 //
390 // If "/jre/lib/" does NOT appear at the right place in the path
391 // instead of exit check for $JAVA_HOME environment variable.
392 //
393 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
394 // then we append a fake suffix "hotspot/libjvm.so" to this path so
395 // it looks like libjvm.so is installed there
396 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
397 //
398 // Otherwise exit.
399 //
400 // Important note: if the location of libjvm.so changes this
401 // code needs to be changed accordingly.
402
403 // See ld(1):
404 // The linker uses the following search paths to locate required
405 // shared libraries:
406 // 1: ...
407 // ...
408 // 7: The default directories, normally /lib and /usr/lib.
409 #ifndef OVERRIDE_LIBPATH
410 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
411 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
412 #else
413 #define DEFAULT_LIBPATH "/lib:/usr/lib"
414 #endif
415 #else
416 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
417 #endif
418
419 // Base path of extensions installed on the system.
420 #define SYS_EXT_DIR "/usr/java/packages"
421 #define EXTENSIONS_DIR "/lib/ext"
422
423 // Buffer that fits several sprintfs.
424 // Note that the space for the colon and the trailing null are provided
425 // by the nulls included by the sizeof operator.
426 const size_t bufsize =
427 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
428 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
429 char *buf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
430
431 // sysclasspath, java_home, dll_dir
432 {
433 char *pslash;
434 os::jvm_path(buf, bufsize);
435
436 // Found the full path to libjvm.so.
437 // Now cut the path to <java_home>/jre if we can.
438 pslash = strrchr(buf, '/');
439 if (pslash != NULL) {
440 *pslash = '\0'; // Get rid of /libjvm.so.
441 }
442 pslash = strrchr(buf, '/');
443 if (pslash != NULL) {
444 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
445 }
446 Arguments::set_dll_dir(buf);
447
448 if (pslash != NULL) {
449 pslash = strrchr(buf, '/');
450 if (pslash != NULL) {
451 *pslash = '\0'; // Get rid of /lib.
452 }
453 }
454 Arguments::set_java_home(buf);
455 if (!set_boot_path('/', ':')) {
456 vm_exit_during_initialization("Failed setting boot class path.", NULL);
457 }
458 }
459
460 // Where to look for native libraries.
461 //
462 // Note: Due to a legacy implementation, most of the library path
463 // is set in the launcher. This was to accomodate linking restrictions
464 // on legacy Linux implementations (which are no longer supported).
465 // Eventually, all the library path setting will be done here.
466 //
467 // However, to prevent the proliferation of improperly built native
468 // libraries, the new path component /usr/java/packages is added here.
469 // Eventually, all the library path setting will be done here.
470 {
471 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
472 // should always exist (until the legacy problem cited above is
473 // addressed).
474 const char *v = ::getenv("LD_LIBRARY_PATH");
475 const char *v_colon = ":";
476 if (v == NULL) { v = ""; v_colon = ""; }
477 // That's +1 for the colon and +1 for the trailing '\0'.
478 char *ld_library_path = NEW_C_HEAP_ARRAY(char,
479 strlen(v) + 1 +
480 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
481 mtInternal);
482 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
483 Arguments::set_library_path(ld_library_path);
484 FREE_C_HEAP_ARRAY(char, ld_library_path);
485 }
486
487 // Extensions directories.
488 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
489 Arguments::set_ext_dirs(buf);
490
491 FREE_C_HEAP_ARRAY(char, buf);
492
493 #undef DEFAULT_LIBPATH
494 #undef SYS_EXT_DIR
495 #undef EXTENSIONS_DIR
496 }
497
498 ////////////////////////////////////////////////////////////////////////////////
499 // breakpoint support
500
501 void os::breakpoint() {
502 BREAKPOINT;
503 }
504
505 extern "C" void breakpoint() {
506 // use debugger to set breakpoint here
507 }
508
509 ////////////////////////////////////////////////////////////////////////////////
510 // signal support
511
512 debug_only(static bool signal_sets_initialized = false);
513 static sigset_t unblocked_sigs, vm_sigs;
514
515 void os::Linux::signal_sets_init() {
516 // Should also have an assertion stating we are still single-threaded.
517 assert(!signal_sets_initialized, "Already initialized");
518 // Fill in signals that are necessarily unblocked for all threads in
519 // the VM. Currently, we unblock the following signals:
520 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
521 // by -Xrs (=ReduceSignalUsage));
522 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
523 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
524 // the dispositions or masks wrt these signals.
525 // Programs embedding the VM that want to use the above signals for their
526 // own purposes must, at this time, use the "-Xrs" option to prevent
527 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
528 // (See bug 4345157, and other related bugs).
529 // In reality, though, unblocking these signals is really a nop, since
530 // these signals are not blocked by default.
531 sigemptyset(&unblocked_sigs);
532 sigaddset(&unblocked_sigs, SIGILL);
533 sigaddset(&unblocked_sigs, SIGSEGV);
534 sigaddset(&unblocked_sigs, SIGBUS);
535 sigaddset(&unblocked_sigs, SIGFPE);
536 #if defined(PPC64)
537 sigaddset(&unblocked_sigs, SIGTRAP);
538 #endif
539 sigaddset(&unblocked_sigs, SR_signum);
540
541 if (!ReduceSignalUsage) {
542 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
543 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
544 }
545 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
546 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
547 }
548 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
549 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
550 }
551 }
552 // Fill in signals that are blocked by all but the VM thread.
553 sigemptyset(&vm_sigs);
554 if (!ReduceSignalUsage) {
555 sigaddset(&vm_sigs, BREAK_SIGNAL);
556 }
557 debug_only(signal_sets_initialized = true);
558
559 }
560
561 // These are signals that are unblocked while a thread is running Java.
562 // (For some reason, they get blocked by default.)
563 sigset_t* os::Linux::unblocked_signals() {
564 assert(signal_sets_initialized, "Not initialized");
565 return &unblocked_sigs;
566 }
567
568 // These are the signals that are blocked while a (non-VM) thread is
569 // running Java. Only the VM thread handles these signals.
570 sigset_t* os::Linux::vm_signals() {
571 assert(signal_sets_initialized, "Not initialized");
572 return &vm_sigs;
573 }
574
575 void os::Linux::hotspot_sigmask(Thread* thread) {
576
577 //Save caller's signal mask before setting VM signal mask
578 sigset_t caller_sigmask;
579 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
580
581 OSThread* osthread = thread->osthread();
582 osthread->set_caller_sigmask(caller_sigmask);
583
584 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
585
586 if (!ReduceSignalUsage) {
587 if (thread->is_VM_thread()) {
588 // Only the VM thread handles BREAK_SIGNAL ...
589 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
590 } else {
591 // ... all other threads block BREAK_SIGNAL
592 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
593 }
594 }
595 }
596
597 //////////////////////////////////////////////////////////////////////////////
598 // detecting pthread library
599
600 void os::Linux::libpthread_init() {
601 // Save glibc and pthread version strings.
602 #if !defined(_CS_GNU_LIBC_VERSION) || \
603 !defined(_CS_GNU_LIBPTHREAD_VERSION)
604 #error "glibc too old (< 2.3.2)"
605 #endif
606
607 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
608 assert(n > 0, "cannot retrieve glibc version");
609 char *str = (char *)malloc(n, mtInternal);
610 confstr(_CS_GNU_LIBC_VERSION, str, n);
611 os::Linux::set_glibc_version(str);
612
613 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
614 assert(n > 0, "cannot retrieve pthread version");
615 str = (char *)malloc(n, mtInternal);
616 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
617 os::Linux::set_libpthread_version(str);
618 }
619
620 /////////////////////////////////////////////////////////////////////////////
621 // thread stack expansion
622
623 // os::Linux::manually_expand_stack() takes care of expanding the thread
624 // stack. Note that this is normally not needed: pthread stacks allocate
625 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
626 // committed. Therefore it is not necessary to expand the stack manually.
627 //
628 // Manually expanding the stack was historically needed on LinuxThreads
629 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
630 // it is kept to deal with very rare corner cases:
631 //
632 // For one, user may run the VM on an own implementation of threads
633 // whose stacks are - like the old LinuxThreads - implemented using
634 // mmap(MAP_GROWSDOWN).
635 //
636 // Also, this coding may be needed if the VM is running on the primordial
637 // thread. Normally we avoid running on the primordial thread; however,
638 // user may still invoke the VM on the primordial thread.
639 //
640 // The following historical comment describes the details about running
641 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
642
643
644 // Force Linux kernel to expand current thread stack. If "bottom" is close
645 // to the stack guard, caller should block all signals.
646 //
647 // MAP_GROWSDOWN:
648 // A special mmap() flag that is used to implement thread stacks. It tells
649 // kernel that the memory region should extend downwards when needed. This
650 // allows early versions of LinuxThreads to only mmap the first few pages
651 // when creating a new thread. Linux kernel will automatically expand thread
652 // stack as needed (on page faults).
653 //
654 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
655 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
656 // region, it's hard to tell if the fault is due to a legitimate stack
657 // access or because of reading/writing non-exist memory (e.g. buffer
658 // overrun). As a rule, if the fault happens below current stack pointer,
659 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
660 // application (see Linux kernel fault.c).
661 //
662 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
663 // stack overflow detection.
664 //
665 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
666 // not use MAP_GROWSDOWN.
667 //
668 // To get around the problem and allow stack banging on Linux, we need to
669 // manually expand thread stack after receiving the SIGSEGV.
670 //
671 // There are two ways to expand thread stack to address "bottom", we used
672 // both of them in JVM before 1.5:
673 // 1. adjust stack pointer first so that it is below "bottom", and then
674 // touch "bottom"
675 // 2. mmap() the page in question
676 //
677 // Now alternate signal stack is gone, it's harder to use 2. For instance,
678 // if current sp is already near the lower end of page 101, and we need to
679 // call mmap() to map page 100, it is possible that part of the mmap() frame
680 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
681 // That will destroy the mmap() frame and cause VM to crash.
682 //
683 // The following code works by adjusting sp first, then accessing the "bottom"
684 // page to force a page fault. Linux kernel will then automatically expand the
685 // stack mapping.
686 //
687 // _expand_stack_to() assumes its frame size is less than page size, which
688 // should always be true if the function is not inlined.
689
690 static void NOINLINE _expand_stack_to(address bottom) {
691 address sp;
692 size_t size;
693 volatile char *p;
694
695 // Adjust bottom to point to the largest address within the same page, it
696 // gives us a one-page buffer if alloca() allocates slightly more memory.
697 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
698 bottom += os::Linux::page_size() - 1;
699
700 // sp might be slightly above current stack pointer; if that's the case, we
701 // will alloca() a little more space than necessary, which is OK. Don't use
702 // os::current_stack_pointer(), as its result can be slightly below current
703 // stack pointer, causing us to not alloca enough to reach "bottom".
704 sp = (address)&sp;
705
706 if (sp > bottom) {
707 size = sp - bottom;
708 p = (volatile char *)alloca(size);
709 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
710 p[0] = '\0';
711 }
712 }
713
714 void os::Linux::expand_stack_to(address bottom) {
715 _expand_stack_to(bottom);
716 }
717
718 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
719 assert(t!=NULL, "just checking");
720 assert(t->osthread()->expanding_stack(), "expand should be set");
721
722 if (t->is_in_usable_stack(addr)) {
723 sigset_t mask_all, old_sigset;
724 sigfillset(&mask_all);
725 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
726 _expand_stack_to(addr);
727 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
728 return true;
729 }
730 return false;
731 }
732
733 //////////////////////////////////////////////////////////////////////////////
734 // create new thread
735
736 // Thread start routine for all newly created threads
737 static void *thread_native_entry(Thread *thread) {
738
739 thread->record_stack_base_and_size();
740
741 // Try to randomize the cache line index of hot stack frames.
742 // This helps when threads of the same stack traces evict each other's
743 // cache lines. The threads can be either from the same JVM instance, or
744 // from different JVM instances. The benefit is especially true for
745 // processors with hyperthreading technology.
746 static int counter = 0;
747 int pid = os::current_process_id();
748 alloca(((pid ^ counter++) & 7) * 128);
749
750 thread->initialize_thread_current();
751
752 OSThread* osthread = thread->osthread();
753 Monitor* sync = osthread->startThread_lock();
754
755 osthread->set_thread_id(os::current_thread_id());
756
757 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
758 os::current_thread_id(), (uintx) pthread_self());
759
760 if (UseNUMA) {
761 int lgrp_id = os::numa_get_group_id();
762 if (lgrp_id != -1) {
763 thread->set_lgrp_id(lgrp_id);
764 }
765 }
766 // initialize signal mask for this thread
767 os::Linux::hotspot_sigmask(thread);
768
769 // initialize floating point control register
770 os::Linux::init_thread_fpu_state();
771
772 // handshaking with parent thread
773 {
774 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
775
776 // notify parent thread
777 osthread->set_state(INITIALIZED);
778 sync->notify_all();
779
780 // wait until os::start_thread()
781 while (osthread->get_state() == INITIALIZED) {
782 sync->wait_without_safepoint_check();
783 }
784 }
785
786 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
787
788 // call one more level start routine
789 thread->call_run();
790
791 // Note: at this point the thread object may already have deleted itself.
792 // Prevent dereferencing it from here on out.
793 thread = NULL;
794
795 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
796 os::current_thread_id(), (uintx) pthread_self());
797
798 return 0;
799 }
800
801 // On Linux, glibc places static TLS blocks (for __thread variables) on
802 // the thread stack. This decreases the stack size actually available
803 // to threads.
804 //
805 // For large static TLS sizes, this may cause threads to malfunction due
806 // to insufficient stack space. This is a well-known issue in glibc:
807 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
808 //
809 // As a workaround, we call a private but assumed-stable glibc function,
810 // __pthread_get_minstack() to obtain the minstack size and derive the
811 // static TLS size from it. We then increase the user requested stack
812 // size by this TLS size.
813 //
814 // Due to compatibility concerns, this size adjustment is opt-in and
815 // controlled via AdjustStackSizeForTLS.
816 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
817
818 GetMinStack _get_minstack_func = NULL;
819
820 static void get_minstack_init() {
821 _get_minstack_func =
822 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
823 log_info(os, thread)("Lookup of __pthread_get_minstack %s",
824 _get_minstack_func == NULL ? "failed" : "succeeded");
825 }
826
827 // Returns the size of the static TLS area glibc puts on thread stacks.
828 // The value is cached on first use, which occurs when the first thread
829 // is created during VM initialization.
830 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
831 size_t tls_size = 0;
832 if (_get_minstack_func != NULL) {
833 // Obtain the pthread minstack size by calling __pthread_get_minstack.
834 size_t minstack_size = _get_minstack_func(attr);
835
836 // Remove non-TLS area size included in minstack size returned
837 // by __pthread_get_minstack() to get the static TLS size.
838 // In glibc before 2.27, minstack size includes guard_size.
839 // In glibc 2.27 and later, guard_size is automatically added
840 // to the stack size by pthread_create and is no longer included
841 // in minstack size. In both cases, the guard_size is taken into
842 // account, so there is no need to adjust the result for that.
843 //
844 // Although __pthread_get_minstack() is a private glibc function,
845 // it is expected to have a stable behavior across future glibc
846 // versions while glibc still allocates the static TLS blocks off
847 // the stack. Following is glibc 2.28 __pthread_get_minstack():
848 //
849 // size_t
850 // __pthread_get_minstack (const pthread_attr_t *attr)
851 // {
852 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
853 // }
854 //
855 //
856 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
857 // if check is done for precaution.
858 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
859 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
860 }
861 }
862
863 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
864 tls_size);
865 return tls_size;
866 }
867
868 bool os::create_thread(Thread* thread, ThreadType thr_type,
869 size_t req_stack_size) {
870 assert(thread->osthread() == NULL, "caller responsible");
871
872 // Allocate the OSThread object
873 OSThread* osthread = new OSThread(NULL, NULL);
874 if (osthread == NULL) {
875 return false;
876 }
877
878 // set the correct thread state
879 osthread->set_thread_type(thr_type);
880
881 // Initial state is ALLOCATED but not INITIALIZED
882 osthread->set_state(ALLOCATED);
883
884 thread->set_osthread(osthread);
885
886 // init thread attributes
887 pthread_attr_t attr;
888 pthread_attr_init(&attr);
889 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
890
891 // Calculate stack size if it's not specified by caller.
892 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
893 // In glibc versions prior to 2.7 the guard size mechanism
894 // is not implemented properly. The posix standard requires adding
895 // the size of the guard pages to the stack size, instead Linux
896 // takes the space out of 'stacksize'. Thus we adapt the requested
897 // stack_size by the size of the guard pages to mimick proper
898 // behaviour. However, be careful not to end up with a size
899 // of zero due to overflow. Don't add the guard page in that case.
900 size_t guard_size = os::Linux::default_guard_size(thr_type);
901 // Configure glibc guard page. Must happen before calling
902 // get_static_tls_area_size(), which uses the guard_size.
903 pthread_attr_setguardsize(&attr, guard_size);
904
905 size_t stack_adjust_size = 0;
906 if (AdjustStackSizeForTLS) {
907 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
908 stack_adjust_size += get_static_tls_area_size(&attr);
909 } else {
910 stack_adjust_size += guard_size;
911 }
912
913 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
914 if (stack_size <= SIZE_MAX - stack_adjust_size) {
915 stack_size += stack_adjust_size;
916 }
917 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
918
919 int status = pthread_attr_setstacksize(&attr, stack_size);
920 assert_status(status == 0, status, "pthread_attr_setstacksize");
921
922 ThreadState state;
923
924 {
925 pthread_t tid;
926 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
927
928 char buf[64];
929 if (ret == 0) {
930 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
931 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
932 } else {
933 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
934 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
935 // Log some OS information which might explain why creating the thread failed.
936 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
937 LogStream st(Log(os, thread)::info());
938 os::Posix::print_rlimit_info(&st);
939 os::print_memory_info(&st);
940 os::Linux::print_proc_sys_info(&st);
941 os::Linux::print_container_info(&st);
942 }
943
944 pthread_attr_destroy(&attr);
945
946 if (ret != 0) {
947 // Need to clean up stuff we've allocated so far
948 thread->set_osthread(NULL);
949 delete osthread;
950 return false;
951 }
952
953 // Store pthread info into the OSThread
954 osthread->set_pthread_id(tid);
955
956 // Wait until child thread is either initialized or aborted
957 {
958 Monitor* sync_with_child = osthread->startThread_lock();
959 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
960 while ((state = osthread->get_state()) == ALLOCATED) {
961 sync_with_child->wait_without_safepoint_check();
962 }
963 }
964 }
965
966 // Aborted due to thread limit being reached
967 if (state == ZOMBIE) {
968 thread->set_osthread(NULL);
969 delete osthread;
970 return false;
971 }
972
973 // The thread is returned suspended (in state INITIALIZED),
974 // and is started higher up in the call chain
975 assert(state == INITIALIZED, "race condition");
976 return true;
977 }
978
979 /////////////////////////////////////////////////////////////////////////////
980 // attach existing thread
981
982 // bootstrap the main thread
983 bool os::create_main_thread(JavaThread* thread) {
984 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
985 return create_attached_thread(thread);
986 }
987
988 bool os::create_attached_thread(JavaThread* thread) {
989 #ifdef ASSERT
990 thread->verify_not_published();
991 #endif
992
993 // Allocate the OSThread object
994 OSThread* osthread = new OSThread(NULL, NULL);
995
996 if (osthread == NULL) {
997 return false;
998 }
999
1000 // Store pthread info into the OSThread
1001 osthread->set_thread_id(os::Linux::gettid());
1002 osthread->set_pthread_id(::pthread_self());
1003
1004 // initialize floating point control register
1005 os::Linux::init_thread_fpu_state();
1006
1007 // Initial thread state is RUNNABLE
1008 osthread->set_state(RUNNABLE);
1009
1010 thread->set_osthread(osthread);
1011
1012 if (UseNUMA) {
1013 int lgrp_id = os::numa_get_group_id();
1014 if (lgrp_id != -1) {
1015 thread->set_lgrp_id(lgrp_id);
1016 }
1017 }
1018
1019 if (os::is_primordial_thread()) {
1020 // If current thread is primordial thread, its stack is mapped on demand,
1021 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1022 // the entire stack region to avoid SEGV in stack banging.
1023 // It is also useful to get around the heap-stack-gap problem on SuSE
1024 // kernel (see 4821821 for details). We first expand stack to the top
1025 // of yellow zone, then enable stack yellow zone (order is significant,
1026 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1027 // is no gap between the last two virtual memory regions.
1028
1029 JavaThread *jt = (JavaThread *)thread;
1030 address addr = jt->stack_reserved_zone_base();
1031 assert(addr != NULL, "initialization problem?");
1032 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1033
1034 osthread->set_expanding_stack();
1035 os::Linux::manually_expand_stack(jt, addr);
1036 osthread->clear_expanding_stack();
1037 }
1038
1039 // initialize signal mask for this thread
1040 // and save the caller's signal mask
1041 os::Linux::hotspot_sigmask(thread);
1042
1043 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1044 os::current_thread_id(), (uintx) pthread_self());
1045
1046 return true;
1047 }
1048
1049 void os::pd_start_thread(Thread* thread) {
1050 OSThread * osthread = thread->osthread();
1051 assert(osthread->get_state() != INITIALIZED, "just checking");
1052 Monitor* sync_with_child = osthread->startThread_lock();
1053 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1054 sync_with_child->notify();
1055 }
1056
1057 // Free Linux resources related to the OSThread
1058 void os::free_thread(OSThread* osthread) {
1059 assert(osthread != NULL, "osthread not set");
1060
1061 // We are told to free resources of the argument thread,
1062 // but we can only really operate on the current thread.
1063 assert(Thread::current()->osthread() == osthread,
1064 "os::free_thread but not current thread");
1065
1066 #ifdef ASSERT
1067 sigset_t current;
1068 sigemptyset(¤t);
1069 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1070 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1071 #endif
1072
1073 // Restore caller's signal mask
1074 sigset_t sigmask = osthread->caller_sigmask();
1075 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1076
1077 delete osthread;
1078 }
1079
1080 //////////////////////////////////////////////////////////////////////////////
1081 // primordial thread
1082
1083 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1084 bool os::is_primordial_thread(void) {
1085 if (suppress_primordial_thread_resolution) {
1086 return false;
1087 }
1088 char dummy;
1089 // If called before init complete, thread stack bottom will be null.
1090 // Can be called if fatal error occurs before initialization.
1091 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1092 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1093 os::Linux::initial_thread_stack_size() != 0,
1094 "os::init did not locate primordial thread's stack region");
1095 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1096 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1097 os::Linux::initial_thread_stack_size()) {
1098 return true;
1099 } else {
1100 return false;
1101 }
1102 }
1103
1104 // Find the virtual memory area that contains addr
1105 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1106 FILE *fp = fopen("/proc/self/maps", "r");
1107 if (fp) {
1108 address low, high;
1109 while (!feof(fp)) {
1110 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1111 if (low <= addr && addr < high) {
1112 if (vma_low) *vma_low = low;
1113 if (vma_high) *vma_high = high;
1114 fclose(fp);
1115 return true;
1116 }
1117 }
1118 for (;;) {
1119 int ch = fgetc(fp);
1120 if (ch == EOF || ch == (int)'\n') break;
1121 }
1122 }
1123 fclose(fp);
1124 }
1125 return false;
1126 }
1127
1128 // Locate primordial thread stack. This special handling of primordial thread stack
1129 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1130 // bogus value for the primordial process thread. While the launcher has created
1131 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1132 // JNI invocation API from a primordial thread.
1133 void os::Linux::capture_initial_stack(size_t max_size) {
1134
1135 // max_size is either 0 (which means accept OS default for thread stacks) or
1136 // a user-specified value known to be at least the minimum needed. If we
1137 // are actually on the primordial thread we can make it appear that we have a
1138 // smaller max_size stack by inserting the guard pages at that location. But we
1139 // cannot do anything to emulate a larger stack than what has been provided by
1140 // the OS or threading library. In fact if we try to use a stack greater than
1141 // what is set by rlimit then we will crash the hosting process.
1142
1143 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1144 // If this is "unlimited" then it will be a huge value.
1145 struct rlimit rlim;
1146 getrlimit(RLIMIT_STACK, &rlim);
1147 size_t stack_size = rlim.rlim_cur;
1148
1149 // 6308388: a bug in ld.so will relocate its own .data section to the
1150 // lower end of primordial stack; reduce ulimit -s value a little bit
1151 // so we won't install guard page on ld.so's data section.
1152 // But ensure we don't underflow the stack size - allow 1 page spare
1153 if (stack_size >= (size_t)(3 * page_size())) {
1154 stack_size -= 2 * page_size();
1155 }
1156
1157 // Try to figure out where the stack base (top) is. This is harder.
1158 //
1159 // When an application is started, glibc saves the initial stack pointer in
1160 // a global variable "__libc_stack_end", which is then used by system
1161 // libraries. __libc_stack_end should be pretty close to stack top. The
1162 // variable is available since the very early days. However, because it is
1163 // a private interface, it could disappear in the future.
1164 //
1165 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1166 // to __libc_stack_end, it is very close to stack top, but isn't the real
1167 // stack top. Note that /proc may not exist if VM is running as a chroot
1168 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1169 // /proc/<pid>/stat could change in the future (though unlikely).
1170 //
1171 // We try __libc_stack_end first. If that doesn't work, look for
1172 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1173 // as a hint, which should work well in most cases.
1174
1175 uintptr_t stack_start;
1176
1177 // try __libc_stack_end first
1178 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1179 if (p && *p) {
1180 stack_start = *p;
1181 } else {
1182 // see if we can get the start_stack field from /proc/self/stat
1183 FILE *fp;
1184 int pid;
1185 char state;
1186 int ppid;
1187 int pgrp;
1188 int session;
1189 int nr;
1190 int tpgrp;
1191 unsigned long flags;
1192 unsigned long minflt;
1193 unsigned long cminflt;
1194 unsigned long majflt;
1195 unsigned long cmajflt;
1196 unsigned long utime;
1197 unsigned long stime;
1198 long cutime;
1199 long cstime;
1200 long prio;
1201 long nice;
1202 long junk;
1203 long it_real;
1204 uintptr_t start;
1205 uintptr_t vsize;
1206 intptr_t rss;
1207 uintptr_t rsslim;
1208 uintptr_t scodes;
1209 uintptr_t ecode;
1210 int i;
1211
1212 // Figure what the primordial thread stack base is. Code is inspired
1213 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1214 // followed by command name surrounded by parentheses, state, etc.
1215 char stat[2048];
1216 int statlen;
1217
1218 fp = fopen("/proc/self/stat", "r");
1219 if (fp) {
1220 statlen = fread(stat, 1, 2047, fp);
1221 stat[statlen] = '\0';
1222 fclose(fp);
1223
1224 // Skip pid and the command string. Note that we could be dealing with
1225 // weird command names, e.g. user could decide to rename java launcher
1226 // to "java 1.4.2 :)", then the stat file would look like
1227 // 1234 (java 1.4.2 :)) R ... ...
1228 // We don't really need to know the command string, just find the last
1229 // occurrence of ")" and then start parsing from there. See bug 4726580.
1230 char * s = strrchr(stat, ')');
1231
1232 i = 0;
1233 if (s) {
1234 // Skip blank chars
1235 do { s++; } while (s && isspace(*s));
1236
1237 #define _UFM UINTX_FORMAT
1238 #define _DFM INTX_FORMAT
1239
1240 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1241 // 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
1242 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,
1243 &state, // 3 %c
1244 &ppid, // 4 %d
1245 &pgrp, // 5 %d
1246 &session, // 6 %d
1247 &nr, // 7 %d
1248 &tpgrp, // 8 %d
1249 &flags, // 9 %lu
1250 &minflt, // 10 %lu
1251 &cminflt, // 11 %lu
1252 &majflt, // 12 %lu
1253 &cmajflt, // 13 %lu
1254 &utime, // 14 %lu
1255 &stime, // 15 %lu
1256 &cutime, // 16 %ld
1257 &cstime, // 17 %ld
1258 &prio, // 18 %ld
1259 &nice, // 19 %ld
1260 &junk, // 20 %ld
1261 &it_real, // 21 %ld
1262 &start, // 22 UINTX_FORMAT
1263 &vsize, // 23 UINTX_FORMAT
1264 &rss, // 24 INTX_FORMAT
1265 &rsslim, // 25 UINTX_FORMAT
1266 &scodes, // 26 UINTX_FORMAT
1267 &ecode, // 27 UINTX_FORMAT
1268 &stack_start); // 28 UINTX_FORMAT
1269 }
1270
1271 #undef _UFM
1272 #undef _DFM
1273
1274 if (i != 28 - 2) {
1275 assert(false, "Bad conversion from /proc/self/stat");
1276 // product mode - assume we are the primordial thread, good luck in the
1277 // embedded case.
1278 warning("Can't detect primordial thread stack location - bad conversion");
1279 stack_start = (uintptr_t) &rlim;
1280 }
1281 } else {
1282 // For some reason we can't open /proc/self/stat (for example, running on
1283 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1284 // most cases, so don't abort:
1285 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1286 stack_start = (uintptr_t) &rlim;
1287 }
1288 }
1289
1290 // Now we have a pointer (stack_start) very close to the stack top, the
1291 // next thing to do is to figure out the exact location of stack top. We
1292 // can find out the virtual memory area that contains stack_start by
1293 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1294 // and its upper limit is the real stack top. (again, this would fail if
1295 // running inside chroot, because /proc may not exist.)
1296
1297 uintptr_t stack_top;
1298 address low, high;
1299 if (find_vma((address)stack_start, &low, &high)) {
1300 // success, "high" is the true stack top. (ignore "low", because initial
1301 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1302 stack_top = (uintptr_t)high;
1303 } else {
1304 // failed, likely because /proc/self/maps does not exist
1305 warning("Can't detect primordial thread stack location - find_vma failed");
1306 // best effort: stack_start is normally within a few pages below the real
1307 // stack top, use it as stack top, and reduce stack size so we won't put
1308 // guard page outside stack.
1309 stack_top = stack_start;
1310 stack_size -= 16 * page_size();
1311 }
1312
1313 // stack_top could be partially down the page so align it
1314 stack_top = align_up(stack_top, page_size());
1315
1316 // Allowed stack value is minimum of max_size and what we derived from rlimit
1317 if (max_size > 0) {
1318 _initial_thread_stack_size = MIN2(max_size, stack_size);
1319 } else {
1320 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1321 // clamp it at 8MB as we do on Solaris
1322 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1323 }
1324 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1325 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1326
1327 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1328
1329 if (log_is_enabled(Info, os, thread)) {
1330 // See if we seem to be on primordial process thread
1331 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1332 uintptr_t(&rlim) < stack_top;
1333
1334 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1335 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1336 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1337 stack_top, intptr_t(_initial_thread_stack_bottom));
1338 }
1339 }
1340
1341 ////////////////////////////////////////////////////////////////////////////////
1342 // time support
1343
1344 #ifndef SUPPORTS_CLOCK_MONOTONIC
1345 #error "Build platform doesn't support clock_gettime and related functionality"
1346 #endif
1347
1348 // Time since start-up in seconds to a fine granularity.
1349 // Used by VMSelfDestructTimer and the MemProfiler.
1350 double os::elapsedTime() {
1351
1352 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1353 }
1354
1355 jlong os::elapsed_counter() {
1356 return javaTimeNanos() - initial_time_count;
1357 }
1358
1359 jlong os::elapsed_frequency() {
1360 return NANOSECS_PER_SEC; // nanosecond resolution
1361 }
1362
1363 bool os::supports_vtime() { return true; }
1364
1365 double os::elapsedVTime() {
1366 struct rusage usage;
1367 int retval = getrusage(RUSAGE_THREAD, &usage);
1368 if (retval == 0) {
1369 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);
1370 } else {
1371 // better than nothing, but not much
1372 return elapsedTime();
1373 }
1374 }
1375
1376 jlong os::javaTimeMillis() {
1377 timeval time;
1378 int status = gettimeofday(&time, NULL);
1379 assert(status != -1, "linux error");
1380 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1381 }
1382
1383 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1384 timeval time;
1385 int status = gettimeofday(&time, NULL);
1386 assert(status != -1, "linux error");
1387 seconds = jlong(time.tv_sec);
1388 nanos = jlong(time.tv_usec) * 1000;
1389 }
1390
1391 void os::Linux::fast_thread_clock_init() {
1392 if (!UseLinuxPosixThreadCPUClocks) {
1393 return;
1394 }
1395 clockid_t clockid;
1396 struct timespec tp;
1397 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1398 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1399
1400 // Switch to using fast clocks for thread cpu time if
1401 // the clock_getres() returns 0 error code.
1402 // Note, that some kernels may support the current thread
1403 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1404 // returned by the pthread_getcpuclockid().
1405 // If the fast Posix clocks are supported then the clock_getres()
1406 // must return at least tp.tv_sec == 0 which means a resolution
1407 // better than 1 sec. This is extra check for reliability.
1408
1409 if (pthread_getcpuclockid_func &&
1410 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1411 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1412 _supports_fast_thread_cpu_time = true;
1413 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1414 }
1415 }
1416
1417 jlong os::javaTimeNanos() {
1418 if (os::supports_monotonic_clock()) {
1419 struct timespec tp;
1420 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1421 assert(status == 0, "gettime error");
1422 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1423 return result;
1424 } else {
1425 timeval time;
1426 int status = gettimeofday(&time, NULL);
1427 assert(status != -1, "linux error");
1428 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1429 return 1000 * usecs;
1430 }
1431 }
1432
1433 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1434 if (os::supports_monotonic_clock()) {
1435 info_ptr->max_value = ALL_64_BITS;
1436
1437 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1438 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1439 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1440 } else {
1441 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1442 info_ptr->max_value = ALL_64_BITS;
1443
1444 // gettimeofday is a real time clock so it skips
1445 info_ptr->may_skip_backward = true;
1446 info_ptr->may_skip_forward = true;
1447 }
1448
1449 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1450 }
1451
1452 // Return the real, user, and system times in seconds from an
1453 // arbitrary fixed point in the past.
1454 bool os::getTimesSecs(double* process_real_time,
1455 double* process_user_time,
1456 double* process_system_time) {
1457 struct tms ticks;
1458 clock_t real_ticks = times(&ticks);
1459
1460 if (real_ticks == (clock_t) (-1)) {
1461 return false;
1462 } else {
1463 double ticks_per_second = (double) clock_tics_per_sec;
1464 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1465 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1466 *process_real_time = ((double) real_ticks) / ticks_per_second;
1467
1468 return true;
1469 }
1470 }
1471
1472
1473 char * os::local_time_string(char *buf, size_t buflen) {
1474 struct tm t;
1475 time_t long_time;
1476 time(&long_time);
1477 localtime_r(&long_time, &t);
1478 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1479 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1480 t.tm_hour, t.tm_min, t.tm_sec);
1481 return buf;
1482 }
1483
1484 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1485 return localtime_r(clock, res);
1486 }
1487
1488 ////////////////////////////////////////////////////////////////////////////////
1489 // runtime exit support
1490
1491 // Note: os::shutdown() might be called very early during initialization, or
1492 // called from signal handler. Before adding something to os::shutdown(), make
1493 // sure it is async-safe and can handle partially initialized VM.
1494 void os::shutdown() {
1495
1496 // allow PerfMemory to attempt cleanup of any persistent resources
1497 perfMemory_exit();
1498
1499 // needs to remove object in file system
1500 AttachListener::abort();
1501
1502 // flush buffered output, finish log files
1503 ostream_abort();
1504
1505 // Check for abort hook
1506 abort_hook_t abort_hook = Arguments::abort_hook();
1507 if (abort_hook != NULL) {
1508 abort_hook();
1509 }
1510
1511 }
1512
1513 // Note: os::abort() might be called very early during initialization, or
1514 // called from signal handler. Before adding something to os::abort(), make
1515 // sure it is async-safe and can handle partially initialized VM.
1516 void os::abort(bool dump_core, void* siginfo, const void* context) {
1517 os::shutdown();
1518 if (dump_core) {
1519 if (DumpPrivateMappingsInCore) {
1520 ClassLoader::close_jrt_image();
1521 }
1522 #ifndef PRODUCT
1523 fdStream out(defaultStream::output_fd());
1524 out.print_raw("Current thread is ");
1525 char buf[16];
1526 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1527 out.print_raw_cr(buf);
1528 out.print_raw_cr("Dumping core ...");
1529 #endif
1530 ::abort(); // dump core
1531 }
1532
1533 ::exit(1);
1534 }
1535
1536 // Die immediately, no exit hook, no abort hook, no cleanup.
1537 // Dump a core file, if possible, for debugging.
1538 void os::die() {
1539 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1540 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1541 // and don't take the time to generate a core file.
1542 os::signal_raise(SIGKILL);
1543 } else {
1544 ::abort();
1545 }
1546 }
1547
1548 // thread_id is kernel thread id (similar to Solaris LWP id)
1549 intx os::current_thread_id() { return os::Linux::gettid(); }
1550 int os::current_process_id() {
1551 return ::getpid();
1552 }
1553
1554 // DLL functions
1555
1556 const char* os::dll_file_extension() { return ".so"; }
1557
1558 // This must be hard coded because it's the system's temporary
1559 // directory not the java application's temp directory, ala java.io.tmpdir.
1560 const char* os::get_temp_directory() { return "/tmp"; }
1561
1562 static bool file_exists(const char* filename) {
1563 struct stat statbuf;
1564 if (filename == NULL || strlen(filename) == 0) {
1565 return false;
1566 }
1567 return os::stat(filename, &statbuf) == 0;
1568 }
1569
1570 // check if addr is inside libjvm.so
1571 bool os::address_is_in_vm(address addr) {
1572 static address libjvm_base_addr;
1573 Dl_info dlinfo;
1574
1575 if (libjvm_base_addr == NULL) {
1576 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1577 libjvm_base_addr = (address)dlinfo.dli_fbase;
1578 }
1579 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1580 }
1581
1582 if (dladdr((void *)addr, &dlinfo) != 0) {
1583 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1584 }
1585
1586 return false;
1587 }
1588
1589 bool os::dll_address_to_function_name(address addr, char *buf,
1590 int buflen, int *offset,
1591 bool demangle) {
1592 // buf is not optional, but offset is optional
1593 assert(buf != NULL, "sanity check");
1594
1595 Dl_info dlinfo;
1596
1597 if (dladdr((void*)addr, &dlinfo) != 0) {
1598 // see if we have a matching symbol
1599 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1600 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1601 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1602 }
1603 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1604 return true;
1605 }
1606 // no matching symbol so try for just file info
1607 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1608 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1609 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1610 return true;
1611 }
1612 }
1613 }
1614
1615 buf[0] = '\0';
1616 if (offset != NULL) *offset = -1;
1617 return false;
1618 }
1619
1620 struct _address_to_library_name {
1621 address addr; // input : memory address
1622 size_t buflen; // size of fname
1623 char* fname; // output: library name
1624 address base; // library base addr
1625 };
1626
1627 static int address_to_library_name_callback(struct dl_phdr_info *info,
1628 size_t size, void *data) {
1629 int i;
1630 bool found = false;
1631 address libbase = NULL;
1632 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1633
1634 // iterate through all loadable segments
1635 for (i = 0; i < info->dlpi_phnum; i++) {
1636 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1637 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1638 // base address of a library is the lowest address of its loaded
1639 // segments.
1640 if (libbase == NULL || libbase > segbase) {
1641 libbase = segbase;
1642 }
1643 // see if 'addr' is within current segment
1644 if (segbase <= d->addr &&
1645 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1646 found = true;
1647 }
1648 }
1649 }
1650
1651 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1652 // so dll_address_to_library_name() can fall through to use dladdr() which
1653 // can figure out executable name from argv[0].
1654 if (found && info->dlpi_name && info->dlpi_name[0]) {
1655 d->base = libbase;
1656 if (d->fname) {
1657 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1658 }
1659 return 1;
1660 }
1661 return 0;
1662 }
1663
1664 bool os::dll_address_to_library_name(address addr, char* buf,
1665 int buflen, int* offset) {
1666 // buf is not optional, but offset is optional
1667 assert(buf != NULL, "sanity check");
1668
1669 Dl_info dlinfo;
1670 struct _address_to_library_name data;
1671
1672 // There is a bug in old glibc dladdr() implementation that it could resolve
1673 // to wrong library name if the .so file has a base address != NULL. Here
1674 // we iterate through the program headers of all loaded libraries to find
1675 // out which library 'addr' really belongs to. This workaround can be
1676 // removed once the minimum requirement for glibc is moved to 2.3.x.
1677 data.addr = addr;
1678 data.fname = buf;
1679 data.buflen = buflen;
1680 data.base = NULL;
1681 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1682
1683 if (rslt) {
1684 // buf already contains library name
1685 if (offset) *offset = addr - data.base;
1686 return true;
1687 }
1688 if (dladdr((void*)addr, &dlinfo) != 0) {
1689 if (dlinfo.dli_fname != NULL) {
1690 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1691 }
1692 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1693 *offset = addr - (address)dlinfo.dli_fbase;
1694 }
1695 return true;
1696 }
1697
1698 buf[0] = '\0';
1699 if (offset) *offset = -1;
1700 return false;
1701 }
1702
1703 // Loads .dll/.so and
1704 // in case of error it checks if .dll/.so was built for the
1705 // same architecture as Hotspot is running on
1706
1707
1708 // Remember the stack's state. The Linux dynamic linker will change
1709 // the stack to 'executable' at most once, so we must safepoint only once.
1710 bool os::Linux::_stack_is_executable = false;
1711
1712 // VM operation that loads a library. This is necessary if stack protection
1713 // of the Java stacks can be lost during loading the library. If we
1714 // do not stop the Java threads, they can stack overflow before the stacks
1715 // are protected again.
1716 class VM_LinuxDllLoad: public VM_Operation {
1717 private:
1718 const char *_filename;
1719 char *_ebuf;
1720 int _ebuflen;
1721 void *_lib;
1722 public:
1723 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1724 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1725 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1726 void doit() {
1727 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1728 os::Linux::_stack_is_executable = true;
1729 }
1730 void* loaded_library() { return _lib; }
1731 };
1732
1733 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1734 void * result = NULL;
1735 bool load_attempted = false;
1736
1737 log_info(os)("attempting shared library load of %s", filename);
1738
1739 // Check whether the library to load might change execution rights
1740 // of the stack. If they are changed, the protection of the stack
1741 // guard pages will be lost. We need a safepoint to fix this.
1742 //
1743 // See Linux man page execstack(8) for more info.
1744 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1745 if (!ElfFile::specifies_noexecstack(filename)) {
1746 if (!is_init_completed()) {
1747 os::Linux::_stack_is_executable = true;
1748 // This is OK - No Java threads have been created yet, and hence no
1749 // stack guard pages to fix.
1750 //
1751 // Dynamic loader will make all stacks executable after
1752 // this function returns, and will not do that again.
1753 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1754 } else {
1755 warning("You have loaded library %s which might have disabled stack guard. "
1756 "The VM will try to fix the stack guard now.\n"
1757 "It's highly recommended that you fix the library with "
1758 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1759 filename);
1760
1761 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1762 JavaThread *jt = JavaThread::current();
1763 if (jt->thread_state() != _thread_in_native) {
1764 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1765 // that requires ExecStack. Cannot enter safe point. Let's give up.
1766 warning("Unable to fix stack guard. Giving up.");
1767 } else {
1768 if (!LoadExecStackDllInVMThread) {
1769 // This is for the case where the DLL has an static
1770 // constructor function that executes JNI code. We cannot
1771 // load such DLLs in the VMThread.
1772 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1773 }
1774
1775 ThreadInVMfromNative tiv(jt);
1776 debug_only(VMNativeEntryWrapper vew;)
1777
1778 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1779 VMThread::execute(&op);
1780 if (LoadExecStackDllInVMThread) {
1781 result = op.loaded_library();
1782 }
1783 load_attempted = true;
1784 }
1785 }
1786 }
1787 }
1788
1789 if (!load_attempted) {
1790 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1791 }
1792
1793 if (result != NULL) {
1794 // Successful loading
1795 return result;
1796 }
1797
1798 Elf32_Ehdr elf_head;
1799 int diag_msg_max_length=ebuflen-strlen(ebuf);
1800 char* diag_msg_buf=ebuf+strlen(ebuf);
1801
1802 if (diag_msg_max_length==0) {
1803 // No more space in ebuf for additional diagnostics message
1804 return NULL;
1805 }
1806
1807
1808 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1809
1810 if (file_descriptor < 0) {
1811 // Can't open library, report dlerror() message
1812 return NULL;
1813 }
1814
1815 bool failed_to_read_elf_head=
1816 (sizeof(elf_head)!=
1817 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1818
1819 ::close(file_descriptor);
1820 if (failed_to_read_elf_head) {
1821 // file i/o error - report dlerror() msg
1822 return NULL;
1823 }
1824
1825 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1826 // handle invalid/out of range endianness values
1827 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1828 return NULL;
1829 }
1830
1831 #if defined(VM_LITTLE_ENDIAN)
1832 // VM is LE, shared object BE
1833 elf_head.e_machine = be16toh(elf_head.e_machine);
1834 #else
1835 // VM is BE, shared object LE
1836 elf_head.e_machine = le16toh(elf_head.e_machine);
1837 #endif
1838 }
1839
1840 typedef struct {
1841 Elf32_Half code; // Actual value as defined in elf.h
1842 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1843 unsigned char elf_class; // 32 or 64 bit
1844 unsigned char endianness; // MSB or LSB
1845 char* name; // String representation
1846 } arch_t;
1847
1848 #ifndef EM_486
1849 #define EM_486 6 /* Intel 80486 */
1850 #endif
1851 #ifndef EM_AARCH64
1852 #define EM_AARCH64 183 /* ARM AARCH64 */
1853 #endif
1854 #ifndef EM_RISCV
1855 #define EM_RISCV 243 /* RISC-V */
1856 #endif
1857
1858 static const arch_t arch_array[]={
1859 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1860 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1861 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1862 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1863 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1864 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1865 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1866 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1867 #if defined(VM_LITTLE_ENDIAN)
1868 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1869 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1870 #else
1871 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1872 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1873 #endif
1874 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1875 // we only support 64 bit z architecture
1876 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1877 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1878 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1879 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1880 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1881 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1882 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1883 {EM_RISCV, EM_RISCV, ELFCLASS64, ELFDATA2LSB, (char*)"RISC-V"},
1884 };
1885
1886 #if (defined IA32)
1887 static Elf32_Half running_arch_code=EM_386;
1888 #elif (defined AMD64) || (defined X32)
1889 static Elf32_Half running_arch_code=EM_X86_64;
1890 #elif (defined IA64)
1891 static Elf32_Half running_arch_code=EM_IA_64;
1892 #elif (defined __sparc) && (defined _LP64)
1893 static Elf32_Half running_arch_code=EM_SPARCV9;
1894 #elif (defined __sparc) && (!defined _LP64)
1895 static Elf32_Half running_arch_code=EM_SPARC;
1896 #elif (defined __powerpc64__)
1897 static Elf32_Half running_arch_code=EM_PPC64;
1898 #elif (defined __powerpc__)
1899 static Elf32_Half running_arch_code=EM_PPC;
1900 #elif (defined AARCH64)
1901 static Elf32_Half running_arch_code=EM_AARCH64;
1902 #elif (defined ARM)
1903 static Elf32_Half running_arch_code=EM_ARM;
1904 #elif (defined S390)
1905 static Elf32_Half running_arch_code=EM_S390;
1906 #elif (defined ALPHA)
1907 static Elf32_Half running_arch_code=EM_ALPHA;
1908 #elif (defined MIPSEL)
1909 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1910 #elif (defined PARISC)
1911 static Elf32_Half running_arch_code=EM_PARISC;
1912 #elif (defined MIPS)
1913 static Elf32_Half running_arch_code=EM_MIPS;
1914 #elif (defined M68K)
1915 static Elf32_Half running_arch_code=EM_68K;
1916 #elif (defined SH)
1917 static Elf32_Half running_arch_code=EM_SH;
1918 #elif (defined RISCV)
1919 static Elf32_Half running_arch_code=EM_RISCV;
1920 #else
1921 #error Method os::dll_load requires that one of following is defined:\
1922 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, RISCV, S390, SH, __sparc
1923 #endif
1924
1925 // Identify compatibility class for VM's architecture and library's architecture
1926 // Obtain string descriptions for architectures
1927
1928 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1929 int running_arch_index=-1;
1930
1931 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1932 if (running_arch_code == arch_array[i].code) {
1933 running_arch_index = i;
1934 }
1935 if (lib_arch.code == arch_array[i].code) {
1936 lib_arch.compat_class = arch_array[i].compat_class;
1937 lib_arch.name = arch_array[i].name;
1938 }
1939 }
1940
1941 assert(running_arch_index != -1,
1942 "Didn't find running architecture code (running_arch_code) in arch_array");
1943 if (running_arch_index == -1) {
1944 // Even though running architecture detection failed
1945 // we may still continue with reporting dlerror() message
1946 return NULL;
1947 }
1948
1949 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1950 if (lib_arch.name != NULL) {
1951 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1952 " (Possible cause: can't load %s .so on a %s platform)",
1953 lib_arch.name, arch_array[running_arch_index].name);
1954 } else {
1955 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1956 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1957 lib_arch.code, arch_array[running_arch_index].name);
1958 }
1959 return NULL;
1960 }
1961
1962 if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1963 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1964 return NULL;
1965 }
1966
1967 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1968 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1969 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1970 return NULL;
1971 }
1972
1973 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1974 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1975 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1976 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1977 return NULL;
1978 }
1979
1980 return NULL;
1981 }
1982
1983 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1984 int ebuflen) {
1985 void * result = ::dlopen(filename, RTLD_LAZY);
1986 if (result == NULL) {
1987 const char* error_report = ::dlerror();
1988 if (error_report == NULL) {
1989 error_report = "dlerror returned no error description";
1990 }
1991 if (ebuf != NULL && ebuflen > 0) {
1992 ::strncpy(ebuf, error_report, ebuflen-1);
1993 ebuf[ebuflen-1]='\0';
1994 }
1995 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1996 log_info(os)("shared library load of %s failed, %s", filename, error_report);
1997 } else {
1998 Events::log(NULL, "Loaded shared library %s", filename);
1999 log_info(os)("shared library load of %s was successful", filename);
2000 }
2001 return result;
2002 }
2003
2004 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2005 int ebuflen) {
2006 void * result = NULL;
2007 if (LoadExecStackDllInVMThread) {
2008 result = dlopen_helper(filename, ebuf, ebuflen);
2009 }
2010
2011 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2012 // library that requires an executable stack, or which does not have this
2013 // stack attribute set, dlopen changes the stack attribute to executable. The
2014 // read protection of the guard pages gets lost.
2015 //
2016 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2017 // may have been queued at the same time.
2018
2019 if (!_stack_is_executable) {
2020 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2021 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2022 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
2023 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2024 warning("Attempt to reguard stack yellow zone failed.");
2025 }
2026 }
2027 }
2028 }
2029
2030 return result;
2031 }
2032
2033 void* os::dll_lookup(void* handle, const char* name) {
2034 void* res = dlsym(handle, name);
2035 return res;
2036 }
2037
2038 void* os::get_default_process_handle() {
2039 return (void*)::dlopen(NULL, RTLD_LAZY);
2040 }
2041
2042 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2043 int fd = ::open(filename, O_RDONLY);
2044 if (fd == -1) {
2045 return false;
2046 }
2047
2048 if (hdr != NULL) {
2049 st->print_cr("%s", hdr);
2050 }
2051
2052 char buf[33];
2053 int bytes;
2054 buf[32] = '\0';
2055 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2056 st->print_raw(buf, bytes);
2057 }
2058
2059 ::close(fd);
2060
2061 return true;
2062 }
2063
2064 void os::print_dll_info(outputStream *st) {
2065 st->print_cr("Dynamic libraries:");
2066
2067 char fname[32];
2068 pid_t pid = os::Linux::gettid();
2069
2070 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2071
2072 if (!_print_ascii_file(fname, st)) {
2073 st->print("Can not get library information for pid = %d\n", pid);
2074 }
2075 }
2076
2077 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2078 FILE *procmapsFile = NULL;
2079
2080 // Open the procfs maps file for the current process
2081 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2082 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2083 char line[PATH_MAX + 100];
2084
2085 // Read line by line from 'file'
2086 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2087 u8 base, top, inode;
2088 char name[sizeof(line)];
2089
2090 // Parse fields from line, discard perms, offset and device
2091 int matches = sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %*s %*s %*s " INT64_FORMAT " %s",
2092 &base, &top, &inode, name);
2093 // the last entry 'name' is empty for some entries, so we might have 3 matches instead of 4 for some lines
2094 if (matches < 3) continue;
2095 if (matches == 3) name[0] = '\0';
2096
2097 // Filter by inode 0 so that we only get file system mapped files.
2098 if (inode != 0) {
2099
2100 // Call callback with the fields of interest
2101 if(callback(name, (address)base, (address)top, param)) {
2102 // Oops abort, callback aborted
2103 fclose(procmapsFile);
2104 return 1;
2105 }
2106 }
2107 }
2108 fclose(procmapsFile);
2109 }
2110 return 0;
2111 }
2112
2113 void os::print_os_info_brief(outputStream* st) {
2114 os::Linux::print_distro_info(st);
2115
2116 os::Posix::print_uname_info(st);
2117
2118 os::Linux::print_libversion_info(st);
2119
2120 }
2121
2122 void os::print_os_info(outputStream* st) {
2123 st->print("OS:");
2124
2125 os::Linux::print_distro_info(st);
2126
2127 os::Posix::print_uname_info(st);
2128
2129 os::Linux::print_uptime_info(st);
2130
2131 // Print warning if unsafe chroot environment detected
2132 if (unsafe_chroot_detected) {
2133 st->print("WARNING!! ");
2134 st->print_cr("%s", unstable_chroot_error);
2135 }
2136
2137 os::Linux::print_libversion_info(st);
2138
2139 os::Posix::print_rlimit_info(st);
2140
2141 os::Posix::print_load_average(st);
2142
2143 os::Linux::print_full_memory_info(st);
2144
2145 os::Linux::print_proc_sys_info(st);
2146
2147 os::Linux::print_ld_preload_file(st);
2148
2149 os::Linux::print_container_info(st);
2150
2151 VM_Version::print_platform_virtualization_info(st);
2152
2153 os::Linux::print_steal_info(st);
2154 }
2155
2156 // Try to identify popular distros.
2157 // Most Linux distributions have a /etc/XXX-release file, which contains
2158 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2159 // file that also contains the OS version string. Some have more than one
2160 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2161 // /etc/redhat-release.), so the order is important.
2162 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2163 // their own specific XXX-release file as well as a redhat-release file.
2164 // Because of this the XXX-release file needs to be searched for before the
2165 // redhat-release file.
2166 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2167 // search for redhat-release / SuSE-release needs to be before lsb-release.
2168 // Since the lsb-release file is the new standard it needs to be searched
2169 // before the older style release files.
2170 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2171 // next to last resort. The os-release file is a new standard that contains
2172 // distribution information and the system-release file seems to be an old
2173 // standard that has been replaced by the lsb-release and os-release files.
2174 // Searching for the debian_version file is the last resort. It contains
2175 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2176 // "Debian " is printed before the contents of the debian_version file.
2177
2178 const char* distro_files[] = {
2179 "/etc/oracle-release",
2180 "/etc/mandriva-release",
2181 "/etc/mandrake-release",
2182 "/etc/sun-release",
2183 "/etc/redhat-release",
2184 "/etc/SuSE-release",
2185 "/etc/lsb-release",
2186 "/etc/turbolinux-release",
2187 "/etc/gentoo-release",
2188 "/etc/ltib-release",
2189 "/etc/angstrom-version",
2190 "/etc/system-release",
2191 "/etc/os-release",
2192 NULL };
2193
2194 void os::Linux::print_distro_info(outputStream* st) {
2195 for (int i = 0;; i++) {
2196 const char* file = distro_files[i];
2197 if (file == NULL) {
2198 break; // done
2199 }
2200 // If file prints, we found it.
2201 if (_print_ascii_file(file, st)) {
2202 return;
2203 }
2204 }
2205
2206 if (file_exists("/etc/debian_version")) {
2207 st->print("Debian ");
2208 _print_ascii_file("/etc/debian_version", st);
2209 } else {
2210 st->print("Linux");
2211 }
2212 st->cr();
2213 }
2214
2215 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2216 char buf[256];
2217 while (fgets(buf, sizeof(buf), fp)) {
2218 // Edit out extra stuff in expected format
2219 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2220 char* ptr = strstr(buf, "\""); // the name is in quotes
2221 if (ptr != NULL) {
2222 ptr++; // go beyond first quote
2223 char* nl = strchr(ptr, '\"');
2224 if (nl != NULL) *nl = '\0';
2225 strncpy(distro, ptr, length);
2226 } else {
2227 ptr = strstr(buf, "=");
2228 ptr++; // go beyond equals then
2229 char* nl = strchr(ptr, '\n');
2230 if (nl != NULL) *nl = '\0';
2231 strncpy(distro, ptr, length);
2232 }
2233 return;
2234 } else if (get_first_line) {
2235 char* nl = strchr(buf, '\n');
2236 if (nl != NULL) *nl = '\0';
2237 strncpy(distro, buf, length);
2238 return;
2239 }
2240 }
2241 // print last line and close
2242 char* nl = strchr(buf, '\n');
2243 if (nl != NULL) *nl = '\0';
2244 strncpy(distro, buf, length);
2245 }
2246
2247 static void parse_os_info(char* distro, size_t length, const char* file) {
2248 FILE* fp = fopen(file, "r");
2249 if (fp != NULL) {
2250 // if suse format, print out first line
2251 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2252 parse_os_info_helper(fp, distro, length, get_first_line);
2253 fclose(fp);
2254 }
2255 }
2256
2257 void os::get_summary_os_info(char* buf, size_t buflen) {
2258 for (int i = 0;; i++) {
2259 const char* file = distro_files[i];
2260 if (file == NULL) {
2261 break; // ran out of distro_files
2262 }
2263 if (file_exists(file)) {
2264 parse_os_info(buf, buflen, file);
2265 return;
2266 }
2267 }
2268 // special case for debian
2269 if (file_exists("/etc/debian_version")) {
2270 strncpy(buf, "Debian ", buflen);
2271 if (buflen > 7) {
2272 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2273 }
2274 } else {
2275 strncpy(buf, "Linux", buflen);
2276 }
2277 }
2278
2279 void os::Linux::print_libversion_info(outputStream* st) {
2280 // libc, pthread
2281 st->print("libc:");
2282 st->print("%s ", os::Linux::glibc_version());
2283 st->print("%s ", os::Linux::libpthread_version());
2284 st->cr();
2285 }
2286
2287 void os::Linux::print_proc_sys_info(outputStream* st) {
2288 st->cr();
2289 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2290 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2291 st->cr();
2292 st->cr();
2293
2294 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2295 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2296 st->cr();
2297 st->cr();
2298
2299 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2300 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2301 st->cr();
2302 st->cr();
2303 }
2304
2305 void os::Linux::print_full_memory_info(outputStream* st) {
2306 st->print("\n/proc/meminfo:\n");
2307 _print_ascii_file("/proc/meminfo", st);
2308 st->cr();
2309
2310 // some information regarding THPs; for details see
2311 // https://www.kernel.org/doc/Documentation/vm/transhuge.txt
2312 st->print_cr("/sys/kernel/mm/transparent_hugepage/enabled:");
2313 if (!_print_ascii_file("/sys/kernel/mm/transparent_hugepage/enabled", st)) {
2314 st->print_cr(" <Not Available>");
2315 }
2316 st->cr();
2317 st->print_cr("/sys/kernel/mm/transparent_hugepage/defrag (defrag/compaction efforts parameter):");
2318 if (!_print_ascii_file("/sys/kernel/mm/transparent_hugepage/defrag", st)) {
2319 st->print_cr(" <Not Available>");
2320 }
2321 st->cr();
2322 }
2323
2324 void os::Linux::print_ld_preload_file(outputStream* st) {
2325 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2326 st->cr();
2327 }
2328
2329 void os::Linux::print_uptime_info(outputStream* st) {
2330 struct sysinfo sinfo;
2331 int ret = sysinfo(&sinfo);
2332 if (ret == 0) {
2333 os::print_dhm(st, "OS uptime:", (long) sinfo.uptime);
2334 }
2335 }
2336
2337
2338 void os::Linux::print_container_info(outputStream* st) {
2339 if (!OSContainer::is_containerized()) {
2340 return;
2341 }
2342
2343 st->print("container (cgroup) information:\n");
2344
2345 const char *p_ct = OSContainer::container_type();
2346 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2347
2348 char *p = OSContainer::cpu_cpuset_cpus();
2349 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2350 free(p);
2351
2352 p = OSContainer::cpu_cpuset_memory_nodes();
2353 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2354 free(p);
2355
2356 int i = OSContainer::active_processor_count();
2357 st->print("active_processor_count: ");
2358 if (i > 0) {
2359 st->print("%d\n", i);
2360 } else {
2361 st->print("not supported\n");
2362 }
2363
2364 i = OSContainer::cpu_quota();
2365 st->print("cpu_quota: ");
2366 if (i > 0) {
2367 st->print("%d\n", i);
2368 } else {
2369 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2370 }
2371
2372 i = OSContainer::cpu_period();
2373 st->print("cpu_period: ");
2374 if (i > 0) {
2375 st->print("%d\n", i);
2376 } else {
2377 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2378 }
2379
2380 i = OSContainer::cpu_shares();
2381 st->print("cpu_shares: ");
2382 if (i > 0) {
2383 st->print("%d\n", i);
2384 } else {
2385 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2386 }
2387
2388 jlong j = OSContainer::memory_limit_in_bytes();
2389 st->print("memory_limit_in_bytes: ");
2390 if (j > 0) {
2391 st->print(JLONG_FORMAT "\n", j);
2392 } else {
2393 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2394 }
2395
2396 j = OSContainer::memory_and_swap_limit_in_bytes();
2397 st->print("memory_and_swap_limit_in_bytes: ");
2398 if (j > 0) {
2399 st->print(JLONG_FORMAT "\n", j);
2400 } else {
2401 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2402 }
2403
2404 j = OSContainer::memory_soft_limit_in_bytes();
2405 st->print("memory_soft_limit_in_bytes: ");
2406 if (j > 0) {
2407 st->print(JLONG_FORMAT "\n", j);
2408 } else {
2409 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2410 }
2411
2412 j = OSContainer::OSContainer::memory_usage_in_bytes();
2413 st->print("memory_usage_in_bytes: ");
2414 if (j > 0) {
2415 st->print(JLONG_FORMAT "\n", j);
2416 } else {
2417 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2418 }
2419
2420 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2421 st->print("memory_max_usage_in_bytes: ");
2422 if (j > 0) {
2423 st->print(JLONG_FORMAT "\n", j);
2424 } else {
2425 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2426 }
2427 st->cr();
2428 }
2429
2430 void os::Linux::print_steal_info(outputStream* st) {
2431 if (has_initial_tick_info) {
2432 CPUPerfTicks pticks;
2433 bool res = os::Linux::get_tick_information(&pticks, -1);
2434
2435 if (res && pticks.has_steal_ticks) {
2436 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2437 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2438 double steal_ticks_perc = 0.0;
2439 if (total_ticks_difference != 0) {
2440 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2441 }
2442 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2443 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2444 }
2445 }
2446 }
2447
2448 void os::print_memory_info(outputStream* st) {
2449
2450 st->print("Memory:");
2451 st->print(" %dk page", os::vm_page_size()>>10);
2452
2453 // values in struct sysinfo are "unsigned long"
2454 struct sysinfo si;
2455 sysinfo(&si);
2456
2457 st->print(", physical " UINT64_FORMAT "k",
2458 os::physical_memory() >> 10);
2459 st->print("(" UINT64_FORMAT "k free)",
2460 os::available_memory() >> 10);
2461 st->print(", swap " UINT64_FORMAT "k",
2462 ((jlong)si.totalswap * si.mem_unit) >> 10);
2463 st->print("(" UINT64_FORMAT "k free)",
2464 ((jlong)si.freeswap * si.mem_unit) >> 10);
2465 st->cr();
2466 }
2467
2468 // Print the first "model name" line and the first "flags" line
2469 // that we find and nothing more. We assume "model name" comes
2470 // before "flags" so if we find a second "model name", then the
2471 // "flags" field is considered missing.
2472 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2473 #if defined(IA32) || defined(AMD64)
2474 // Other platforms have less repetitive cpuinfo files
2475 FILE *fp = fopen("/proc/cpuinfo", "r");
2476 if (fp) {
2477 while (!feof(fp)) {
2478 if (fgets(buf, buflen, fp)) {
2479 // Assume model name comes before flags
2480 bool model_name_printed = false;
2481 if (strstr(buf, "model name") != NULL) {
2482 if (!model_name_printed) {
2483 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2484 st->print_raw(buf);
2485 model_name_printed = true;
2486 } else {
2487 // model name printed but not flags? Odd, just return
2488 fclose(fp);
2489 return true;
2490 }
2491 }
2492 // print the flags line too
2493 if (strstr(buf, "flags") != NULL) {
2494 st->print_raw(buf);
2495 fclose(fp);
2496 return true;
2497 }
2498 }
2499 }
2500 fclose(fp);
2501 }
2502 #endif // x86 platforms
2503 return false;
2504 }
2505
2506 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2507 // Only print the model name if the platform provides this as a summary
2508 if (!print_model_name_and_flags(st, buf, buflen)) {
2509 st->print("\n/proc/cpuinfo:\n");
2510 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2511 st->print_cr(" <Not Available>");
2512 }
2513 }
2514 }
2515
2516 #if defined(AMD64) || defined(IA32) || defined(X32)
2517 const char* search_string = "model name";
2518 #elif defined(M68K)
2519 const char* search_string = "CPU";
2520 #elif defined(PPC64)
2521 const char* search_string = "cpu";
2522 #elif defined(S390)
2523 const char* search_string = "machine =";
2524 #elif defined(SPARC)
2525 const char* search_string = "cpu";
2526 #else
2527 const char* search_string = "Processor";
2528 #endif
2529
2530 // Parses the cpuinfo file for string representing the model name.
2531 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2532 FILE* fp = fopen("/proc/cpuinfo", "r");
2533 if (fp != NULL) {
2534 while (!feof(fp)) {
2535 char buf[256];
2536 if (fgets(buf, sizeof(buf), fp)) {
2537 char* start = strstr(buf, search_string);
2538 if (start != NULL) {
2539 char *ptr = start + strlen(search_string);
2540 char *end = buf + strlen(buf);
2541 while (ptr != end) {
2542 // skip whitespace and colon for the rest of the name.
2543 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2544 break;
2545 }
2546 ptr++;
2547 }
2548 if (ptr != end) {
2549 // reasonable string, get rid of newline and keep the rest
2550 char* nl = strchr(buf, '\n');
2551 if (nl != NULL) *nl = '\0';
2552 strncpy(cpuinfo, ptr, length);
2553 fclose(fp);
2554 return;
2555 }
2556 }
2557 }
2558 }
2559 fclose(fp);
2560 }
2561 // cpuinfo not found or parsing failed, just print generic string. The entire
2562 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2563 #if defined(AARCH64)
2564 strncpy(cpuinfo, "AArch64", length);
2565 #elif defined(AMD64)
2566 strncpy(cpuinfo, "x86_64", length);
2567 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2568 strncpy(cpuinfo, "ARM", length);
2569 #elif defined(IA32)
2570 strncpy(cpuinfo, "x86_32", length);
2571 #elif defined(IA64)
2572 strncpy(cpuinfo, "IA64", length);
2573 #elif defined(PPC)
2574 strncpy(cpuinfo, "PPC64", length);
2575 #elif defined(S390)
2576 strncpy(cpuinfo, "S390", length);
2577 #elif defined(SPARC)
2578 strncpy(cpuinfo, "sparcv9", length);
2579 #elif defined(ZERO_LIBARCH)
2580 strncpy(cpuinfo, ZERO_LIBARCH, length);
2581 #else
2582 strncpy(cpuinfo, "unknown", length);
2583 #endif
2584 }
2585
2586 static void print_signal_handler(outputStream* st, int sig,
2587 char* buf, size_t buflen);
2588
2589 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2590 st->print_cr("Signal Handlers:");
2591 print_signal_handler(st, SIGSEGV, buf, buflen);
2592 print_signal_handler(st, SIGBUS , buf, buflen);
2593 print_signal_handler(st, SIGFPE , buf, buflen);
2594 print_signal_handler(st, SIGPIPE, buf, buflen);
2595 print_signal_handler(st, SIGXFSZ, buf, buflen);
2596 print_signal_handler(st, SIGILL , buf, buflen);
2597 print_signal_handler(st, SR_signum, buf, buflen);
2598 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2599 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2600 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2601 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2602 #if defined(PPC64)
2603 print_signal_handler(st, SIGTRAP, buf, buflen);
2604 #endif
2605 }
2606
2607 static char saved_jvm_path[MAXPATHLEN] = {0};
2608
2609 // Find the full path to the current module, libjvm.so
2610 void os::jvm_path(char *buf, jint buflen) {
2611 // Error checking.
2612 if (buflen < MAXPATHLEN) {
2613 assert(false, "must use a large-enough buffer");
2614 buf[0] = '\0';
2615 return;
2616 }
2617 // Lazy resolve the path to current module.
2618 if (saved_jvm_path[0] != 0) {
2619 strcpy(buf, saved_jvm_path);
2620 return;
2621 }
2622
2623 char dli_fname[MAXPATHLEN];
2624 bool ret = dll_address_to_library_name(
2625 CAST_FROM_FN_PTR(address, os::jvm_path),
2626 dli_fname, sizeof(dli_fname), NULL);
2627 assert(ret, "cannot locate libjvm");
2628 char *rp = NULL;
2629 if (ret && dli_fname[0] != '\0') {
2630 rp = os::Posix::realpath(dli_fname, buf, buflen);
2631 }
2632 if (rp == NULL) {
2633 return;
2634 }
2635
2636 if (Arguments::sun_java_launcher_is_altjvm()) {
2637 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2638 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2639 // If "/jre/lib/" appears at the right place in the string, then
2640 // assume we are installed in a JDK and we're done. Otherwise, check
2641 // for a JAVA_HOME environment variable and fix up the path so it
2642 // looks like libjvm.so is installed there (append a fake suffix
2643 // hotspot/libjvm.so).
2644 const char *p = buf + strlen(buf) - 1;
2645 for (int count = 0; p > buf && count < 5; ++count) {
2646 for (--p; p > buf && *p != '/'; --p)
2647 /* empty */ ;
2648 }
2649
2650 if (strncmp(p, "/jre/lib/", 9) != 0) {
2651 // Look for JAVA_HOME in the environment.
2652 char* java_home_var = ::getenv("JAVA_HOME");
2653 if (java_home_var != NULL && java_home_var[0] != 0) {
2654 char* jrelib_p;
2655 int len;
2656
2657 // Check the current module name "libjvm.so".
2658 p = strrchr(buf, '/');
2659 if (p == NULL) {
2660 return;
2661 }
2662 assert(strstr(p, "/libjvm") == p, "invalid library name");
2663
2664 rp = os::Posix::realpath(java_home_var, buf, buflen);
2665 if (rp == NULL) {
2666 return;
2667 }
2668
2669 // determine if this is a legacy image or modules image
2670 // modules image doesn't have "jre" subdirectory
2671 len = strlen(buf);
2672 assert(len < buflen, "Ran out of buffer room");
2673 jrelib_p = buf + len;
2674 snprintf(jrelib_p, buflen-len, "/jre/lib");
2675 if (0 != access(buf, F_OK)) {
2676 snprintf(jrelib_p, buflen-len, "/lib");
2677 }
2678
2679 if (0 == access(buf, F_OK)) {
2680 // Use current module name "libjvm.so"
2681 len = strlen(buf);
2682 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2683 } else {
2684 // Go back to path of .so
2685 rp = os::Posix::realpath(dli_fname, buf, buflen);
2686 if (rp == NULL) {
2687 return;
2688 }
2689 }
2690 }
2691 }
2692 }
2693
2694 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2695 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2696 }
2697
2698 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2699 // no prefix required, not even "_"
2700 }
2701
2702 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2703 // no suffix required
2704 }
2705
2706 ////////////////////////////////////////////////////////////////////////////////
2707 // sun.misc.Signal support
2708
2709 static void UserHandler(int sig, void *siginfo, void *context) {
2710 // Ctrl-C is pressed during error reporting, likely because the error
2711 // handler fails to abort. Let VM die immediately.
2712 if (sig == SIGINT && VMError::is_error_reported()) {
2713 os::die();
2714 }
2715
2716 os::signal_notify(sig);
2717 }
2718
2719 void* os::user_handler() {
2720 return CAST_FROM_FN_PTR(void*, UserHandler);
2721 }
2722
2723 extern "C" {
2724 typedef void (*sa_handler_t)(int);
2725 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2726 }
2727
2728 void* os::signal(int signal_number, void* handler) {
2729 struct sigaction sigAct, oldSigAct;
2730
2731 sigfillset(&(sigAct.sa_mask));
2732 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2733 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2734
2735 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2736 // -1 means registration failed
2737 return (void *)-1;
2738 }
2739
2740 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2741 }
2742
2743 void os::signal_raise(int signal_number) {
2744 ::raise(signal_number);
2745 }
2746
2747 // The following code is moved from os.cpp for making this
2748 // code platform specific, which it is by its very nature.
2749
2750 // Will be modified when max signal is changed to be dynamic
2751 int os::sigexitnum_pd() {
2752 return NSIG;
2753 }
2754
2755 // a counter for each possible signal value
2756 static volatile jint pending_signals[NSIG+1] = { 0 };
2757
2758 // Linux(POSIX) specific hand shaking semaphore.
2759 static Semaphore* sig_sem = NULL;
2760 static PosixSemaphore sr_semaphore;
2761
2762 static void jdk_misc_signal_init() {
2763 // Initialize signal structures
2764 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2765
2766 // Initialize signal semaphore
2767 sig_sem = new Semaphore();
2768 }
2769
2770 void os::signal_notify(int sig) {
2771 if (sig_sem != NULL) {
2772 Atomic::inc(&pending_signals[sig]);
2773 sig_sem->signal();
2774 } else {
2775 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2776 // initialization isn't called.
2777 assert(ReduceSignalUsage, "signal semaphore should be created");
2778 }
2779 }
2780
2781 static int check_pending_signals() {
2782 for (;;) {
2783 for (int i = 0; i < NSIG + 1; i++) {
2784 jint n = pending_signals[i];
2785 if (n > 0 && n == Atomic::cmpxchg(&pending_signals[i], n, n - 1)) {
2786 return i;
2787 }
2788 }
2789 JavaThread *thread = JavaThread::current();
2790 ThreadBlockInVM tbivm(thread);
2791
2792 bool threadIsSuspended;
2793 do {
2794 thread->set_suspend_equivalent();
2795 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2796 sig_sem->wait();
2797
2798 // were we externally suspended while we were waiting?
2799 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2800 if (threadIsSuspended) {
2801 // The semaphore has been incremented, but while we were waiting
2802 // another thread suspended us. We don't want to continue running
2803 // while suspended because that would surprise the thread that
2804 // suspended us.
2805 sig_sem->signal();
2806
2807 thread->java_suspend_self();
2808 }
2809 } while (threadIsSuspended);
2810 }
2811 }
2812
2813 int os::signal_wait() {
2814 return check_pending_signals();
2815 }
2816
2817 ////////////////////////////////////////////////////////////////////////////////
2818 // Virtual Memory
2819
2820 int os::vm_page_size() {
2821 // Seems redundant as all get out
2822 assert(os::Linux::page_size() != -1, "must call os::init");
2823 return os::Linux::page_size();
2824 }
2825
2826 // Solaris allocates memory by pages.
2827 int os::vm_allocation_granularity() {
2828 assert(os::Linux::page_size() != -1, "must call os::init");
2829 return os::Linux::page_size();
2830 }
2831
2832 // Rationale behind this function:
2833 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2834 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2835 // samples for JITted code. Here we create private executable mapping over the code cache
2836 // and then we can use standard (well, almost, as mapping can change) way to provide
2837 // info for the reporting script by storing timestamp and location of symbol
2838 void linux_wrap_code(char* base, size_t size) {
2839 static volatile jint cnt = 0;
2840
2841 if (!UseOprofile) {
2842 return;
2843 }
2844
2845 char buf[PATH_MAX+1];
2846 int num = Atomic::add(&cnt, 1);
2847
2848 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2849 os::get_temp_directory(), os::current_process_id(), num);
2850 unlink(buf);
2851
2852 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2853
2854 if (fd != -1) {
2855 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2856 if (rv != (off_t)-1) {
2857 if (::write(fd, "", 1) == 1) {
2858 mmap(base, size,
2859 PROT_READ|PROT_WRITE|PROT_EXEC,
2860 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2861 }
2862 }
2863 ::close(fd);
2864 unlink(buf);
2865 }
2866 }
2867
2868 static bool recoverable_mmap_error(int err) {
2869 // See if the error is one we can let the caller handle. This
2870 // list of errno values comes from JBS-6843484. I can't find a
2871 // Linux man page that documents this specific set of errno
2872 // values so while this list currently matches Solaris, it may
2873 // change as we gain experience with this failure mode.
2874 switch (err) {
2875 case EBADF:
2876 case EINVAL:
2877 case ENOTSUP:
2878 // let the caller deal with these errors
2879 return true;
2880
2881 default:
2882 // Any remaining errors on this OS can cause our reserved mapping
2883 // to be lost. That can cause confusion where different data
2884 // structures think they have the same memory mapped. The worst
2885 // scenario is if both the VM and a library think they have the
2886 // same memory mapped.
2887 return false;
2888 }
2889 }
2890
2891 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2892 int err) {
2893 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2894 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2895 os::strerror(err), err);
2896 }
2897
2898 static void warn_fail_commit_memory(char* addr, size_t size,
2899 size_t alignment_hint, bool exec,
2900 int err) {
2901 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2902 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2903 alignment_hint, exec, os::strerror(err), err);
2904 }
2905
2906 // NOTE: Linux kernel does not really reserve the pages for us.
2907 // All it does is to check if there are enough free pages
2908 // left at the time of mmap(). This could be a potential
2909 // problem.
2910 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2911 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2912 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2913 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2914 if (res != (uintptr_t) MAP_FAILED) {
2915 if (UseNUMAInterleaving) {
2916 numa_make_global(addr, size);
2917 }
2918 return 0;
2919 }
2920
2921 int err = errno; // save errno from mmap() call above
2922
2923 if (!recoverable_mmap_error(err)) {
2924 warn_fail_commit_memory(addr, size, exec, err);
2925 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2926 }
2927
2928 return err;
2929 }
2930
2931 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2932 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2933 }
2934
2935 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2936 const char* mesg) {
2937 assert(mesg != NULL, "mesg must be specified");
2938 int err = os::Linux::commit_memory_impl(addr, size, exec);
2939 if (err != 0) {
2940 // the caller wants all commit errors to exit with the specified mesg:
2941 warn_fail_commit_memory(addr, size, exec, err);
2942 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2943 }
2944 }
2945
2946 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2947 #ifndef MAP_HUGETLB
2948 #define MAP_HUGETLB 0x40000
2949 #endif
2950
2951 // mmap: If MAP_HUGETLB is set, and the system supports multiple huge page sizes,
2952 // flag bits [26:31] can be used to encode the log2 of the desired huge page size.
2953 // Otherwise the system's default huge page size will be used.
2954 #ifndef MAP_HUGE_SHIFT
2955 #define MAP_HUGE_SHIFT 26
2956 #endif
2957
2958 #ifndef MAP_HUGE_2MB
2959 #define MAP_HUGE_2MB (21 << MAP_HUGE_SHIFT)
2960 #endif
2961
2962 #ifndef MAP_HUGE_1GB
2963 #define MAP_HUGE_1GB (30 << MAP_HUGE_SHIFT)
2964 #endif
2965
2966 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2967 #ifndef MADV_HUGEPAGE
2968 #define MADV_HUGEPAGE 14
2969 #endif
2970
2971 int os::Linux::commit_memory_impl(char* addr, size_t size,
2972 size_t alignment_hint, bool exec) {
2973 int err = os::Linux::commit_memory_impl(addr, size, exec);
2974 if (err == 0) {
2975 realign_memory(addr, size, alignment_hint);
2976 }
2977 return err;
2978 }
2979
2980 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2981 bool exec) {
2982 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2983 }
2984
2985 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2986 size_t alignment_hint, bool exec,
2987 const char* mesg) {
2988 assert(mesg != NULL, "mesg must be specified");
2989 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2990 if (err != 0) {
2991 // the caller wants all commit errors to exit with the specified mesg:
2992 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2993 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2994 }
2995 }
2996
2997 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2998 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2999 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
3000 // be supported or the memory may already be backed by huge pages.
3001 ::madvise(addr, bytes, MADV_HUGEPAGE);
3002 }
3003 }
3004
3005 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
3006 // This method works by doing an mmap over an existing mmaping and effectively discarding
3007 // the existing pages. However it won't work for SHM-based large pages that cannot be
3008 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
3009 // small pages on top of the SHM segment. This method always works for small pages, so we
3010 // allow that in any case.
3011 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
3012 commit_memory(addr, bytes, alignment_hint, !ExecMem);
3013 }
3014 }
3015
3016 void os::numa_make_global(char *addr, size_t bytes) {
3017 Linux::numa_interleave_memory(addr, bytes);
3018 }
3019
3020 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
3021 // bind policy to MPOL_PREFERRED for the current thread.
3022 #define USE_MPOL_PREFERRED 0
3023
3024 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
3025 // To make NUMA and large pages more robust when both enabled, we need to ease
3026 // the requirements on where the memory should be allocated. MPOL_BIND is the
3027 // default policy and it will force memory to be allocated on the specified
3028 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
3029 // the specified node, but will not force it. Using this policy will prevent
3030 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
3031 // free large pages.
3032 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
3033 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
3034 }
3035
3036 bool os::numa_topology_changed() { return false; }
3037
3038 size_t os::numa_get_groups_num() {
3039 // Return just the number of nodes in which it's possible to allocate memory
3040 // (in numa terminology, configured nodes).
3041 return Linux::numa_num_configured_nodes();
3042 }
3043
3044 int os::numa_get_group_id() {
3045 int cpu_id = Linux::sched_getcpu();
3046 if (cpu_id != -1) {
3047 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3048 if (lgrp_id != -1) {
3049 return lgrp_id;
3050 }
3051 }
3052 return 0;
3053 }
3054
3055 int os::numa_get_group_id_for_address(const void* address) {
3056 void** pages = const_cast<void**>(&address);
3057 int id = -1;
3058
3059 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) {
3060 return -1;
3061 }
3062 if (id < 0) {
3063 return -1;
3064 }
3065 return id;
3066 }
3067
3068 int os::Linux::get_existing_num_nodes() {
3069 int node;
3070 int highest_node_number = Linux::numa_max_node();
3071 int num_nodes = 0;
3072
3073 // Get the total number of nodes in the system including nodes without memory.
3074 for (node = 0; node <= highest_node_number; node++) {
3075 if (is_node_in_existing_nodes(node)) {
3076 num_nodes++;
3077 }
3078 }
3079 return num_nodes;
3080 }
3081
3082 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3083 int highest_node_number = Linux::numa_max_node();
3084 size_t i = 0;
3085
3086 // Map all node ids in which it is possible to allocate memory. Also nodes are
3087 // not always consecutively available, i.e. available from 0 to the highest
3088 // node number. If the nodes have been bound explicitly using numactl membind,
3089 // then allocate memory from those nodes only.
3090 for (int node = 0; node <= highest_node_number; node++) {
3091 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3092 ids[i++] = node;
3093 }
3094 }
3095 return i;
3096 }
3097
3098 bool os::get_page_info(char *start, page_info* info) {
3099 return false;
3100 }
3101
3102 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3103 page_info* page_found) {
3104 return end;
3105 }
3106
3107
3108 int os::Linux::sched_getcpu_syscall(void) {
3109 unsigned int cpu = 0;
3110 int retval = -1;
3111
3112 #if defined(IA32)
3113 #ifndef SYS_getcpu
3114 #define SYS_getcpu 318
3115 #endif
3116 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3117 #elif defined(AMD64)
3118 // Unfortunately we have to bring all these macros here from vsyscall.h
3119 // to be able to compile on old linuxes.
3120 #define __NR_vgetcpu 2
3121 #define VSYSCALL_START (-10UL << 20)
3122 #define VSYSCALL_SIZE 1024
3123 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3124 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3125 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3126 retval = vgetcpu(&cpu, NULL, NULL);
3127 #endif
3128
3129 return (retval == -1) ? retval : cpu;
3130 }
3131
3132 void os::Linux::sched_getcpu_init() {
3133 // sched_getcpu() should be in libc.
3134 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3135 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3136
3137 // If it's not, try a direct syscall.
3138 if (sched_getcpu() == -1) {
3139 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3140 (void*)&sched_getcpu_syscall));
3141 }
3142
3143 if (sched_getcpu() == -1) {
3144 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3145 }
3146 }
3147
3148 // Something to do with the numa-aware allocator needs these symbols
3149 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3150 extern "C" JNIEXPORT void numa_error(char *where) { }
3151
3152 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3153 // load symbol from base version instead.
3154 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3155 void *f = dlvsym(handle, name, "libnuma_1.1");
3156 if (f == NULL) {
3157 f = dlsym(handle, name);
3158 }
3159 return f;
3160 }
3161
3162 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3163 // Return NULL if the symbol is not defined in this particular version.
3164 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3165 return dlvsym(handle, name, "libnuma_1.2");
3166 }
3167
3168 bool os::Linux::libnuma_init() {
3169 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3170 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3171 if (handle != NULL) {
3172 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3173 libnuma_dlsym(handle, "numa_node_to_cpus")));
3174 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3175 libnuma_dlsym(handle, "numa_max_node")));
3176 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3177 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3178 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3179 libnuma_dlsym(handle, "numa_available")));
3180 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3181 libnuma_dlsym(handle, "numa_tonode_memory")));
3182 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3183 libnuma_dlsym(handle, "numa_interleave_memory")));
3184 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3185 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3186 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3187 libnuma_dlsym(handle, "numa_set_bind_policy")));
3188 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3189 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3190 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3191 libnuma_dlsym(handle, "numa_distance")));
3192 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3193 libnuma_v2_dlsym(handle, "numa_get_membind")));
3194 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3195 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3196 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t,
3197 libnuma_dlsym(handle, "numa_move_pages")));
3198 set_numa_set_preferred(CAST_TO_FN_PTR(numa_set_preferred_func_t,
3199 libnuma_dlsym(handle, "numa_set_preferred")));
3200
3201 if (numa_available() != -1) {
3202 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3203 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3204 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3205 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3206 set_numa_membind_bitmask(_numa_get_membind());
3207 // Create an index -> node mapping, since nodes are not always consecutive
3208 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3209 rebuild_nindex_to_node_map();
3210 // Create a cpu -> node mapping
3211 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3212 rebuild_cpu_to_node_map();
3213 return true;
3214 }
3215 }
3216 }
3217 return false;
3218 }
3219
3220 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3221 // Creating guard page is very expensive. Java thread has HotSpot
3222 // guard pages, only enable glibc guard page for non-Java threads.
3223 // (Remember: compiler thread is a Java thread, too!)
3224 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3225 }
3226
3227 void os::Linux::rebuild_nindex_to_node_map() {
3228 int highest_node_number = Linux::numa_max_node();
3229
3230 nindex_to_node()->clear();
3231 for (int node = 0; node <= highest_node_number; node++) {
3232 if (Linux::is_node_in_existing_nodes(node)) {
3233 nindex_to_node()->append(node);
3234 }
3235 }
3236 }
3237
3238 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3239 // The table is later used in get_node_by_cpu().
3240 void os::Linux::rebuild_cpu_to_node_map() {
3241 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3242 // in libnuma (possible values are starting from 16,
3243 // and continuing up with every other power of 2, but less
3244 // than the maximum number of CPUs supported by kernel), and
3245 // is a subject to change (in libnuma version 2 the requirements
3246 // are more reasonable) we'll just hardcode the number they use
3247 // in the library.
3248 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3249
3250 size_t cpu_num = processor_count();
3251 size_t cpu_map_size = NCPUS / BitsPerCLong;
3252 size_t cpu_map_valid_size =
3253 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3254
3255 cpu_to_node()->clear();
3256 cpu_to_node()->at_grow(cpu_num - 1);
3257
3258 size_t node_num = get_existing_num_nodes();
3259
3260 int distance = 0;
3261 int closest_distance = INT_MAX;
3262 int closest_node = 0;
3263 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3264 for (size_t i = 0; i < node_num; i++) {
3265 // Check if node is configured (not a memory-less node). If it is not, find
3266 // the closest configured node. Check also if node is bound, i.e. it's allowed
3267 // to allocate memory from the node. If it's not allowed, map cpus in that node
3268 // to the closest node from which memory allocation is allowed.
3269 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3270 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3271 closest_distance = INT_MAX;
3272 // Check distance from all remaining nodes in the system. Ignore distance
3273 // from itself, from another non-configured node, and from another non-bound
3274 // node.
3275 for (size_t m = 0; m < node_num; m++) {
3276 if (m != i &&
3277 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3278 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3279 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3280 // If a closest node is found, update. There is always at least one
3281 // configured and bound node in the system so there is always at least
3282 // one node close.
3283 if (distance != 0 && distance < closest_distance) {
3284 closest_distance = distance;
3285 closest_node = nindex_to_node()->at(m);
3286 }
3287 }
3288 }
3289 } else {
3290 // Current node is already a configured node.
3291 closest_node = nindex_to_node()->at(i);
3292 }
3293
3294 // Get cpus from the original node and map them to the closest node. If node
3295 // is a configured node (not a memory-less node), then original node and
3296 // closest node are the same.
3297 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3298 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3299 if (cpu_map[j] != 0) {
3300 for (size_t k = 0; k < BitsPerCLong; k++) {
3301 if (cpu_map[j] & (1UL << k)) {
3302 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3303 }
3304 }
3305 }
3306 }
3307 }
3308 }
3309 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3310 }
3311
3312 int os::Linux::get_node_by_cpu(int cpu_id) {
3313 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3314 return cpu_to_node()->at(cpu_id);
3315 }
3316 return -1;
3317 }
3318
3319 GrowableArray<int>* os::Linux::_cpu_to_node;
3320 GrowableArray<int>* os::Linux::_nindex_to_node;
3321 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3322 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3323 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3324 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3325 os::Linux::numa_available_func_t os::Linux::_numa_available;
3326 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3327 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3328 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3329 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3330 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3331 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3332 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3333 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3334 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages;
3335 os::Linux::numa_set_preferred_func_t os::Linux::_numa_set_preferred;
3336 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3337 unsigned long* os::Linux::_numa_all_nodes;
3338 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3339 struct bitmask* os::Linux::_numa_nodes_ptr;
3340 struct bitmask* os::Linux::_numa_interleave_bitmask;
3341 struct bitmask* os::Linux::_numa_membind_bitmask;
3342
3343 bool os::pd_uncommit_memory(char* addr, size_t size) {
3344 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3345 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3346 return res != (uintptr_t) MAP_FAILED;
3347 }
3348
3349 static address get_stack_commited_bottom(address bottom, size_t size) {
3350 address nbot = bottom;
3351 address ntop = bottom + size;
3352
3353 size_t page_sz = os::vm_page_size();
3354 unsigned pages = size / page_sz;
3355
3356 unsigned char vec[1];
3357 unsigned imin = 1, imax = pages + 1, imid;
3358 int mincore_return_value = 0;
3359
3360 assert(imin <= imax, "Unexpected page size");
3361
3362 while (imin < imax) {
3363 imid = (imax + imin) / 2;
3364 nbot = ntop - (imid * page_sz);
3365
3366 // Use a trick with mincore to check whether the page is mapped or not.
3367 // mincore sets vec to 1 if page resides in memory and to 0 if page
3368 // is swapped output but if page we are asking for is unmapped
3369 // it returns -1,ENOMEM
3370 mincore_return_value = mincore(nbot, page_sz, vec);
3371
3372 if (mincore_return_value == -1) {
3373 // Page is not mapped go up
3374 // to find first mapped page
3375 if (errno != EAGAIN) {
3376 assert(errno == ENOMEM, "Unexpected mincore errno");
3377 imax = imid;
3378 }
3379 } else {
3380 // Page is mapped go down
3381 // to find first not mapped page
3382 imin = imid + 1;
3383 }
3384 }
3385
3386 nbot = nbot + page_sz;
3387
3388 // Adjust stack bottom one page up if last checked page is not mapped
3389 if (mincore_return_value == -1) {
3390 nbot = nbot + page_sz;
3391 }
3392
3393 return nbot;
3394 }
3395
3396 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3397 int mincore_return_value;
3398 const size_t stripe = 1024; // query this many pages each time
3399 unsigned char vec[stripe + 1];
3400 // set a guard
3401 vec[stripe] = 'X';
3402
3403 const size_t page_sz = os::vm_page_size();
3404 size_t pages = size / page_sz;
3405
3406 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3407 assert(is_aligned(size, page_sz), "Size must be page aligned");
3408
3409 committed_start = NULL;
3410
3411 int loops = (pages + stripe - 1) / stripe;
3412 int committed_pages = 0;
3413 address loop_base = start;
3414 bool found_range = false;
3415
3416 for (int index = 0; index < loops && !found_range; index ++) {
3417 assert(pages > 0, "Nothing to do");
3418 int pages_to_query = (pages >= stripe) ? stripe : pages;
3419 pages -= pages_to_query;
3420
3421 // Get stable read
3422 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3423
3424 // During shutdown, some memory goes away without properly notifying NMT,
3425 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3426 // Bailout and return as not committed for now.
3427 if (mincore_return_value == -1 && errno == ENOMEM) {
3428 return false;
3429 }
3430
3431 assert(vec[stripe] == 'X', "overflow guard");
3432 assert(mincore_return_value == 0, "Range must be valid");
3433 // Process this stripe
3434 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3435 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3436 // End of current contiguous region
3437 if (committed_start != NULL) {
3438 found_range = true;
3439 break;
3440 }
3441 } else { // committed
3442 // Start of region
3443 if (committed_start == NULL) {
3444 committed_start = loop_base + page_sz * vecIdx;
3445 }
3446 committed_pages ++;
3447 }
3448 }
3449
3450 loop_base += pages_to_query * page_sz;
3451 }
3452
3453 if (committed_start != NULL) {
3454 assert(committed_pages > 0, "Must have committed region");
3455 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3456 assert(committed_start >= start && committed_start < start + size, "Out of range");
3457 committed_size = page_sz * committed_pages;
3458 return true;
3459 } else {
3460 assert(committed_pages == 0, "Should not have committed region");
3461 return false;
3462 }
3463 }
3464
3465
3466 // Linux uses a growable mapping for the stack, and if the mapping for
3467 // the stack guard pages is not removed when we detach a thread the
3468 // stack cannot grow beyond the pages where the stack guard was
3469 // mapped. If at some point later in the process the stack expands to
3470 // that point, the Linux kernel cannot expand the stack any further
3471 // because the guard pages are in the way, and a segfault occurs.
3472 //
3473 // However, it's essential not to split the stack region by unmapping
3474 // a region (leaving a hole) that's already part of the stack mapping,
3475 // so if the stack mapping has already grown beyond the guard pages at
3476 // the time we create them, we have to truncate the stack mapping.
3477 // So, we need to know the extent of the stack mapping when
3478 // create_stack_guard_pages() is called.
3479
3480 // We only need this for stacks that are growable: at the time of
3481 // writing thread stacks don't use growable mappings (i.e. those
3482 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3483 // only applies to the main thread.
3484
3485 // If the (growable) stack mapping already extends beyond the point
3486 // where we're going to put our guard pages, truncate the mapping at
3487 // that point by munmap()ping it. This ensures that when we later
3488 // munmap() the guard pages we don't leave a hole in the stack
3489 // mapping. This only affects the main/primordial thread
3490
3491 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3492 if (os::is_primordial_thread()) {
3493 // As we manually grow stack up to bottom inside create_attached_thread(),
3494 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3495 // we don't need to do anything special.
3496 // Check it first, before calling heavy function.
3497 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3498 unsigned char vec[1];
3499
3500 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3501 // Fallback to slow path on all errors, including EAGAIN
3502 stack_extent = (uintptr_t) get_stack_commited_bottom(
3503 os::Linux::initial_thread_stack_bottom(),
3504 (size_t)addr - stack_extent);
3505 }
3506
3507 if (stack_extent < (uintptr_t)addr) {
3508 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3509 }
3510 }
3511
3512 return os::commit_memory(addr, size, !ExecMem);
3513 }
3514
3515 // If this is a growable mapping, remove the guard pages entirely by
3516 // munmap()ping them. If not, just call uncommit_memory(). This only
3517 // affects the main/primordial thread, but guard against future OS changes.
3518 // It's safe to always unmap guard pages for primordial thread because we
3519 // always place it right after end of the mapped region.
3520
3521 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3522 uintptr_t stack_extent, stack_base;
3523
3524 if (os::is_primordial_thread()) {
3525 return ::munmap(addr, size) == 0;
3526 }
3527
3528 return os::uncommit_memory(addr, size);
3529 }
3530
3531 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3532 // at 'requested_addr'. If there are existing memory mappings at the same
3533 // location, however, they will be overwritten. If 'fixed' is false,
3534 // 'requested_addr' is only treated as a hint, the return value may or
3535 // may not start from the requested address. Unlike Linux mmap(), this
3536 // function returns NULL to indicate failure.
3537 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3538 char * addr;
3539 int flags;
3540
3541 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3542 if (fixed) {
3543 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3544 flags |= MAP_FIXED;
3545 }
3546
3547 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3548 // touch an uncommitted page. Otherwise, the read/write might
3549 // succeed if we have enough swap space to back the physical page.
3550 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3551 flags, -1, 0);
3552
3553 return addr == MAP_FAILED ? NULL : addr;
3554 }
3555
3556 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3557 // (req_addr != NULL) or with a given alignment.
3558 // - bytes shall be a multiple of alignment.
3559 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3560 // - alignment sets the alignment at which memory shall be allocated.
3561 // It must be a multiple of allocation granularity.
3562 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3563 // req_addr or NULL.
3564 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3565
3566 size_t extra_size = bytes;
3567 if (req_addr == NULL && alignment > 0) {
3568 extra_size += alignment;
3569 }
3570
3571 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3572 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3573 -1, 0);
3574 if (start == MAP_FAILED) {
3575 start = NULL;
3576 } else {
3577 if (req_addr != NULL) {
3578 if (start != req_addr) {
3579 ::munmap(start, extra_size);
3580 start = NULL;
3581 }
3582 } else {
3583 char* const start_aligned = align_up(start, alignment);
3584 char* const end_aligned = start_aligned + bytes;
3585 char* const end = start + extra_size;
3586 if (start_aligned > start) {
3587 ::munmap(start, start_aligned - start);
3588 }
3589 if (end_aligned < end) {
3590 ::munmap(end_aligned, end - end_aligned);
3591 }
3592 start = start_aligned;
3593 }
3594 }
3595 return start;
3596 }
3597
3598 static int anon_munmap(char * addr, size_t size) {
3599 return ::munmap(addr, size) == 0;
3600 }
3601
3602 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3603 size_t alignment_hint) {
3604 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3605 }
3606
3607 bool os::pd_release_memory(char* addr, size_t size) {
3608 return anon_munmap(addr, size);
3609 }
3610
3611 static bool linux_mprotect(char* addr, size_t size, int prot) {
3612 // Linux wants the mprotect address argument to be page aligned.
3613 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3614
3615 // According to SUSv3, mprotect() should only be used with mappings
3616 // established by mmap(), and mmap() always maps whole pages. Unaligned
3617 // 'addr' likely indicates problem in the VM (e.g. trying to change
3618 // protection of malloc'ed or statically allocated memory). Check the
3619 // caller if you hit this assert.
3620 assert(addr == bottom, "sanity check");
3621
3622 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3623 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3624 return ::mprotect(bottom, size, prot) == 0;
3625 }
3626
3627 // Set protections specified
3628 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3629 bool is_committed) {
3630 unsigned int p = 0;
3631 switch (prot) {
3632 case MEM_PROT_NONE: p = PROT_NONE; break;
3633 case MEM_PROT_READ: p = PROT_READ; break;
3634 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3635 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3636 default:
3637 ShouldNotReachHere();
3638 }
3639 // is_committed is unused.
3640 return linux_mprotect(addr, bytes, p);
3641 }
3642
3643 bool os::guard_memory(char* addr, size_t size) {
3644 return linux_mprotect(addr, size, PROT_NONE);
3645 }
3646
3647 bool os::unguard_memory(char* addr, size_t size) {
3648 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3649 }
3650
3651 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3652 size_t page_size) {
3653 bool result = false;
3654 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3655 MAP_ANONYMOUS|MAP_PRIVATE,
3656 -1, 0);
3657 if (p != MAP_FAILED) {
3658 void *aligned_p = align_up(p, page_size);
3659
3660 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3661
3662 munmap(p, page_size * 2);
3663 }
3664
3665 if (warn && !result) {
3666 warning("TransparentHugePages is not supported by the operating system.");
3667 }
3668
3669 return result;
3670 }
3671
3672 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3673 bool result = false;
3674 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3675 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3676 -1, 0);
3677
3678 if (p != MAP_FAILED) {
3679 // We don't know if this really is a huge page or not.
3680 FILE *fp = fopen("/proc/self/maps", "r");
3681 if (fp) {
3682 while (!feof(fp)) {
3683 char chars[257];
3684 long x = 0;
3685 if (fgets(chars, sizeof(chars), fp)) {
3686 if (sscanf(chars, "%lx-%*x", &x) == 1
3687 && x == (long)p) {
3688 if (strstr (chars, "hugepage")) {
3689 result = true;
3690 break;
3691 }
3692 }
3693 }
3694 }
3695 fclose(fp);
3696 }
3697 munmap(p, page_size);
3698 }
3699
3700 if (warn && !result) {
3701 warning("HugeTLBFS is not supported by the operating system.");
3702 }
3703
3704 return result;
3705 }
3706
3707 // From the coredump_filter documentation:
3708 //
3709 // - (bit 0) anonymous private memory
3710 // - (bit 1) anonymous shared memory
3711 // - (bit 2) file-backed private memory
3712 // - (bit 3) file-backed shared memory
3713 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3714 // effective only if the bit 2 is cleared)
3715 // - (bit 5) hugetlb private memory
3716 // - (bit 6) hugetlb shared memory
3717 // - (bit 7) dax private memory
3718 // - (bit 8) dax shared memory
3719 //
3720 static void set_coredump_filter(CoredumpFilterBit bit) {
3721 FILE *f;
3722 long cdm;
3723
3724 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3725 return;
3726 }
3727
3728 if (fscanf(f, "%lx", &cdm) != 1) {
3729 fclose(f);
3730 return;
3731 }
3732
3733 long saved_cdm = cdm;
3734 rewind(f);
3735 cdm |= bit;
3736
3737 if (cdm != saved_cdm) {
3738 fprintf(f, "%#lx", cdm);
3739 }
3740
3741 fclose(f);
3742 }
3743
3744 // Large page support
3745
3746 static size_t _large_page_size = 0;
3747
3748 size_t os::Linux::find_default_large_page_size() {
3749 size_t large_page_size = 0;
3750
3751 // large_page_size on Linux is used to round up heap size. x86 uses either
3752 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3753 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3754 // page as large as 256M.
3755 //
3756 // Here we try to figure out page size by parsing /proc/meminfo and looking
3757 // for a line with the following format:
3758 // Hugepagesize: 2048 kB
3759 //
3760 // If we can't determine the value (e.g. /proc is not mounted, or the text
3761 // format has been changed), we'll use the largest page size supported by
3762 // the processor.
3763
3764 #ifndef ZERO
3765 large_page_size =
3766 AARCH64_ONLY(2 * M)
3767 AMD64_ONLY(2 * M)
3768 ARM32_ONLY(2 * M)
3769 IA32_ONLY(4 * M)
3770 IA64_ONLY(256 * M)
3771 PPC_ONLY(4 * M)
3772 S390_ONLY(1 * M)
3773 SPARC_ONLY(4 * M);
3774 #endif // ZERO
3775
3776 FILE *fp = fopen("/proc/meminfo", "r");
3777 if (fp) {
3778 while (!feof(fp)) {
3779 int x = 0;
3780 char buf[16];
3781 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3782 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3783 large_page_size = x * K;
3784 break;
3785 }
3786 } else {
3787 // skip to next line
3788 for (;;) {
3789 int ch = fgetc(fp);
3790 if (ch == EOF || ch == (int)'\n') break;
3791 }
3792 }
3793 }
3794 fclose(fp);
3795 }
3796 return large_page_size;
3797 }
3798
3799 void os::Linux::find_large_page_sizes() {
3800 _page_sizes[0] = 0;
3801 // We need to scan /sys/kernel/mm/hugepages
3802 // to discover the available page sizes
3803 const char* sys_hugepages = "/sys/kernel/mm/hugepages";
3804 if (dir_is_empty(sys_hugepages)) {
3805 return;
3806 }
3807
3808 DIR *dir = opendir(sys_hugepages);
3809 if (dir == NULL) {
3810 return;
3811 }
3812
3813 struct dirent *entry;
3814 size_t page_size;
3815 int count = 0;
3816
3817 while (count < page_sizes_max - 1 && (entry = readdir(dir)) != NULL) {
3818 if(entry->d_type == DT_DIR &&
3819 sscanf(entry->d_name, "hugepages-%zukB", &page_size) ) {
3820
3821 // The kernel is using kB, hotspot uses bytes
3822 _page_sizes[count] = page_size * K;
3823 count++;
3824 }
3825 }
3826 _page_sizes[count] = 0;
3827 closedir(dir);
3828 return;
3829 }
3830
3831 bool os::Linux::is_valid_large_page_size(size_t page_size) {
3832 find_large_page_sizes();
3833 int count = 0;
3834 while (_page_sizes[count] != 0) {
3835 if (_page_sizes[count++] == page_size) {
3836 return true;
3837 }
3838 }
3839 return false;
3840 }
3841
3842 size_t os::Linux::setup_large_page_size() {
3843 _large_page_size = Linux::find_default_large_page_size();
3844
3845 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != _large_page_size ) {
3846 if (is_valid_large_page_size(LargePageSizeInBytes)) {
3847 _large_page_size = LargePageSizeInBytes;
3848 } else {
3849 warning("Setting LargePageSizeInBytes=" SIZE_FORMAT " has no effect on this OS. Default large page size is "
3850 SIZE_FORMAT "%s.",
3851 LargePageSizeInBytes,
3852 byte_size_in_proper_unit(_large_page_size), proper_unit_for_byte_size(_large_page_size));
3853 }
3854 }
3855
3856 const size_t default_page_size = (size_t)Linux::page_size();
3857 if (_large_page_size > default_page_size) {
3858 _page_sizes[0] = _large_page_size;
3859 _page_sizes[1] = default_page_size;
3860 _page_sizes[2] = 0;
3861 }
3862
3863 return _large_page_size;
3864 }
3865
3866 bool os::Linux::setup_large_page_type(size_t page_size) {
3867 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3868 FLAG_IS_DEFAULT(UseSHM) &&
3869 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3870
3871 // The type of large pages has not been specified by the user.
3872
3873 // Try UseHugeTLBFS and then UseSHM.
3874 UseHugeTLBFS = UseSHM = true;
3875
3876 // Don't try UseTransparentHugePages since there are known
3877 // performance issues with it turned on. This might change in the future.
3878 UseTransparentHugePages = false;
3879 }
3880
3881 if (UseTransparentHugePages) {
3882 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3883 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3884 UseHugeTLBFS = false;
3885 UseSHM = false;
3886 return true;
3887 }
3888 UseTransparentHugePages = false;
3889 }
3890
3891 if (UseHugeTLBFS) {
3892 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3893 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3894 UseSHM = false;
3895 return true;
3896 }
3897 UseHugeTLBFS = false;
3898 }
3899
3900 return UseSHM;
3901 }
3902
3903 void os::large_page_init() {
3904 if (!UseLargePages &&
3905 !UseTransparentHugePages &&
3906 !UseHugeTLBFS &&
3907 !UseSHM) {
3908 // Not using large pages.
3909 return;
3910 }
3911
3912 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3913 // The user explicitly turned off large pages.
3914 // Ignore the rest of the large pages flags.
3915 UseTransparentHugePages = false;
3916 UseHugeTLBFS = false;
3917 UseSHM = false;
3918 return;
3919 }
3920
3921 size_t large_page_size = Linux::setup_large_page_size();
3922 UseLargePages = Linux::setup_large_page_type(large_page_size);
3923
3924 set_coredump_filter(LARGEPAGES_BIT);
3925 }
3926
3927 #ifndef SHM_HUGETLB
3928 #define SHM_HUGETLB 04000
3929 #endif
3930
3931 #define shm_warning_format(format, ...) \
3932 do { \
3933 if (UseLargePages && \
3934 (!FLAG_IS_DEFAULT(UseLargePages) || \
3935 !FLAG_IS_DEFAULT(UseSHM) || \
3936 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3937 warning(format, __VA_ARGS__); \
3938 } \
3939 } while (0)
3940
3941 #define shm_warning(str) shm_warning_format("%s", str)
3942
3943 #define shm_warning_with_errno(str) \
3944 do { \
3945 int err = errno; \
3946 shm_warning_format(str " (error = %d)", err); \
3947 } while (0)
3948
3949 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3950 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3951
3952 if (!is_aligned(alignment, SHMLBA)) {
3953 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3954 return NULL;
3955 }
3956
3957 // To ensure that we get 'alignment' aligned memory from shmat,
3958 // we pre-reserve aligned virtual memory and then attach to that.
3959
3960 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3961 if (pre_reserved_addr == NULL) {
3962 // Couldn't pre-reserve aligned memory.
3963 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3964 return NULL;
3965 }
3966
3967 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3968 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3969
3970 if ((intptr_t)addr == -1) {
3971 int err = errno;
3972 shm_warning_with_errno("Failed to attach shared memory.");
3973
3974 assert(err != EACCES, "Unexpected error");
3975 assert(err != EIDRM, "Unexpected error");
3976 assert(err != EINVAL, "Unexpected error");
3977
3978 // Since we don't know if the kernel unmapped the pre-reserved memory area
3979 // we can't unmap it, since that would potentially unmap memory that was
3980 // mapped from other threads.
3981 return NULL;
3982 }
3983
3984 return addr;
3985 }
3986
3987 static char* shmat_at_address(int shmid, char* req_addr) {
3988 if (!is_aligned(req_addr, SHMLBA)) {
3989 assert(false, "Requested address needs to be SHMLBA aligned");
3990 return NULL;
3991 }
3992
3993 char* addr = (char*)shmat(shmid, req_addr, 0);
3994
3995 if ((intptr_t)addr == -1) {
3996 shm_warning_with_errno("Failed to attach shared memory.");
3997 return NULL;
3998 }
3999
4000 return addr;
4001 }
4002
4003 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
4004 // If a req_addr has been provided, we assume that the caller has already aligned the address.
4005 if (req_addr != NULL) {
4006 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
4007 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
4008 return shmat_at_address(shmid, req_addr);
4009 }
4010
4011 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
4012 // return large page size aligned memory addresses when req_addr == NULL.
4013 // However, if the alignment is larger than the large page size, we have
4014 // to manually ensure that the memory returned is 'alignment' aligned.
4015 if (alignment > os::large_page_size()) {
4016 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
4017 return shmat_with_alignment(shmid, bytes, alignment);
4018 } else {
4019 return shmat_at_address(shmid, NULL);
4020 }
4021 }
4022
4023 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
4024 char* req_addr, bool exec) {
4025 // "exec" is passed in but not used. Creating the shared image for
4026 // the code cache doesn't have an SHM_X executable permission to check.
4027 assert(UseLargePages && UseSHM, "only for SHM large pages");
4028 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4029 assert(is_aligned(req_addr, alignment), "Unaligned address");
4030
4031 if (!is_aligned(bytes, os::large_page_size())) {
4032 return NULL; // Fallback to small pages.
4033 }
4034
4035 // Create a large shared memory region to attach to based on size.
4036 // Currently, size is the total size of the heap.
4037 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
4038 if (shmid == -1) {
4039 // Possible reasons for shmget failure:
4040 // 1. shmmax is too small for Java heap.
4041 // > check shmmax value: cat /proc/sys/kernel/shmmax
4042 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
4043 // 2. not enough large page memory.
4044 // > check available large pages: cat /proc/meminfo
4045 // > increase amount of large pages:
4046 // echo new_value > /proc/sys/vm/nr_hugepages
4047 // Note 1: different Linux may use different name for this property,
4048 // e.g. on Redhat AS-3 it is "hugetlb_pool".
4049 // Note 2: it's possible there's enough physical memory available but
4050 // they are so fragmented after a long run that they can't
4051 // coalesce into large pages. Try to reserve large pages when
4052 // the system is still "fresh".
4053 shm_warning_with_errno("Failed to reserve shared memory.");
4054 return NULL;
4055 }
4056
4057 // Attach to the region.
4058 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
4059
4060 // Remove shmid. If shmat() is successful, the actual shared memory segment
4061 // will be deleted when it's detached by shmdt() or when the process
4062 // terminates. If shmat() is not successful this will remove the shared
4063 // segment immediately.
4064 shmctl(shmid, IPC_RMID, NULL);
4065
4066 return addr;
4067 }
4068
4069 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
4070 int error) {
4071 assert(error == ENOMEM, "Only expect to fail if no memory is available");
4072
4073 bool warn_on_failure = UseLargePages &&
4074 (!FLAG_IS_DEFAULT(UseLargePages) ||
4075 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
4076 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
4077
4078 if (warn_on_failure) {
4079 char msg[128];
4080 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
4081 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
4082 warning("%s", msg);
4083 }
4084 }
4085
4086 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
4087 char* req_addr,
4088 bool exec) {
4089 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4090 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
4091 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4092
4093 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4094 int flags = MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB;
4095
4096 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes)) {
4097 flags |= (os::large_page_size() > (1 << (MAP_HUGE_2MB >> MAP_HUGE_SHIFT)))
4098 ? MAP_HUGE_1GB
4099 : MAP_HUGE_2MB;
4100 }
4101 char* addr = (char*)::mmap(req_addr, bytes, prot, flags, -1, 0);
4102
4103 if (addr == MAP_FAILED) {
4104 warn_on_large_pages_failure(req_addr, bytes, errno);
4105 return NULL;
4106 }
4107
4108 assert(is_aligned(addr, os::large_page_size()), "Must be");
4109
4110 return addr;
4111 }
4112
4113 // Reserve memory using mmap(MAP_HUGETLB).
4114 // - bytes shall be a multiple of alignment.
4115 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4116 // - alignment sets the alignment at which memory shall be allocated.
4117 // It must be a multiple of allocation granularity.
4118 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4119 // req_addr or NULL.
4120 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4121 size_t alignment,
4122 char* req_addr,
4123 bool exec) {
4124 size_t large_page_size = os::large_page_size();
4125 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4126
4127 assert(is_aligned(req_addr, alignment), "Must be");
4128 assert(is_aligned(bytes, alignment), "Must be");
4129
4130 // First reserve - but not commit - the address range in small pages.
4131 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4132
4133 if (start == NULL) {
4134 return NULL;
4135 }
4136
4137 assert(is_aligned(start, alignment), "Must be");
4138
4139 char* end = start + bytes;
4140
4141 // Find the regions of the allocated chunk that can be promoted to large pages.
4142 char* lp_start = align_up(start, large_page_size);
4143 char* lp_end = align_down(end, large_page_size);
4144
4145 size_t lp_bytes = lp_end - lp_start;
4146
4147 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4148
4149 if (lp_bytes == 0) {
4150 // The mapped region doesn't even span the start and the end of a large page.
4151 // Fall back to allocate a non-special area.
4152 ::munmap(start, end - start);
4153 return NULL;
4154 }
4155
4156 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4157 int flags = MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED;
4158 void* result;
4159
4160 // Commit small-paged leading area.
4161 if (start != lp_start) {
4162 result = ::mmap(start, lp_start - start, prot, flags, -1, 0);
4163 if (result == MAP_FAILED) {
4164 ::munmap(lp_start, end - lp_start);
4165 return NULL;
4166 }
4167 }
4168
4169 // Commit large-paged area.
4170 flags |= MAP_HUGETLB;
4171
4172 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes)) {
4173 flags |= (os::large_page_size() > (1 << (MAP_HUGE_2MB >> MAP_HUGE_SHIFT)))
4174 ? MAP_HUGE_1GB
4175 : MAP_HUGE_2MB;
4176 }
4177
4178 result = ::mmap(lp_start, lp_bytes, prot, flags, -1, 0);
4179 if (result == MAP_FAILED) {
4180 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4181 // If the mmap above fails, the large pages region will be unmapped and we
4182 // have regions before and after with small pages. Release these regions.
4183 //
4184 // | mapped | unmapped | mapped |
4185 // ^ ^ ^ ^
4186 // start lp_start lp_end end
4187 //
4188 ::munmap(start, lp_start - start);
4189 ::munmap(lp_end, end - lp_end);
4190 return NULL;
4191 }
4192
4193 // Commit small-paged trailing area.
4194 if (lp_end != end) {
4195 result = ::mmap(lp_end, end - lp_end, prot,
4196 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4197 -1, 0);
4198 if (result == MAP_FAILED) {
4199 ::munmap(start, lp_end - start);
4200 return NULL;
4201 }
4202 }
4203
4204 return start;
4205 }
4206
4207 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4208 size_t alignment,
4209 char* req_addr,
4210 bool exec) {
4211 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4212 assert(is_aligned(req_addr, alignment), "Must be");
4213 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4214 assert(is_power_of_2(os::large_page_size()), "Must be");
4215 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4216
4217 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4218 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4219 } else {
4220 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4221 }
4222 }
4223
4224 char* os::pd_reserve_memory_special(size_t bytes, size_t alignment,
4225 char* req_addr, bool exec) {
4226 assert(UseLargePages, "only for large pages");
4227
4228 char* addr;
4229 if (UseSHM) {
4230 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4231 } else {
4232 assert(UseHugeTLBFS, "must be");
4233 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4234 }
4235
4236 if (addr != NULL) {
4237 if (UseNUMAInterleaving) {
4238 numa_make_global(addr, bytes);
4239 }
4240 }
4241
4242 return addr;
4243 }
4244
4245 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4246 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4247 return shmdt(base) == 0;
4248 }
4249
4250 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4251 return pd_release_memory(base, bytes);
4252 }
4253
4254 bool os::pd_release_memory_special(char* base, size_t bytes) {
4255 assert(UseLargePages, "only for large pages");
4256 bool res;
4257
4258 if (UseSHM) {
4259 res = os::Linux::release_memory_special_shm(base, bytes);
4260 } else {
4261 assert(UseHugeTLBFS, "must be");
4262 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4263 }
4264 return res;
4265 }
4266
4267 size_t os::large_page_size() {
4268 return _large_page_size;
4269 }
4270
4271 // With SysV SHM the entire memory region must be allocated as shared
4272 // memory.
4273 // HugeTLBFS allows application to commit large page memory on demand.
4274 // However, when committing memory with HugeTLBFS fails, the region
4275 // that was supposed to be committed will lose the old reservation
4276 // and allow other threads to steal that memory region. Because of this
4277 // behavior we can't commit HugeTLBFS memory.
4278 bool os::can_commit_large_page_memory() {
4279 return UseTransparentHugePages;
4280 }
4281
4282 bool os::can_execute_large_page_memory() {
4283 return UseTransparentHugePages || UseHugeTLBFS;
4284 }
4285
4286 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4287 assert(file_desc >= 0, "file_desc is not valid");
4288 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4289 if (result != NULL) {
4290 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4291 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4292 }
4293 }
4294 return result;
4295 }
4296
4297 // Reserve memory at an arbitrary address, only if that area is
4298 // available (and not reserved for something else).
4299
4300 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4301 // Assert only that the size is a multiple of the page size, since
4302 // that's all that mmap requires, and since that's all we really know
4303 // about at this low abstraction level. If we need higher alignment,
4304 // we can either pass an alignment to this method or verify alignment
4305 // in one of the methods further up the call chain. See bug 5044738.
4306 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4307
4308 // Repeatedly allocate blocks until the block is allocated at the
4309 // right spot.
4310
4311 // Linux mmap allows caller to pass an address as hint; give it a try first,
4312 // if kernel honors the hint then we can return immediately.
4313 char * addr = anon_mmap(requested_addr, bytes, false);
4314 if (addr == requested_addr) {
4315 return requested_addr;
4316 }
4317
4318 if (addr != NULL) {
4319 // mmap() is successful but it fails to reserve at the requested address
4320 anon_munmap(addr, bytes);
4321 }
4322
4323 return NULL;
4324 }
4325
4326 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4327 void os::infinite_sleep() {
4328 while (true) { // sleep forever ...
4329 ::sleep(100); // ... 100 seconds at a time
4330 }
4331 }
4332
4333 // Used to convert frequent JVM_Yield() to nops
4334 bool os::dont_yield() {
4335 return DontYieldALot;
4336 }
4337
4338 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4339 // actually give up the CPU. Since skip buddy (v2.6.28):
4340 //
4341 // * Sets the yielding task as skip buddy for current CPU's run queue.
4342 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4343 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4344 //
4345 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4346 // getting re-scheduled immediately.
4347 //
4348 void os::naked_yield() {
4349 sched_yield();
4350 }
4351
4352 ////////////////////////////////////////////////////////////////////////////////
4353 // thread priority support
4354
4355 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4356 // only supports dynamic priority, static priority must be zero. For real-time
4357 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4358 // However, for large multi-threaded applications, SCHED_RR is not only slower
4359 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4360 // of 5 runs - Sep 2005).
4361 //
4362 // The following code actually changes the niceness of kernel-thread/LWP. It
4363 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4364 // not the entire user process, and user level threads are 1:1 mapped to kernel
4365 // threads. It has always been the case, but could change in the future. For
4366 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4367 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4368 // (e.g., root privilege or CAP_SYS_NICE capability).
4369
4370 int os::java_to_os_priority[CriticalPriority + 1] = {
4371 19, // 0 Entry should never be used
4372
4373 4, // 1 MinPriority
4374 3, // 2
4375 2, // 3
4376
4377 1, // 4
4378 0, // 5 NormPriority
4379 -1, // 6
4380
4381 -2, // 7
4382 -3, // 8
4383 -4, // 9 NearMaxPriority
4384
4385 -5, // 10 MaxPriority
4386
4387 -5 // 11 CriticalPriority
4388 };
4389
4390 static int prio_init() {
4391 if (ThreadPriorityPolicy == 1) {
4392 if (geteuid() != 0) {
4393 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy) && !FLAG_IS_JIMAGE_RESOURCE(ThreadPriorityPolicy)) {
4394 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4395 "e.g., being the root user. If the necessary permission is not " \
4396 "possessed, changes to priority will be silently ignored.");
4397 }
4398 }
4399 }
4400 if (UseCriticalJavaThreadPriority) {
4401 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4402 }
4403 return 0;
4404 }
4405
4406 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4407 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4408
4409 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4410 return (ret == 0) ? OS_OK : OS_ERR;
4411 }
4412
4413 OSReturn os::get_native_priority(const Thread* const thread,
4414 int *priority_ptr) {
4415 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4416 *priority_ptr = java_to_os_priority[NormPriority];
4417 return OS_OK;
4418 }
4419
4420 errno = 0;
4421 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4422 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4423 }
4424
4425 ////////////////////////////////////////////////////////////////////////////////
4426 // suspend/resume support
4427
4428 // The low-level signal-based suspend/resume support is a remnant from the
4429 // old VM-suspension that used to be for java-suspension, safepoints etc,
4430 // within hotspot. Currently used by JFR's OSThreadSampler
4431 //
4432 // The remaining code is greatly simplified from the more general suspension
4433 // code that used to be used.
4434 //
4435 // The protocol is quite simple:
4436 // - suspend:
4437 // - sends a signal to the target thread
4438 // - polls the suspend state of the osthread using a yield loop
4439 // - target thread signal handler (SR_handler) sets suspend state
4440 // and blocks in sigsuspend until continued
4441 // - resume:
4442 // - sets target osthread state to continue
4443 // - sends signal to end the sigsuspend loop in the SR_handler
4444 //
4445 // Note that the SR_lock plays no role in this suspend/resume protocol,
4446 // but is checked for NULL in SR_handler as a thread termination indicator.
4447 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4448 //
4449 // Note that resume_clear_context() and suspend_save_context() are needed
4450 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4451 // which in part is used by:
4452 // - Forte Analyzer: AsyncGetCallTrace()
4453 // - StackBanging: get_frame_at_stack_banging_point()
4454
4455 static void resume_clear_context(OSThread *osthread) {
4456 osthread->set_ucontext(NULL);
4457 osthread->set_siginfo(NULL);
4458 }
4459
4460 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4461 ucontext_t* context) {
4462 osthread->set_ucontext(context);
4463 osthread->set_siginfo(siginfo);
4464 }
4465
4466 // Handler function invoked when a thread's execution is suspended or
4467 // resumed. We have to be careful that only async-safe functions are
4468 // called here (Note: most pthread functions are not async safe and
4469 // should be avoided.)
4470 //
4471 // Note: sigwait() is a more natural fit than sigsuspend() from an
4472 // interface point of view, but sigwait() prevents the signal hander
4473 // from being run. libpthread would get very confused by not having
4474 // its signal handlers run and prevents sigwait()'s use with the
4475 // mutex granting granting signal.
4476 //
4477 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4478 //
4479 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4480 // Save and restore errno to avoid confusing native code with EINTR
4481 // after sigsuspend.
4482 int old_errno = errno;
4483
4484 Thread* thread = Thread::current_or_null_safe();
4485 assert(thread != NULL, "Missing current thread in SR_handler");
4486
4487 // On some systems we have seen signal delivery get "stuck" until the signal
4488 // mask is changed as part of thread termination. Check that the current thread
4489 // has not already terminated (via SR_lock()) - else the following assertion
4490 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4491 // destructor has completed.
4492
4493 if (thread->SR_lock() == NULL) {
4494 return;
4495 }
4496
4497 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4498
4499 OSThread* osthread = thread->osthread();
4500
4501 os::SuspendResume::State current = osthread->sr.state();
4502 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4503 suspend_save_context(osthread, siginfo, context);
4504
4505 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4506 os::SuspendResume::State state = osthread->sr.suspended();
4507 if (state == os::SuspendResume::SR_SUSPENDED) {
4508 sigset_t suspend_set; // signals for sigsuspend()
4509 sigemptyset(&suspend_set);
4510 // get current set of blocked signals and unblock resume signal
4511 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4512 sigdelset(&suspend_set, SR_signum);
4513
4514 sr_semaphore.signal();
4515 // wait here until we are resumed
4516 while (1) {
4517 sigsuspend(&suspend_set);
4518
4519 os::SuspendResume::State result = osthread->sr.running();
4520 if (result == os::SuspendResume::SR_RUNNING) {
4521 sr_semaphore.signal();
4522 break;
4523 }
4524 }
4525
4526 } else if (state == os::SuspendResume::SR_RUNNING) {
4527 // request was cancelled, continue
4528 } else {
4529 ShouldNotReachHere();
4530 }
4531
4532 resume_clear_context(osthread);
4533 } else if (current == os::SuspendResume::SR_RUNNING) {
4534 // request was cancelled, continue
4535 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4536 // ignore
4537 } else {
4538 // ignore
4539 }
4540
4541 errno = old_errno;
4542 }
4543
4544 static int SR_initialize() {
4545 struct sigaction act;
4546 char *s;
4547
4548 // Get signal number to use for suspend/resume
4549 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4550 int sig = ::strtol(s, 0, 10);
4551 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4552 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4553 SR_signum = sig;
4554 } else {
4555 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4556 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4557 }
4558 }
4559
4560 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4561 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4562
4563 sigemptyset(&SR_sigset);
4564 sigaddset(&SR_sigset, SR_signum);
4565
4566 // Set up signal handler for suspend/resume
4567 act.sa_flags = SA_RESTART|SA_SIGINFO;
4568 act.sa_handler = (void (*)(int)) SR_handler;
4569
4570 // SR_signum is blocked by default.
4571 // 4528190 - We also need to block pthread restart signal (32 on all
4572 // supported Linux platforms). Note that LinuxThreads need to block
4573 // this signal for all threads to work properly. So we don't have
4574 // to use hard-coded signal number when setting up the mask.
4575 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4576
4577 if (sigaction(SR_signum, &act, 0) == -1) {
4578 return -1;
4579 }
4580
4581 // Save signal flag
4582 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4583 return 0;
4584 }
4585
4586 static int sr_notify(OSThread* osthread) {
4587 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4588 assert_status(status == 0, status, "pthread_kill");
4589 return status;
4590 }
4591
4592 // "Randomly" selected value for how long we want to spin
4593 // before bailing out on suspending a thread, also how often
4594 // we send a signal to a thread we want to resume
4595 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4596 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4597
4598 // returns true on success and false on error - really an error is fatal
4599 // but this seems the normal response to library errors
4600 static bool do_suspend(OSThread* osthread) {
4601 assert(osthread->sr.is_running(), "thread should be running");
4602 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4603
4604 // mark as suspended and send signal
4605 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4606 // failed to switch, state wasn't running?
4607 ShouldNotReachHere();
4608 return false;
4609 }
4610
4611 if (sr_notify(osthread) != 0) {
4612 ShouldNotReachHere();
4613 }
4614
4615 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4616 while (true) {
4617 if (sr_semaphore.timedwait(2)) {
4618 break;
4619 } else {
4620 // timeout
4621 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4622 if (cancelled == os::SuspendResume::SR_RUNNING) {
4623 return false;
4624 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4625 // make sure that we consume the signal on the semaphore as well
4626 sr_semaphore.wait();
4627 break;
4628 } else {
4629 ShouldNotReachHere();
4630 return false;
4631 }
4632 }
4633 }
4634
4635 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4636 return true;
4637 }
4638
4639 static void do_resume(OSThread* osthread) {
4640 assert(osthread->sr.is_suspended(), "thread should be suspended");
4641 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4642
4643 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4644 // failed to switch to WAKEUP_REQUEST
4645 ShouldNotReachHere();
4646 return;
4647 }
4648
4649 while (true) {
4650 if (sr_notify(osthread) == 0) {
4651 if (sr_semaphore.timedwait(2)) {
4652 if (osthread->sr.is_running()) {
4653 return;
4654 }
4655 }
4656 } else {
4657 ShouldNotReachHere();
4658 }
4659 }
4660
4661 guarantee(osthread->sr.is_running(), "Must be running!");
4662 }
4663
4664 ///////////////////////////////////////////////////////////////////////////////////
4665 // signal handling (except suspend/resume)
4666
4667 // This routine may be used by user applications as a "hook" to catch signals.
4668 // The user-defined signal handler must pass unrecognized signals to this
4669 // routine, and if it returns true (non-zero), then the signal handler must
4670 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4671 // routine will never retun false (zero), but instead will execute a VM panic
4672 // routine kill the process.
4673 //
4674 // If this routine returns false, it is OK to call it again. This allows
4675 // the user-defined signal handler to perform checks either before or after
4676 // the VM performs its own checks. Naturally, the user code would be making
4677 // a serious error if it tried to handle an exception (such as a null check
4678 // or breakpoint) that the VM was generating for its own correct operation.
4679 //
4680 // This routine may recognize any of the following kinds of signals:
4681 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4682 // It should be consulted by handlers for any of those signals.
4683 //
4684 // The caller of this routine must pass in the three arguments supplied
4685 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4686 // field of the structure passed to sigaction(). This routine assumes that
4687 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4688 //
4689 // Note that the VM will print warnings if it detects conflicting signal
4690 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4691 //
4692 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4693 siginfo_t* siginfo,
4694 void* ucontext,
4695 int abort_if_unrecognized);
4696
4697 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4698 assert(info != NULL && uc != NULL, "it must be old kernel");
4699 int orig_errno = errno; // Preserve errno value over signal handler.
4700 JVM_handle_linux_signal(sig, info, uc, true);
4701 errno = orig_errno;
4702 }
4703
4704
4705 // This boolean allows users to forward their own non-matching signals
4706 // to JVM_handle_linux_signal, harmlessly.
4707 bool os::Linux::signal_handlers_are_installed = false;
4708
4709 // For signal-chaining
4710 bool os::Linux::libjsig_is_loaded = false;
4711 typedef struct sigaction *(*get_signal_t)(int);
4712 get_signal_t os::Linux::get_signal_action = NULL;
4713
4714 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4715 struct sigaction *actp = NULL;
4716
4717 if (libjsig_is_loaded) {
4718 // Retrieve the old signal handler from libjsig
4719 actp = (*get_signal_action)(sig);
4720 }
4721 if (actp == NULL) {
4722 // Retrieve the preinstalled signal handler from jvm
4723 actp = os::Posix::get_preinstalled_handler(sig);
4724 }
4725
4726 return actp;
4727 }
4728
4729 static bool call_chained_handler(struct sigaction *actp, int sig,
4730 siginfo_t *siginfo, void *context) {
4731 // Call the old signal handler
4732 if (actp->sa_handler == SIG_DFL) {
4733 // It's more reasonable to let jvm treat it as an unexpected exception
4734 // instead of taking the default action.
4735 return false;
4736 } else if (actp->sa_handler != SIG_IGN) {
4737 if ((actp->sa_flags & SA_NODEFER) == 0) {
4738 // automaticlly block the signal
4739 sigaddset(&(actp->sa_mask), sig);
4740 }
4741
4742 sa_handler_t hand = NULL;
4743 sa_sigaction_t sa = NULL;
4744 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4745 // retrieve the chained handler
4746 if (siginfo_flag_set) {
4747 sa = actp->sa_sigaction;
4748 } else {
4749 hand = actp->sa_handler;
4750 }
4751
4752 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4753 actp->sa_handler = SIG_DFL;
4754 }
4755
4756 // try to honor the signal mask
4757 sigset_t oset;
4758 sigemptyset(&oset);
4759 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4760
4761 // call into the chained handler
4762 if (siginfo_flag_set) {
4763 (*sa)(sig, siginfo, context);
4764 } else {
4765 (*hand)(sig);
4766 }
4767
4768 // restore the signal mask
4769 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4770 }
4771 // Tell jvm's signal handler the signal is taken care of.
4772 return true;
4773 }
4774
4775 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4776 bool chained = false;
4777 // signal-chaining
4778 if (UseSignalChaining) {
4779 struct sigaction *actp = get_chained_signal_action(sig);
4780 if (actp != NULL) {
4781 chained = call_chained_handler(actp, sig, siginfo, context);
4782 }
4783 }
4784 return chained;
4785 }
4786
4787 // for diagnostic
4788 int sigflags[NSIG];
4789
4790 int os::Linux::get_our_sigflags(int sig) {
4791 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4792 return sigflags[sig];
4793 }
4794
4795 void os::Linux::set_our_sigflags(int sig, int flags) {
4796 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4797 if (sig > 0 && sig < NSIG) {
4798 sigflags[sig] = flags;
4799 }
4800 }
4801
4802 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4803 // Check for overwrite.
4804 struct sigaction oldAct;
4805 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4806
4807 void* oldhand = oldAct.sa_sigaction
4808 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4809 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4810 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4811 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4812 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4813 if (AllowUserSignalHandlers || !set_installed) {
4814 // Do not overwrite; user takes responsibility to forward to us.
4815 return;
4816 } else if (UseSignalChaining) {
4817 // save the old handler in jvm
4818 os::Posix::save_preinstalled_handler(sig, oldAct);
4819 // libjsig also interposes the sigaction() call below and saves the
4820 // old sigaction on it own.
4821 } else {
4822 fatal("Encountered unexpected pre-existing sigaction handler "
4823 "%#lx for signal %d.", (long)oldhand, sig);
4824 }
4825 }
4826
4827 struct sigaction sigAct;
4828 sigfillset(&(sigAct.sa_mask));
4829 sigAct.sa_handler = SIG_DFL;
4830 if (!set_installed) {
4831 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4832 } else {
4833 sigAct.sa_sigaction = signalHandler;
4834 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4835 }
4836 // Save flags, which are set by ours
4837 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4838 sigflags[sig] = sigAct.sa_flags;
4839
4840 int ret = sigaction(sig, &sigAct, &oldAct);
4841 assert(ret == 0, "check");
4842
4843 void* oldhand2 = oldAct.sa_sigaction
4844 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4845 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4846 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4847 }
4848
4849 // install signal handlers for signals that HotSpot needs to
4850 // handle in order to support Java-level exception handling.
4851
4852 void os::Linux::install_signal_handlers() {
4853 if (!signal_handlers_are_installed) {
4854 signal_handlers_are_installed = true;
4855
4856 // signal-chaining
4857 typedef void (*signal_setting_t)();
4858 signal_setting_t begin_signal_setting = NULL;
4859 signal_setting_t end_signal_setting = NULL;
4860 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4861 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4862 if (begin_signal_setting != NULL) {
4863 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4864 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4865 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4866 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4867 libjsig_is_loaded = true;
4868 assert(UseSignalChaining, "should enable signal-chaining");
4869 }
4870 if (libjsig_is_loaded) {
4871 // Tell libjsig jvm is setting signal handlers
4872 (*begin_signal_setting)();
4873 }
4874
4875 set_signal_handler(SIGSEGV, true);
4876 set_signal_handler(SIGPIPE, true);
4877 set_signal_handler(SIGBUS, true);
4878 set_signal_handler(SIGILL, true);
4879 set_signal_handler(SIGFPE, true);
4880 #if defined(PPC64)
4881 set_signal_handler(SIGTRAP, true);
4882 #endif
4883 set_signal_handler(SIGXFSZ, true);
4884
4885 if (libjsig_is_loaded) {
4886 // Tell libjsig jvm finishes setting signal handlers
4887 (*end_signal_setting)();
4888 }
4889
4890 // We don't activate signal checker if libjsig is in place, we trust ourselves
4891 // and if UserSignalHandler is installed all bets are off.
4892 // Log that signal checking is off only if -verbose:jni is specified.
4893 if (CheckJNICalls) {
4894 if (libjsig_is_loaded) {
4895 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled");
4896 check_signals = false;
4897 }
4898 if (AllowUserSignalHandlers) {
4899 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4900 check_signals = false;
4901 }
4902 }
4903 }
4904 }
4905
4906 // This is the fastest way to get thread cpu time on Linux.
4907 // Returns cpu time (user+sys) for any thread, not only for current.
4908 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4909 // It might work on 2.6.10+ with a special kernel/glibc patch.
4910 // For reference, please, see IEEE Std 1003.1-2004:
4911 // http://www.unix.org/single_unix_specification
4912
4913 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4914 struct timespec tp;
4915 int rc = os::Posix::clock_gettime(clockid, &tp);
4916 assert(rc == 0, "clock_gettime is expected to return 0 code");
4917
4918 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4919 }
4920
4921 /////
4922 // glibc on Linux platform uses non-documented flag
4923 // to indicate, that some special sort of signal
4924 // trampoline is used.
4925 // We will never set this flag, and we should
4926 // ignore this flag in our diagnostic
4927 #ifdef SIGNIFICANT_SIGNAL_MASK
4928 #undef SIGNIFICANT_SIGNAL_MASK
4929 #endif
4930 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4931
4932 static const char* get_signal_handler_name(address handler,
4933 char* buf, int buflen) {
4934 int offset = 0;
4935 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4936 if (found) {
4937 // skip directory names
4938 const char *p1, *p2;
4939 p1 = buf;
4940 size_t len = strlen(os::file_separator());
4941 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4942 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4943 } else {
4944 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4945 }
4946 return buf;
4947 }
4948
4949 static void print_signal_handler(outputStream* st, int sig,
4950 char* buf, size_t buflen) {
4951 struct sigaction sa;
4952
4953 sigaction(sig, NULL, &sa);
4954
4955 // See comment for SIGNIFICANT_SIGNAL_MASK define
4956 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4957
4958 st->print("%s: ", os::exception_name(sig, buf, buflen));
4959
4960 address handler = (sa.sa_flags & SA_SIGINFO)
4961 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4962 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4963
4964 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4965 st->print("SIG_DFL");
4966 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4967 st->print("SIG_IGN");
4968 } else {
4969 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4970 }
4971
4972 st->print(", sa_mask[0]=");
4973 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4974
4975 address rh = VMError::get_resetted_sighandler(sig);
4976 // May be, handler was resetted by VMError?
4977 if (rh != NULL) {
4978 handler = rh;
4979 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4980 }
4981
4982 st->print(", sa_flags=");
4983 os::Posix::print_sa_flags(st, sa.sa_flags);
4984
4985 // Check: is it our handler?
4986 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4987 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4988 // It is our signal handler
4989 // check for flags, reset system-used one!
4990 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4991 st->print(
4992 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4993 os::Linux::get_our_sigflags(sig));
4994 }
4995 }
4996 st->cr();
4997 }
4998
4999
5000 #define DO_SIGNAL_CHECK(sig) \
5001 do { \
5002 if (!sigismember(&check_signal_done, sig)) { \
5003 os::Linux::check_signal_handler(sig); \
5004 } \
5005 } while (0)
5006
5007 // This method is a periodic task to check for misbehaving JNI applications
5008 // under CheckJNI, we can add any periodic checks here
5009
5010 void os::run_periodic_checks() {
5011 if (check_signals == false) return;
5012
5013 // SEGV and BUS if overridden could potentially prevent
5014 // generation of hs*.log in the event of a crash, debugging
5015 // such a case can be very challenging, so we absolutely
5016 // check the following for a good measure:
5017 DO_SIGNAL_CHECK(SIGSEGV);
5018 DO_SIGNAL_CHECK(SIGILL);
5019 DO_SIGNAL_CHECK(SIGFPE);
5020 DO_SIGNAL_CHECK(SIGBUS);
5021 DO_SIGNAL_CHECK(SIGPIPE);
5022 DO_SIGNAL_CHECK(SIGXFSZ);
5023 #if defined(PPC64)
5024 DO_SIGNAL_CHECK(SIGTRAP);
5025 #endif
5026
5027 // ReduceSignalUsage allows the user to override these handlers
5028 // see comments at the very top and jvm_md.h
5029 if (!ReduceSignalUsage) {
5030 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
5031 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
5032 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
5033 DO_SIGNAL_CHECK(BREAK_SIGNAL);
5034 }
5035
5036 DO_SIGNAL_CHECK(SR_signum);
5037 }
5038
5039 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
5040
5041 static os_sigaction_t os_sigaction = NULL;
5042
5043 void os::Linux::check_signal_handler(int sig) {
5044 char buf[O_BUFLEN];
5045 address jvmHandler = NULL;
5046
5047
5048 struct sigaction act;
5049 if (os_sigaction == NULL) {
5050 // only trust the default sigaction, in case it has been interposed
5051 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
5052 if (os_sigaction == NULL) return;
5053 }
5054
5055 os_sigaction(sig, (struct sigaction*)NULL, &act);
5056
5057
5058 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
5059
5060 address thisHandler = (act.sa_flags & SA_SIGINFO)
5061 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5062 : CAST_FROM_FN_PTR(address, act.sa_handler);
5063
5064
5065 switch (sig) {
5066 case SIGSEGV:
5067 case SIGBUS:
5068 case SIGFPE:
5069 case SIGPIPE:
5070 case SIGILL:
5071 case SIGXFSZ:
5072 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5073 break;
5074
5075 case SHUTDOWN1_SIGNAL:
5076 case SHUTDOWN2_SIGNAL:
5077 case SHUTDOWN3_SIGNAL:
5078 case BREAK_SIGNAL:
5079 jvmHandler = (address)user_handler();
5080 break;
5081
5082 default:
5083 if (sig == SR_signum) {
5084 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5085 } else {
5086 return;
5087 }
5088 break;
5089 }
5090
5091 if (thisHandler != jvmHandler) {
5092 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5093 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5094 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5095 // No need to check this sig any longer
5096 sigaddset(&check_signal_done, sig);
5097 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5098 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5099 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5100 exception_name(sig, buf, O_BUFLEN));
5101 }
5102 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5103 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5104 tty->print("expected:");
5105 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5106 tty->cr();
5107 tty->print(" found:");
5108 os::Posix::print_sa_flags(tty, act.sa_flags);
5109 tty->cr();
5110 // No need to check this sig any longer
5111 sigaddset(&check_signal_done, sig);
5112 }
5113
5114 // Dump all the signal
5115 if (sigismember(&check_signal_done, sig)) {
5116 print_signal_handlers(tty, buf, O_BUFLEN);
5117 }
5118 }
5119
5120 extern void report_error(char* file_name, int line_no, char* title,
5121 char* format, ...);
5122
5123 // this is called _before_ most of the global arguments have been parsed
5124 void os::init(void) {
5125 char dummy; // used to get a guess on initial stack address
5126
5127 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5128
5129 init_random(1234567);
5130
5131 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5132 if (Linux::page_size() == -1) {
5133 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5134 os::strerror(errno));
5135 }
5136 init_page_sizes((size_t) Linux::page_size());
5137
5138 Linux::initialize_system_info();
5139
5140 os::Linux::CPUPerfTicks pticks;
5141 bool res = os::Linux::get_tick_information(&pticks, -1);
5142
5143 if (res && pticks.has_steal_ticks) {
5144 has_initial_tick_info = true;
5145 initial_total_ticks = pticks.total;
5146 initial_steal_ticks = pticks.steal;
5147 }
5148
5149 // _main_thread points to the thread that created/loaded the JVM.
5150 Linux::_main_thread = pthread_self();
5151
5152 // retrieve entry point for pthread_setname_np
5153 Linux::_pthread_setname_np =
5154 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5155
5156 os::Posix::init();
5157
5158 initial_time_count = javaTimeNanos();
5159
5160 // Always warn if no monotonic clock available
5161 if (!os::Posix::supports_monotonic_clock()) {
5162 warning("No monotonic clock was available - timed services may " \
5163 "be adversely affected if the time-of-day clock changes");
5164 }
5165 }
5166
5167 // To install functions for atexit system call
5168 extern "C" {
5169 static void perfMemory_exit_helper() {
5170 perfMemory_exit();
5171 }
5172 }
5173
5174 void os::pd_init_container_support() {
5175 OSContainer::init();
5176 }
5177
5178 void os::Linux::numa_init() {
5179
5180 // Java can be invoked as
5181 // 1. Without numactl and heap will be allocated/configured on all nodes as
5182 // per the system policy.
5183 // 2. With numactl --interleave:
5184 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5185 // API for membind case bitmask is reset.
5186 // Interleave is only hint and Kernel can fallback to other nodes if
5187 // no memory is available on the target nodes.
5188 // 3. With numactl --membind:
5189 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5190 // interleave case returns bitmask of all nodes.
5191 // numa_all_nodes_ptr holds bitmask of all nodes.
5192 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5193 // bitmask when externally configured to run on all or fewer nodes.
5194
5195 if (!Linux::libnuma_init()) {
5196 UseNUMA = false;
5197 } else {
5198 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5199 // If there's only one node (they start from 0) or if the process
5200 // is bound explicitly to a single node using membind, disable NUMA unless
5201 // user explicilty forces NUMA optimizations on single-node/UMA systems
5202 UseNUMA = ForceNUMA;
5203 } else {
5204
5205 LogTarget(Info,os) log;
5206 LogStream ls(log);
5207
5208 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5209
5210 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5211 const char* numa_mode = "membind";
5212
5213 if (Linux::is_running_in_interleave_mode()) {
5214 bmp = Linux::_numa_interleave_bitmask;
5215 numa_mode = "interleave";
5216 }
5217
5218 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5219 " Heap will be configured using NUMA memory nodes:", numa_mode);
5220
5221 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5222 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5223 ls.print(" %d", node);
5224 }
5225 }
5226 }
5227 }
5228
5229 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5230 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5231 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5232 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5233 // and disable adaptive resizing.
5234 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5235 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5236 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5237 UseAdaptiveSizePolicy = false;
5238 UseAdaptiveNUMAChunkSizing = false;
5239 }
5240 }
5241 }
5242
5243 // this is called _after_ the global arguments have been parsed
5244 jint os::init_2(void) {
5245
5246 // This could be set after os::Posix::init() but all platforms
5247 // have to set it the same so we have to mirror Solaris.
5248 DEBUG_ONLY(os::set_mutex_init_done();)
5249
5250 os::Posix::init_2();
5251
5252 Linux::fast_thread_clock_init();
5253
5254 // initialize suspend/resume support - must do this before signal_sets_init()
5255 if (SR_initialize() != 0) {
5256 perror("SR_initialize failed");
5257 return JNI_ERR;
5258 }
5259
5260 Linux::signal_sets_init();
5261 Linux::install_signal_handlers();
5262 // Initialize data for jdk.internal.misc.Signal
5263 if (!ReduceSignalUsage) {
5264 jdk_misc_signal_init();
5265 }
5266
5267 if (AdjustStackSizeForTLS) {
5268 get_minstack_init();
5269 }
5270
5271 // Check and sets minimum stack sizes against command line options
5272 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5273 return JNI_ERR;
5274 }
5275
5276 #if defined(IA32)
5277 // Need to ensure we've determined the process's initial stack to
5278 // perform the workaround
5279 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5280 workaround_expand_exec_shield_cs_limit();
5281 #else
5282 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5283 if (!suppress_primordial_thread_resolution) {
5284 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5285 }
5286 #endif
5287
5288 Linux::libpthread_init();
5289 Linux::sched_getcpu_init();
5290 log_info(os)("HotSpot is running with %s, %s",
5291 Linux::glibc_version(), Linux::libpthread_version());
5292
5293 if (UseNUMA) {
5294 Linux::numa_init();
5295 }
5296
5297 if (MaxFDLimit) {
5298 // set the number of file descriptors to max. print out error
5299 // if getrlimit/setrlimit fails but continue regardless.
5300 struct rlimit nbr_files;
5301 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5302 if (status != 0) {
5303 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5304 } else {
5305 nbr_files.rlim_cur = nbr_files.rlim_max;
5306 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5307 if (status != 0) {
5308 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5309 }
5310 }
5311 }
5312
5313 // at-exit methods are called in the reverse order of their registration.
5314 // atexit functions are called on return from main or as a result of a
5315 // call to exit(3C). There can be only 32 of these functions registered
5316 // and atexit() does not set errno.
5317
5318 if (PerfAllowAtExitRegistration) {
5319 // only register atexit functions if PerfAllowAtExitRegistration is set.
5320 // atexit functions can be delayed until process exit time, which
5321 // can be problematic for embedded VM situations. Embedded VMs should
5322 // call DestroyJavaVM() to assure that VM resources are released.
5323
5324 // note: perfMemory_exit_helper atexit function may be removed in
5325 // the future if the appropriate cleanup code can be added to the
5326 // VM_Exit VMOperation's doit method.
5327 if (atexit(perfMemory_exit_helper) != 0) {
5328 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5329 }
5330 }
5331
5332 // initialize thread priority policy
5333 prio_init();
5334
5335 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5336 set_coredump_filter(DAX_SHARED_BIT);
5337 }
5338
5339 if (DumpPrivateMappingsInCore) {
5340 set_coredump_filter(FILE_BACKED_PVT_BIT);
5341 }
5342
5343 if (DumpSharedMappingsInCore) {
5344 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5345 }
5346
5347 return JNI_OK;
5348 }
5349
5350 // older glibc versions don't have this macro (which expands to
5351 // an optimized bit-counting function) so we have to roll our own
5352 #ifndef CPU_COUNT
5353
5354 static int _cpu_count(const cpu_set_t* cpus) {
5355 int count = 0;
5356 // only look up to the number of configured processors
5357 for (int i = 0; i < os::processor_count(); i++) {
5358 if (CPU_ISSET(i, cpus)) {
5359 count++;
5360 }
5361 }
5362 return count;
5363 }
5364
5365 #define CPU_COUNT(cpus) _cpu_count(cpus)
5366
5367 #endif // CPU_COUNT
5368
5369 // Get the current number of available processors for this process.
5370 // This value can change at any time during a process's lifetime.
5371 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5372 // If it appears there may be more than 1024 processors then we do a
5373 // dynamic check - see 6515172 for details.
5374 // If anything goes wrong we fallback to returning the number of online
5375 // processors - which can be greater than the number available to the process.
5376 int os::Linux::active_processor_count() {
5377 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5378 cpu_set_t* cpus_p = &cpus;
5379 int cpus_size = sizeof(cpu_set_t);
5380
5381 int configured_cpus = os::processor_count(); // upper bound on available cpus
5382 int cpu_count = 0;
5383
5384 // old build platforms may not support dynamic cpu sets
5385 #ifdef CPU_ALLOC
5386
5387 // To enable easy testing of the dynamic path on different platforms we
5388 // introduce a diagnostic flag: UseCpuAllocPath
5389 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5390 // kernel may use a mask bigger than cpu_set_t
5391 log_trace(os)("active_processor_count: using dynamic path %s"
5392 "- configured processors: %d",
5393 UseCpuAllocPath ? "(forced) " : "",
5394 configured_cpus);
5395 cpus_p = CPU_ALLOC(configured_cpus);
5396 if (cpus_p != NULL) {
5397 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5398 // zero it just to be safe
5399 CPU_ZERO_S(cpus_size, cpus_p);
5400 }
5401 else {
5402 // failed to allocate so fallback to online cpus
5403 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5404 log_trace(os)("active_processor_count: "
5405 "CPU_ALLOC failed (%s) - using "
5406 "online processor count: %d",
5407 os::strerror(errno), online_cpus);
5408 return online_cpus;
5409 }
5410 }
5411 else {
5412 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5413 configured_cpus);
5414 }
5415 #else // CPU_ALLOC
5416 // these stubs won't be executed
5417 #define CPU_COUNT_S(size, cpus) -1
5418 #define CPU_FREE(cpus)
5419
5420 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5421 configured_cpus);
5422 #endif // CPU_ALLOC
5423
5424 // pid 0 means the current thread - which we have to assume represents the process
5425 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5426 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5427 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5428 }
5429 else {
5430 cpu_count = CPU_COUNT(cpus_p);
5431 }
5432 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5433 }
5434 else {
5435 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5436 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5437 "which may exceed available processors", os::strerror(errno), cpu_count);
5438 }
5439
5440 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5441 CPU_FREE(cpus_p);
5442 }
5443
5444 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5445 return cpu_count;
5446 }
5447
5448 // Determine the active processor count from one of
5449 // three different sources:
5450 //
5451 // 1. User option -XX:ActiveProcessorCount
5452 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5453 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5454 //
5455 // Option 1, if specified, will always override.
5456 // If the cgroup subsystem is active and configured, we
5457 // will return the min of the cgroup and option 2 results.
5458 // This is required since tools, such as numactl, that
5459 // alter cpu affinity do not update cgroup subsystem
5460 // cpuset configuration files.
5461 int os::active_processor_count() {
5462 // User has overridden the number of active processors
5463 if (ActiveProcessorCount > 0) {
5464 log_trace(os)("active_processor_count: "
5465 "active processor count set by user : %d",
5466 ActiveProcessorCount);
5467 return ActiveProcessorCount;
5468 }
5469
5470 int active_cpus;
5471 if (OSContainer::is_containerized()) {
5472 active_cpus = OSContainer::active_processor_count();
5473 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5474 active_cpus);
5475 } else {
5476 active_cpus = os::Linux::active_processor_count();
5477 }
5478
5479 return active_cpus;
5480 }
5481
5482 uint os::processor_id() {
5483 const int id = Linux::sched_getcpu();
5484 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5485 return (uint)id;
5486 }
5487
5488 void os::set_native_thread_name(const char *name) {
5489 if (Linux::_pthread_setname_np) {
5490 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5491 snprintf(buf, sizeof(buf), "%s", name);
5492 buf[sizeof(buf) - 1] = '\0';
5493 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5494 // ERANGE should not happen; all other errors should just be ignored.
5495 assert(rc != ERANGE, "pthread_setname_np failed");
5496 }
5497 }
5498
5499 bool os::bind_to_processor(uint processor_id) {
5500 // Not yet implemented.
5501 return false;
5502 }
5503
5504 ///
5505
5506 void os::SuspendedThreadTask::internal_do_task() {
5507 if (do_suspend(_thread->osthread())) {
5508 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5509 do_task(context);
5510 do_resume(_thread->osthread());
5511 }
5512 }
5513
5514 ////////////////////////////////////////////////////////////////////////////////
5515 // debug support
5516
5517 bool os::find(address addr, outputStream* st) {
5518 Dl_info dlinfo;
5519 memset(&dlinfo, 0, sizeof(dlinfo));
5520 if (dladdr(addr, &dlinfo) != 0) {
5521 st->print(PTR_FORMAT ": ", p2i(addr));
5522 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5523 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5524 p2i(addr) - p2i(dlinfo.dli_saddr));
5525 } else if (dlinfo.dli_fbase != NULL) {
5526 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5527 } else {
5528 st->print("<absolute address>");
5529 }
5530 if (dlinfo.dli_fname != NULL) {
5531 st->print(" in %s", dlinfo.dli_fname);
5532 }
5533 if (dlinfo.dli_fbase != NULL) {
5534 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5535 }
5536 st->cr();
5537
5538 if (Verbose) {
5539 // decode some bytes around the PC
5540 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5541 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5542 address lowest = (address) dlinfo.dli_sname;
5543 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5544 if (begin < lowest) begin = lowest;
5545 Dl_info dlinfo2;
5546 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5547 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5548 end = (address) dlinfo2.dli_saddr;
5549 }
5550 Disassembler::decode(begin, end, st);
5551 }
5552 return true;
5553 }
5554 return false;
5555 }
5556
5557 ////////////////////////////////////////////////////////////////////////////////
5558 // misc
5559
5560 // This does not do anything on Linux. This is basically a hook for being
5561 // able to use structured exception handling (thread-local exception filters)
5562 // on, e.g., Win32.
5563 void
5564 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5565 JavaCallArguments* args, Thread* thread) {
5566 f(value, method, args, thread);
5567 }
5568
5569 void os::print_statistics() {
5570 }
5571
5572 bool os::message_box(const char* title, const char* message) {
5573 int i;
5574 fdStream err(defaultStream::error_fd());
5575 for (i = 0; i < 78; i++) err.print_raw("=");
5576 err.cr();
5577 err.print_raw_cr(title);
5578 for (i = 0; i < 78; i++) err.print_raw("-");
5579 err.cr();
5580 err.print_raw_cr(message);
5581 for (i = 0; i < 78; i++) err.print_raw("=");
5582 err.cr();
5583
5584 char buf[16];
5585 // Prevent process from exiting upon "read error" without consuming all CPU
5586 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5587
5588 return buf[0] == 'y' || buf[0] == 'Y';
5589 }
5590
5591 // Is a (classpath) directory empty?
5592 bool os::dir_is_empty(const char* path) {
5593 DIR *dir = NULL;
5594 struct dirent *ptr;
5595
5596 dir = opendir(path);
5597 if (dir == NULL) return true;
5598
5599 // Scan the directory
5600 bool result = true;
5601 while (result && (ptr = readdir(dir)) != NULL) {
5602 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5603 result = false;
5604 }
5605 }
5606 closedir(dir);
5607 return result;
5608 }
5609
5610 // This code originates from JDK's sysOpen and open64_w
5611 // from src/solaris/hpi/src/system_md.c
5612
5613 int os::open(const char *path, int oflag, int mode) {
5614 if (strlen(path) > MAX_PATH - 1) {
5615 errno = ENAMETOOLONG;
5616 return -1;
5617 }
5618
5619 // All file descriptors that are opened in the Java process and not
5620 // specifically destined for a subprocess should have the close-on-exec
5621 // flag set. If we don't set it, then careless 3rd party native code
5622 // might fork and exec without closing all appropriate file descriptors
5623 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5624 // turn might:
5625 //
5626 // - cause end-of-file to fail to be detected on some file
5627 // descriptors, resulting in mysterious hangs, or
5628 //
5629 // - might cause an fopen in the subprocess to fail on a system
5630 // suffering from bug 1085341.
5631 //
5632 // (Yes, the default setting of the close-on-exec flag is a Unix
5633 // design flaw)
5634 //
5635 // See:
5636 // 1085341: 32-bit stdio routines should support file descriptors >255
5637 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5638 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5639 //
5640 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5641 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5642 // because it saves a system call and removes a small window where the flag
5643 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5644 // and we fall back to using FD_CLOEXEC (see below).
5645 #ifdef O_CLOEXEC
5646 oflag |= O_CLOEXEC;
5647 #endif
5648
5649 int fd = ::open64(path, oflag, mode);
5650 if (fd == -1) return -1;
5651
5652 //If the open succeeded, the file might still be a directory
5653 {
5654 struct stat64 buf64;
5655 int ret = ::fstat64(fd, &buf64);
5656 int st_mode = buf64.st_mode;
5657
5658 if (ret != -1) {
5659 if ((st_mode & S_IFMT) == S_IFDIR) {
5660 errno = EISDIR;
5661 ::close(fd);
5662 return -1;
5663 }
5664 } else {
5665 ::close(fd);
5666 return -1;
5667 }
5668 }
5669
5670 #ifdef FD_CLOEXEC
5671 // Validate that the use of the O_CLOEXEC flag on open above worked.
5672 // With recent kernels, we will perform this check exactly once.
5673 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5674 if (!O_CLOEXEC_is_known_to_work) {
5675 int flags = ::fcntl(fd, F_GETFD);
5676 if (flags != -1) {
5677 if ((flags & FD_CLOEXEC) != 0)
5678 O_CLOEXEC_is_known_to_work = 1;
5679 else
5680 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5681 }
5682 }
5683 #endif
5684
5685 return fd;
5686 }
5687
5688
5689 // create binary file, rewriting existing file if required
5690 int os::create_binary_file(const char* path, bool rewrite_existing) {
5691 int oflags = O_WRONLY | O_CREAT;
5692 if (!rewrite_existing) {
5693 oflags |= O_EXCL;
5694 }
5695 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5696 }
5697
5698 // return current position of file pointer
5699 jlong os::current_file_offset(int fd) {
5700 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5701 }
5702
5703 // move file pointer to the specified offset
5704 jlong os::seek_to_file_offset(int fd, jlong offset) {
5705 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5706 }
5707
5708 // This code originates from JDK's sysAvailable
5709 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5710
5711 int os::available(int fd, jlong *bytes) {
5712 jlong cur, end;
5713 int mode;
5714 struct stat64 buf64;
5715
5716 if (::fstat64(fd, &buf64) >= 0) {
5717 mode = buf64.st_mode;
5718 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5719 int n;
5720 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5721 *bytes = n;
5722 return 1;
5723 }
5724 }
5725 }
5726 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5727 return 0;
5728 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5729 return 0;
5730 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5731 return 0;
5732 }
5733 *bytes = end - cur;
5734 return 1;
5735 }
5736
5737 // Map a block of memory.
5738 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5739 char *addr, size_t bytes, bool read_only,
5740 bool allow_exec) {
5741 int prot;
5742 int flags = MAP_PRIVATE;
5743
5744 if (read_only) {
5745 prot = PROT_READ;
5746 } else {
5747 prot = PROT_READ | PROT_WRITE;
5748 }
5749
5750 if (allow_exec) {
5751 prot |= PROT_EXEC;
5752 }
5753
5754 if (addr != NULL) {
5755 flags |= MAP_FIXED;
5756 }
5757
5758 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5759 fd, file_offset);
5760 if (mapped_address == MAP_FAILED) {
5761 return NULL;
5762 }
5763 return mapped_address;
5764 }
5765
5766
5767 // Remap a block of memory.
5768 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5769 char *addr, size_t bytes, bool read_only,
5770 bool allow_exec) {
5771 // same as map_memory() on this OS
5772 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5773 allow_exec);
5774 }
5775
5776
5777 // Unmap a block of memory.
5778 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5779 return munmap(addr, bytes) == 0;
5780 }
5781
5782 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5783
5784 static jlong fast_cpu_time(Thread *thread) {
5785 clockid_t clockid;
5786 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5787 &clockid);
5788 if (rc == 0) {
5789 return os::Linux::fast_thread_cpu_time(clockid);
5790 } else {
5791 // It's possible to encounter a terminated native thread that failed
5792 // to detach itself from the VM - which should result in ESRCH.
5793 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5794 return -1;
5795 }
5796 }
5797
5798 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5799 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5800 // of a thread.
5801 //
5802 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5803 // the fast estimate available on the platform.
5804
5805 jlong os::current_thread_cpu_time() {
5806 if (os::Linux::supports_fast_thread_cpu_time()) {
5807 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5808 } else {
5809 // return user + sys since the cost is the same
5810 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5811 }
5812 }
5813
5814 jlong os::thread_cpu_time(Thread* thread) {
5815 // consistent with what current_thread_cpu_time() returns
5816 if (os::Linux::supports_fast_thread_cpu_time()) {
5817 return fast_cpu_time(thread);
5818 } else {
5819 return slow_thread_cpu_time(thread, true /* user + sys */);
5820 }
5821 }
5822
5823 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5824 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5825 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5826 } else {
5827 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5828 }
5829 }
5830
5831 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5832 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5833 return fast_cpu_time(thread);
5834 } else {
5835 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5836 }
5837 }
5838
5839 // -1 on error.
5840 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5841 pid_t tid = thread->osthread()->thread_id();
5842 char *s;
5843 char stat[2048];
5844 int statlen;
5845 char proc_name[64];
5846 int count;
5847 long sys_time, user_time;
5848 char cdummy;
5849 int idummy;
5850 long ldummy;
5851 FILE *fp;
5852
5853 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5854 fp = fopen(proc_name, "r");
5855 if (fp == NULL) return -1;
5856 statlen = fread(stat, 1, 2047, fp);
5857 stat[statlen] = '\0';
5858 fclose(fp);
5859
5860 // Skip pid and the command string. Note that we could be dealing with
5861 // weird command names, e.g. user could decide to rename java launcher
5862 // to "java 1.4.2 :)", then the stat file would look like
5863 // 1234 (java 1.4.2 :)) R ... ...
5864 // We don't really need to know the command string, just find the last
5865 // occurrence of ")" and then start parsing from there. See bug 4726580.
5866 s = strrchr(stat, ')');
5867 if (s == NULL) return -1;
5868
5869 // Skip blank chars
5870 do { s++; } while (s && isspace(*s));
5871
5872 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5873 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5874 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5875 &user_time, &sys_time);
5876 if (count != 13) return -1;
5877 if (user_sys_cpu_time) {
5878 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5879 } else {
5880 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5881 }
5882 }
5883
5884 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5885 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5886 info_ptr->may_skip_backward = false; // elapsed time not wall time
5887 info_ptr->may_skip_forward = false; // elapsed time not wall time
5888 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5889 }
5890
5891 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5892 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5893 info_ptr->may_skip_backward = false; // elapsed time not wall time
5894 info_ptr->may_skip_forward = false; // elapsed time not wall time
5895 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5896 }
5897
5898 bool os::is_thread_cpu_time_supported() {
5899 return true;
5900 }
5901
5902 // System loadavg support. Returns -1 if load average cannot be obtained.
5903 // Linux doesn't yet have a (official) notion of processor sets,
5904 // so just return the system wide load average.
5905 int os::loadavg(double loadavg[], int nelem) {
5906 return ::getloadavg(loadavg, nelem);
5907 }
5908
5909 void os::pause() {
5910 char filename[MAX_PATH];
5911 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5912 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5913 } else {
5914 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5915 }
5916
5917 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5918 if (fd != -1) {
5919 struct stat buf;
5920 ::close(fd);
5921 while (::stat(filename, &buf) == 0) {
5922 (void)::poll(NULL, 0, 100);
5923 }
5924 } else {
5925 jio_fprintf(stderr,
5926 "Could not open pause file '%s', continuing immediately.\n", filename);
5927 }
5928 }
5929
5930 extern char** environ;
5931
5932 // Run the specified command in a separate process. Return its exit value,
5933 // or -1 on failure (e.g. can't fork a new process).
5934 // Unlike system(), this function can be called from signal handler. It
5935 // doesn't block SIGINT et al.
5936 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5937 const char * argv[4] = {"sh", "-c", cmd, NULL};
5938
5939 pid_t pid ;
5940
5941 if (use_vfork_if_available) {
5942 pid = vfork();
5943 } else {
5944 pid = fork();
5945 }
5946
5947 if (pid < 0) {
5948 // fork failed
5949 return -1;
5950
5951 } else if (pid == 0) {
5952 // child process
5953
5954 execve("/bin/sh", (char* const*)argv, environ);
5955
5956 // execve failed
5957 _exit(-1);
5958
5959 } else {
5960 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5961 // care about the actual exit code, for now.
5962
5963 int status;
5964
5965 // Wait for the child process to exit. This returns immediately if
5966 // the child has already exited. */
5967 while (waitpid(pid, &status, 0) < 0) {
5968 switch (errno) {
5969 case ECHILD: return 0;
5970 case EINTR: break;
5971 default: return -1;
5972 }
5973 }
5974
5975 if (WIFEXITED(status)) {
5976 // The child exited normally; get its exit code.
5977 return WEXITSTATUS(status);
5978 } else if (WIFSIGNALED(status)) {
5979 // The child exited because of a signal
5980 // The best value to return is 0x80 + signal number,
5981 // because that is what all Unix shells do, and because
5982 // it allows callers to distinguish between process exit and
5983 // process death by signal.
5984 return 0x80 + WTERMSIG(status);
5985 } else {
5986 // Unknown exit code; pass it through
5987 return status;
5988 }
5989 }
5990 }
5991
5992 // Get the default path to the core file
5993 // Returns the length of the string
5994 int os::get_core_path(char* buffer, size_t bufferSize) {
5995 /*
5996 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5997 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5998 */
5999 const int core_pattern_len = 129;
6000 char core_pattern[core_pattern_len] = {0};
6001
6002 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6003 if (core_pattern_file == -1) {
6004 return -1;
6005 }
6006
6007 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6008 ::close(core_pattern_file);
6009 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6010 return -1;
6011 }
6012 if (core_pattern[ret-1] == '\n') {
6013 core_pattern[ret-1] = '\0';
6014 } else {
6015 core_pattern[ret] = '\0';
6016 }
6017
6018 // Replace the %p in the core pattern with the process id. NOTE: we do this
6019 // only if the pattern doesn't start with "|", and we support only one %p in
6020 // the pattern.
6021 char *pid_pos = strstr(core_pattern, "%p");
6022 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
6023 int written;
6024
6025 if (core_pattern[0] == '/') {
6026 if (pid_pos != NULL) {
6027 *pid_pos = '\0';
6028 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
6029 current_process_id(), tail);
6030 } else {
6031 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6032 }
6033 } else {
6034 char cwd[PATH_MAX];
6035
6036 const char* p = get_current_directory(cwd, PATH_MAX);
6037 if (p == NULL) {
6038 return -1;
6039 }
6040
6041 if (core_pattern[0] == '|') {
6042 written = jio_snprintf(buffer, bufferSize,
6043 "\"%s\" (or dumping to %s/core.%d)",
6044 &core_pattern[1], p, current_process_id());
6045 } else if (pid_pos != NULL) {
6046 *pid_pos = '\0';
6047 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6048 current_process_id(), tail);
6049 } else {
6050 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6051 }
6052 }
6053
6054 if (written < 0) {
6055 return -1;
6056 }
6057
6058 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6059 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6060
6061 if (core_uses_pid_file != -1) {
6062 char core_uses_pid = 0;
6063 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6064 ::close(core_uses_pid_file);
6065
6066 if (core_uses_pid == '1') {
6067 jio_snprintf(buffer + written, bufferSize - written,
6068 ".%d", current_process_id());
6069 }
6070 }
6071 }
6072
6073 return strlen(buffer);
6074 }
6075
6076 bool os::start_debugging(char *buf, int buflen) {
6077 int len = (int)strlen(buf);
6078 char *p = &buf[len];
6079
6080 jio_snprintf(p, buflen-len,
6081 "\n\n"
6082 "Do you want to debug the problem?\n\n"
6083 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6084 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6085 "Otherwise, press RETURN to abort...",
6086 os::current_process_id(), os::current_process_id(),
6087 os::current_thread_id(), os::current_thread_id());
6088
6089 bool yes = os::message_box("Unexpected Error", buf);
6090
6091 if (yes) {
6092 // yes, user asked VM to launch debugger
6093 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6094 os::current_process_id(), os::current_process_id());
6095
6096 os::fork_and_exec(buf);
6097 yes = false;
6098 }
6099 return yes;
6100 }
6101
6102
6103 // Java/Compiler thread:
6104 //
6105 // Low memory addresses
6106 // P0 +------------------------+
6107 // | |\ Java thread created by VM does not have glibc
6108 // | glibc guard page | - guard page, attached Java thread usually has
6109 // | |/ 1 glibc guard page.
6110 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6111 // | |\
6112 // | HotSpot Guard Pages | - red, yellow and reserved pages
6113 // | |/
6114 // +------------------------+ JavaThread::stack_reserved_zone_base()
6115 // | |\
6116 // | Normal Stack | -
6117 // | |/
6118 // P2 +------------------------+ Thread::stack_base()
6119 //
6120 // Non-Java thread:
6121 //
6122 // Low memory addresses
6123 // P0 +------------------------+
6124 // | |\
6125 // | glibc guard page | - usually 1 page
6126 // | |/
6127 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6128 // | |\
6129 // | Normal Stack | -
6130 // | |/
6131 // P2 +------------------------+ Thread::stack_base()
6132 //
6133 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6134 // returned from pthread_attr_getstack().
6135 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6136 // of the stack size given in pthread_attr. We work around this for
6137 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6138 //
6139 #ifndef ZERO
6140 static void current_stack_region(address * bottom, size_t * size) {
6141 if (os::is_primordial_thread()) {
6142 // primordial thread needs special handling because pthread_getattr_np()
6143 // may return bogus value.
6144 *bottom = os::Linux::initial_thread_stack_bottom();
6145 *size = os::Linux::initial_thread_stack_size();
6146 } else {
6147 pthread_attr_t attr;
6148
6149 int rslt = pthread_getattr_np(pthread_self(), &attr);
6150
6151 // JVM needs to know exact stack location, abort if it fails
6152 if (rslt != 0) {
6153 if (rslt == ENOMEM) {
6154 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6155 } else {
6156 fatal("pthread_getattr_np failed with error = %d", rslt);
6157 }
6158 }
6159
6160 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6161 fatal("Cannot locate current stack attributes!");
6162 }
6163
6164 // Work around NPTL stack guard error.
6165 size_t guard_size = 0;
6166 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6167 if (rslt != 0) {
6168 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6169 }
6170 *bottom += guard_size;
6171 *size -= guard_size;
6172
6173 pthread_attr_destroy(&attr);
6174
6175 }
6176 assert(os::current_stack_pointer() >= *bottom &&
6177 os::current_stack_pointer() < *bottom + *size, "just checking");
6178 }
6179
6180 address os::current_stack_base() {
6181 address bottom;
6182 size_t size;
6183 current_stack_region(&bottom, &size);
6184 return (bottom + size);
6185 }
6186
6187 size_t os::current_stack_size() {
6188 // This stack size includes the usable stack and HotSpot guard pages
6189 // (for the threads that have Hotspot guard pages).
6190 address bottom;
6191 size_t size;
6192 current_stack_region(&bottom, &size);
6193 return size;
6194 }
6195 #endif
6196
6197 static inline struct timespec get_mtime(const char* filename) {
6198 struct stat st;
6199 int ret = os::stat(filename, &st);
6200 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6201 return st.st_mtim;
6202 }
6203
6204 int os::compare_file_modified_times(const char* file1, const char* file2) {
6205 struct timespec filetime1 = get_mtime(file1);
6206 struct timespec filetime2 = get_mtime(file2);
6207 int diff = filetime1.tv_sec - filetime2.tv_sec;
6208 if (diff == 0) {
6209 return filetime1.tv_nsec - filetime2.tv_nsec;
6210 }
6211 return diff;
6212 }
6213
6214 bool os::supports_map_sync() {
6215 return true;
6216 }
6217
6218 /////////////// Unit tests ///////////////
6219
6220 #ifndef PRODUCT
6221
6222 class TestReserveMemorySpecial : AllStatic {
6223 public:
6224 static void small_page_write(void* addr, size_t size) {
6225 size_t page_size = os::vm_page_size();
6226
6227 char* end = (char*)addr + size;
6228 for (char* p = (char*)addr; p < end; p += page_size) {
6229 *p = 1;
6230 }
6231 }
6232
6233 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6234 if (!UseHugeTLBFS) {
6235 return;
6236 }
6237
6238 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6239
6240 if (addr != NULL) {
6241 small_page_write(addr, size);
6242
6243 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6244 }
6245 }
6246
6247 static void test_reserve_memory_special_huge_tlbfs_only() {
6248 if (!UseHugeTLBFS) {
6249 return;
6250 }
6251
6252 size_t lp = os::large_page_size();
6253
6254 for (size_t size = lp; size <= lp * 10; size += lp) {
6255 test_reserve_memory_special_huge_tlbfs_only(size);
6256 }
6257 }
6258
6259 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6260 size_t lp = os::large_page_size();
6261 size_t ag = os::vm_allocation_granularity();
6262
6263 // sizes to test
6264 const size_t sizes[] = {
6265 lp, lp + ag, lp + lp / 2, lp * 2,
6266 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6267 lp * 10, lp * 10 + lp / 2
6268 };
6269 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6270
6271 // For each size/alignment combination, we test three scenarios:
6272 // 1) with req_addr == NULL
6273 // 2) with a non-null req_addr at which we expect to successfully allocate
6274 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6275 // expect the allocation to either fail or to ignore req_addr
6276
6277 // Pre-allocate two areas; they shall be as large as the largest allocation
6278 // and aligned to the largest alignment we will be testing.
6279 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6280 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6281 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6282 -1, 0);
6283 assert(mapping1 != MAP_FAILED, "should work");
6284
6285 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6286 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6287 -1, 0);
6288 assert(mapping2 != MAP_FAILED, "should work");
6289
6290 // Unmap the first mapping, but leave the second mapping intact: the first
6291 // mapping will serve as a value for a "good" req_addr (case 2). The second
6292 // mapping, still intact, as "bad" req_addr (case 3).
6293 ::munmap(mapping1, mapping_size);
6294
6295 // Case 1
6296 for (int i = 0; i < num_sizes; i++) {
6297 const size_t size = sizes[i];
6298 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6299 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6300 if (p != NULL) {
6301 assert(is_aligned(p, alignment), "must be");
6302 small_page_write(p, size);
6303 os::Linux::release_memory_special_huge_tlbfs(p, size);
6304 }
6305 }
6306 }
6307
6308 // Case 2
6309 for (int i = 0; i < num_sizes; i++) {
6310 const size_t size = sizes[i];
6311 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6312 char* const req_addr = align_up(mapping1, alignment);
6313 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6314 if (p != NULL) {
6315 assert(p == req_addr, "must be");
6316 small_page_write(p, size);
6317 os::Linux::release_memory_special_huge_tlbfs(p, size);
6318 }
6319 }
6320 }
6321
6322 // Case 3
6323 for (int i = 0; i < num_sizes; i++) {
6324 const size_t size = sizes[i];
6325 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6326 char* const req_addr = align_up(mapping2, alignment);
6327 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6328 // as the area around req_addr contains already existing mappings, the API should always
6329 // return NULL (as per contract, it cannot return another address)
6330 assert(p == NULL, "must be");
6331 }
6332 }
6333
6334 ::munmap(mapping2, mapping_size);
6335
6336 }
6337
6338 static void test_reserve_memory_special_huge_tlbfs() {
6339 if (!UseHugeTLBFS) {
6340 return;
6341 }
6342
6343 test_reserve_memory_special_huge_tlbfs_only();
6344 test_reserve_memory_special_huge_tlbfs_mixed();
6345 }
6346
6347 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6348 if (!UseSHM) {
6349 return;
6350 }
6351
6352 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6353
6354 if (addr != NULL) {
6355 assert(is_aligned(addr, alignment), "Check");
6356 assert(is_aligned(addr, os::large_page_size()), "Check");
6357
6358 small_page_write(addr, size);
6359
6360 os::Linux::release_memory_special_shm(addr, size);
6361 }
6362 }
6363
6364 static void test_reserve_memory_special_shm() {
6365 size_t lp = os::large_page_size();
6366 size_t ag = os::vm_allocation_granularity();
6367
6368 for (size_t size = ag; size < lp * 3; size += ag) {
6369 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6370 test_reserve_memory_special_shm(size, alignment);
6371 }
6372 }
6373 }
6374
6375 static void test() {
6376 test_reserve_memory_special_huge_tlbfs();
6377 test_reserve_memory_special_shm();
6378 }
6379 };
6380
6381 void TestReserveMemorySpecial_test() {
6382 TestReserveMemorySpecial::test();
6383 }
6384
6385 #endif