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 struct loaded_modules_info_param {
2078 os::LoadedModulesCallbackFunc callback;
2079 void *param;
2080 };
2081
2082 static int dl_iterate_callback(struct dl_phdr_info *info, size_t size, void *data) {
2083 if ((info->dlpi_name == NULL) || (*info->dlpi_name == '\0')) {
2084 return 0;
2085 }
2086
2087 struct loaded_modules_info_param *callback_param = reinterpret_cast<struct loaded_modules_info_param *>(data);
2088 address base = NULL;
2089 address top = NULL;
2090 for (int idx = 0; idx < info->dlpi_phnum; idx++) {
2091 const ElfW(Phdr) *phdr = info->dlpi_phdr + idx;
2092 if (phdr->p_type == PT_LOAD) {
2093 address raw_phdr_base = reinterpret_cast<address>(info->dlpi_addr + phdr->p_vaddr);
2094
2095 address phdr_base = align_down(raw_phdr_base, phdr->p_align);
2096 if ((base == NULL) && (base < phdr_base)) {
2097 base = phdr_base;
2098 }
2099
2100 address phdr_top = align_up(raw_phdr_base + phdr->p_memsz, phdr->p_align);
2101 if (top < phdr_top) {
2102 top = phdr_top;
2103 }
2104 }
2105 }
2106
2107 return callback_param->callback(info->dlpi_name, base, top, callback_param->param);
2108 }
2109
2110 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2111 struct loaded_modules_info_param callback_param = {callback, param};
2112 return dl_iterate_phdr(&dl_iterate_callback, &callback_param);
2113 }
2114
2115 void os::print_os_info_brief(outputStream* st) {
2116 os::Linux::print_distro_info(st);
2117
2118 os::Posix::print_uname_info(st);
2119
2120 os::Linux::print_libversion_info(st);
2121
2122 }
2123
2124 void os::print_os_info(outputStream* st) {
2125 st->print("OS:");
2126
2127 os::Linux::print_distro_info(st);
2128
2129 os::Posix::print_uname_info(st);
2130
2131 os::Linux::print_uptime_info(st);
2132
2133 // Print warning if unsafe chroot environment detected
2134 if (unsafe_chroot_detected) {
2135 st->print("WARNING!! ");
2136 st->print_cr("%s", unstable_chroot_error);
2137 }
2138
2139 os::Linux::print_libversion_info(st);
2140
2141 os::Posix::print_rlimit_info(st);
2142
2143 os::Posix::print_load_average(st);
2144
2145 os::Linux::print_full_memory_info(st);
2146
2147 os::Linux::print_proc_sys_info(st);
2148
2149 os::Linux::print_ld_preload_file(st);
2150
2151 os::Linux::print_container_info(st);
2152
2153 VM_Version::print_platform_virtualization_info(st);
2154
2155 os::Linux::print_steal_info(st);
2156 }
2157
2158 // Try to identify popular distros.
2159 // Most Linux distributions have a /etc/XXX-release file, which contains
2160 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2161 // file that also contains the OS version string. Some have more than one
2162 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2163 // /etc/redhat-release.), so the order is important.
2164 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2165 // their own specific XXX-release file as well as a redhat-release file.
2166 // Because of this the XXX-release file needs to be searched for before the
2167 // redhat-release file.
2168 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2169 // search for redhat-release / SuSE-release needs to be before lsb-release.
2170 // Since the lsb-release file is the new standard it needs to be searched
2171 // before the older style release files.
2172 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2173 // next to last resort. The os-release file is a new standard that contains
2174 // distribution information and the system-release file seems to be an old
2175 // standard that has been replaced by the lsb-release and os-release files.
2176 // Searching for the debian_version file is the last resort. It contains
2177 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2178 // "Debian " is printed before the contents of the debian_version file.
2179
2180 const char* distro_files[] = {
2181 "/etc/oracle-release",
2182 "/etc/mandriva-release",
2183 "/etc/mandrake-release",
2184 "/etc/sun-release",
2185 "/etc/redhat-release",
2186 "/etc/SuSE-release",
2187 "/etc/lsb-release",
2188 "/etc/turbolinux-release",
2189 "/etc/gentoo-release",
2190 "/etc/ltib-release",
2191 "/etc/angstrom-version",
2192 "/etc/system-release",
2193 "/etc/os-release",
2194 NULL };
2195
2196 void os::Linux::print_distro_info(outputStream* st) {
2197 for (int i = 0;; i++) {
2198 const char* file = distro_files[i];
2199 if (file == NULL) {
2200 break; // done
2201 }
2202 // If file prints, we found it.
2203 if (_print_ascii_file(file, st)) {
2204 return;
2205 }
2206 }
2207
2208 if (file_exists("/etc/debian_version")) {
2209 st->print("Debian ");
2210 _print_ascii_file("/etc/debian_version", st);
2211 } else {
2212 st->print("Linux");
2213 }
2214 st->cr();
2215 }
2216
2217 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2218 char buf[256];
2219 while (fgets(buf, sizeof(buf), fp)) {
2220 // Edit out extra stuff in expected format
2221 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2222 char* ptr = strstr(buf, "\""); // the name is in quotes
2223 if (ptr != NULL) {
2224 ptr++; // go beyond first quote
2225 char* nl = strchr(ptr, '\"');
2226 if (nl != NULL) *nl = '\0';
2227 strncpy(distro, ptr, length);
2228 } else {
2229 ptr = strstr(buf, "=");
2230 ptr++; // go beyond equals then
2231 char* nl = strchr(ptr, '\n');
2232 if (nl != NULL) *nl = '\0';
2233 strncpy(distro, ptr, length);
2234 }
2235 return;
2236 } else if (get_first_line) {
2237 char* nl = strchr(buf, '\n');
2238 if (nl != NULL) *nl = '\0';
2239 strncpy(distro, buf, length);
2240 return;
2241 }
2242 }
2243 // print last line and close
2244 char* nl = strchr(buf, '\n');
2245 if (nl != NULL) *nl = '\0';
2246 strncpy(distro, buf, length);
2247 }
2248
2249 static void parse_os_info(char* distro, size_t length, const char* file) {
2250 FILE* fp = fopen(file, "r");
2251 if (fp != NULL) {
2252 // if suse format, print out first line
2253 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2254 parse_os_info_helper(fp, distro, length, get_first_line);
2255 fclose(fp);
2256 }
2257 }
2258
2259 void os::get_summary_os_info(char* buf, size_t buflen) {
2260 for (int i = 0;; i++) {
2261 const char* file = distro_files[i];
2262 if (file == NULL) {
2263 break; // ran out of distro_files
2264 }
2265 if (file_exists(file)) {
2266 parse_os_info(buf, buflen, file);
2267 return;
2268 }
2269 }
2270 // special case for debian
2271 if (file_exists("/etc/debian_version")) {
2272 strncpy(buf, "Debian ", buflen);
2273 if (buflen > 7) {
2274 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2275 }
2276 } else {
2277 strncpy(buf, "Linux", buflen);
2278 }
2279 }
2280
2281 void os::Linux::print_libversion_info(outputStream* st) {
2282 // libc, pthread
2283 st->print("libc:");
2284 st->print("%s ", os::Linux::glibc_version());
2285 st->print("%s ", os::Linux::libpthread_version());
2286 st->cr();
2287 }
2288
2289 void os::Linux::print_proc_sys_info(outputStream* st) {
2290 st->cr();
2291 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2292 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2293 st->cr();
2294 st->cr();
2295
2296 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2297 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2298 st->cr();
2299 st->cr();
2300
2301 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2302 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2303 st->cr();
2304 st->cr();
2305 }
2306
2307 void os::Linux::print_full_memory_info(outputStream* st) {
2308 st->print("\n/proc/meminfo:\n");
2309 _print_ascii_file("/proc/meminfo", st);
2310 st->cr();
2311
2312 // some information regarding THPs; for details see
2313 // https://www.kernel.org/doc/Documentation/vm/transhuge.txt
2314 st->print_cr("/sys/kernel/mm/transparent_hugepage/enabled:");
2315 if (!_print_ascii_file("/sys/kernel/mm/transparent_hugepage/enabled", st)) {
2316 st->print_cr(" <Not Available>");
2317 }
2318 st->cr();
2319 st->print_cr("/sys/kernel/mm/transparent_hugepage/defrag (defrag/compaction efforts parameter):");
2320 if (!_print_ascii_file("/sys/kernel/mm/transparent_hugepage/defrag", st)) {
2321 st->print_cr(" <Not Available>");
2322 }
2323 st->cr();
2324 }
2325
2326 void os::Linux::print_ld_preload_file(outputStream* st) {
2327 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2328 st->cr();
2329 }
2330
2331 void os::Linux::print_uptime_info(outputStream* st) {
2332 struct sysinfo sinfo;
2333 int ret = sysinfo(&sinfo);
2334 if (ret == 0) {
2335 os::print_dhm(st, "OS uptime:", (long) sinfo.uptime);
2336 }
2337 }
2338
2339
2340 void os::Linux::print_container_info(outputStream* st) {
2341 if (!OSContainer::is_containerized()) {
2342 return;
2343 }
2344
2345 st->print("container (cgroup) information:\n");
2346
2347 const char *p_ct = OSContainer::container_type();
2348 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2349
2350 char *p = OSContainer::cpu_cpuset_cpus();
2351 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2352 free(p);
2353
2354 p = OSContainer::cpu_cpuset_memory_nodes();
2355 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2356 free(p);
2357
2358 int i = OSContainer::active_processor_count();
2359 st->print("active_processor_count: ");
2360 if (i > 0) {
2361 st->print("%d\n", i);
2362 } else {
2363 st->print("not supported\n");
2364 }
2365
2366 i = OSContainer::cpu_quota();
2367 st->print("cpu_quota: ");
2368 if (i > 0) {
2369 st->print("%d\n", i);
2370 } else {
2371 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2372 }
2373
2374 i = OSContainer::cpu_period();
2375 st->print("cpu_period: ");
2376 if (i > 0) {
2377 st->print("%d\n", i);
2378 } else {
2379 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2380 }
2381
2382 i = OSContainer::cpu_shares();
2383 st->print("cpu_shares: ");
2384 if (i > 0) {
2385 st->print("%d\n", i);
2386 } else {
2387 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2388 }
2389
2390 jlong j = OSContainer::memory_limit_in_bytes();
2391 st->print("memory_limit_in_bytes: ");
2392 if (j > 0) {
2393 st->print(JLONG_FORMAT "\n", j);
2394 } else {
2395 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2396 }
2397
2398 j = OSContainer::memory_and_swap_limit_in_bytes();
2399 st->print("memory_and_swap_limit_in_bytes: ");
2400 if (j > 0) {
2401 st->print(JLONG_FORMAT "\n", j);
2402 } else {
2403 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2404 }
2405
2406 j = OSContainer::memory_soft_limit_in_bytes();
2407 st->print("memory_soft_limit_in_bytes: ");
2408 if (j > 0) {
2409 st->print(JLONG_FORMAT "\n", j);
2410 } else {
2411 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2412 }
2413
2414 j = OSContainer::OSContainer::memory_usage_in_bytes();
2415 st->print("memory_usage_in_bytes: ");
2416 if (j > 0) {
2417 st->print(JLONG_FORMAT "\n", j);
2418 } else {
2419 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2420 }
2421
2422 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2423 st->print("memory_max_usage_in_bytes: ");
2424 if (j > 0) {
2425 st->print(JLONG_FORMAT "\n", j);
2426 } else {
2427 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2428 }
2429 st->cr();
2430 }
2431
2432 void os::Linux::print_steal_info(outputStream* st) {
2433 if (has_initial_tick_info) {
2434 CPUPerfTicks pticks;
2435 bool res = os::Linux::get_tick_information(&pticks, -1);
2436
2437 if (res && pticks.has_steal_ticks) {
2438 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2439 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2440 double steal_ticks_perc = 0.0;
2441 if (total_ticks_difference != 0) {
2442 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2443 }
2444 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2445 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2446 }
2447 }
2448 }
2449
2450 void os::print_memory_info(outputStream* st) {
2451
2452 st->print("Memory:");
2453 st->print(" %dk page", os::vm_page_size()>>10);
2454
2455 // values in struct sysinfo are "unsigned long"
2456 struct sysinfo si;
2457 sysinfo(&si);
2458
2459 st->print(", physical " UINT64_FORMAT "k",
2460 os::physical_memory() >> 10);
2461 st->print("(" UINT64_FORMAT "k free)",
2462 os::available_memory() >> 10);
2463 st->print(", swap " UINT64_FORMAT "k",
2464 ((jlong)si.totalswap * si.mem_unit) >> 10);
2465 st->print("(" UINT64_FORMAT "k free)",
2466 ((jlong)si.freeswap * si.mem_unit) >> 10);
2467 st->cr();
2468 }
2469
2470 // Print the first "model name" line and the first "flags" line
2471 // that we find and nothing more. We assume "model name" comes
2472 // before "flags" so if we find a second "model name", then the
2473 // "flags" field is considered missing.
2474 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2475 #if defined(IA32) || defined(AMD64)
2476 // Other platforms have less repetitive cpuinfo files
2477 FILE *fp = fopen("/proc/cpuinfo", "r");
2478 if (fp) {
2479 while (!feof(fp)) {
2480 if (fgets(buf, buflen, fp)) {
2481 // Assume model name comes before flags
2482 bool model_name_printed = false;
2483 if (strstr(buf, "model name") != NULL) {
2484 if (!model_name_printed) {
2485 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2486 st->print_raw(buf);
2487 model_name_printed = true;
2488 } else {
2489 // model name printed but not flags? Odd, just return
2490 fclose(fp);
2491 return true;
2492 }
2493 }
2494 // print the flags line too
2495 if (strstr(buf, "flags") != NULL) {
2496 st->print_raw(buf);
2497 fclose(fp);
2498 return true;
2499 }
2500 }
2501 }
2502 fclose(fp);
2503 }
2504 #endif // x86 platforms
2505 return false;
2506 }
2507
2508 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2509 // Only print the model name if the platform provides this as a summary
2510 if (!print_model_name_and_flags(st, buf, buflen)) {
2511 st->print("\n/proc/cpuinfo:\n");
2512 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2513 st->print_cr(" <Not Available>");
2514 }
2515 }
2516 }
2517
2518 #if defined(AMD64) || defined(IA32) || defined(X32)
2519 const char* search_string = "model name";
2520 #elif defined(M68K)
2521 const char* search_string = "CPU";
2522 #elif defined(PPC64)
2523 const char* search_string = "cpu";
2524 #elif defined(S390)
2525 const char* search_string = "machine =";
2526 #elif defined(SPARC)
2527 const char* search_string = "cpu";
2528 #else
2529 const char* search_string = "Processor";
2530 #endif
2531
2532 // Parses the cpuinfo file for string representing the model name.
2533 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2534 FILE* fp = fopen("/proc/cpuinfo", "r");
2535 if (fp != NULL) {
2536 while (!feof(fp)) {
2537 char buf[256];
2538 if (fgets(buf, sizeof(buf), fp)) {
2539 char* start = strstr(buf, search_string);
2540 if (start != NULL) {
2541 char *ptr = start + strlen(search_string);
2542 char *end = buf + strlen(buf);
2543 while (ptr != end) {
2544 // skip whitespace and colon for the rest of the name.
2545 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2546 break;
2547 }
2548 ptr++;
2549 }
2550 if (ptr != end) {
2551 // reasonable string, get rid of newline and keep the rest
2552 char* nl = strchr(buf, '\n');
2553 if (nl != NULL) *nl = '\0';
2554 strncpy(cpuinfo, ptr, length);
2555 fclose(fp);
2556 return;
2557 }
2558 }
2559 }
2560 }
2561 fclose(fp);
2562 }
2563 // cpuinfo not found or parsing failed, just print generic string. The entire
2564 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2565 #if defined(AARCH64)
2566 strncpy(cpuinfo, "AArch64", length);
2567 #elif defined(AMD64)
2568 strncpy(cpuinfo, "x86_64", length);
2569 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2570 strncpy(cpuinfo, "ARM", length);
2571 #elif defined(IA32)
2572 strncpy(cpuinfo, "x86_32", length);
2573 #elif defined(IA64)
2574 strncpy(cpuinfo, "IA64", length);
2575 #elif defined(PPC)
2576 strncpy(cpuinfo, "PPC64", length);
2577 #elif defined(S390)
2578 strncpy(cpuinfo, "S390", length);
2579 #elif defined(SPARC)
2580 strncpy(cpuinfo, "sparcv9", length);
2581 #elif defined(ZERO_LIBARCH)
2582 strncpy(cpuinfo, ZERO_LIBARCH, length);
2583 #else
2584 strncpy(cpuinfo, "unknown", length);
2585 #endif
2586 }
2587
2588 static void print_signal_handler(outputStream* st, int sig,
2589 char* buf, size_t buflen);
2590
2591 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2592 st->print_cr("Signal Handlers:");
2593 print_signal_handler(st, SIGSEGV, buf, buflen);
2594 print_signal_handler(st, SIGBUS , buf, buflen);
2595 print_signal_handler(st, SIGFPE , buf, buflen);
2596 print_signal_handler(st, SIGPIPE, buf, buflen);
2597 print_signal_handler(st, SIGXFSZ, buf, buflen);
2598 print_signal_handler(st, SIGILL , buf, buflen);
2599 print_signal_handler(st, SR_signum, buf, buflen);
2600 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2601 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2602 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2603 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2604 #if defined(PPC64)
2605 print_signal_handler(st, SIGTRAP, buf, buflen);
2606 #endif
2607 }
2608
2609 static char saved_jvm_path[MAXPATHLEN] = {0};
2610
2611 // Find the full path to the current module, libjvm.so
2612 void os::jvm_path(char *buf, jint buflen) {
2613 // Error checking.
2614 if (buflen < MAXPATHLEN) {
2615 assert(false, "must use a large-enough buffer");
2616 buf[0] = '\0';
2617 return;
2618 }
2619 // Lazy resolve the path to current module.
2620 if (saved_jvm_path[0] != 0) {
2621 strcpy(buf, saved_jvm_path);
2622 return;
2623 }
2624
2625 char dli_fname[MAXPATHLEN];
2626 bool ret = dll_address_to_library_name(
2627 CAST_FROM_FN_PTR(address, os::jvm_path),
2628 dli_fname, sizeof(dli_fname), NULL);
2629 assert(ret, "cannot locate libjvm");
2630 char *rp = NULL;
2631 if (ret && dli_fname[0] != '\0') {
2632 rp = os::Posix::realpath(dli_fname, buf, buflen);
2633 }
2634 if (rp == NULL) {
2635 return;
2636 }
2637
2638 if (Arguments::sun_java_launcher_is_altjvm()) {
2639 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2640 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2641 // If "/jre/lib/" appears at the right place in the string, then
2642 // assume we are installed in a JDK and we're done. Otherwise, check
2643 // for a JAVA_HOME environment variable and fix up the path so it
2644 // looks like libjvm.so is installed there (append a fake suffix
2645 // hotspot/libjvm.so).
2646 const char *p = buf + strlen(buf) - 1;
2647 for (int count = 0; p > buf && count < 5; ++count) {
2648 for (--p; p > buf && *p != '/'; --p)
2649 /* empty */ ;
2650 }
2651
2652 if (strncmp(p, "/jre/lib/", 9) != 0) {
2653 // Look for JAVA_HOME in the environment.
2654 char* java_home_var = ::getenv("JAVA_HOME");
2655 if (java_home_var != NULL && java_home_var[0] != 0) {
2656 char* jrelib_p;
2657 int len;
2658
2659 // Check the current module name "libjvm.so".
2660 p = strrchr(buf, '/');
2661 if (p == NULL) {
2662 return;
2663 }
2664 assert(strstr(p, "/libjvm") == p, "invalid library name");
2665
2666 rp = os::Posix::realpath(java_home_var, buf, buflen);
2667 if (rp == NULL) {
2668 return;
2669 }
2670
2671 // determine if this is a legacy image or modules image
2672 // modules image doesn't have "jre" subdirectory
2673 len = strlen(buf);
2674 assert(len < buflen, "Ran out of buffer room");
2675 jrelib_p = buf + len;
2676 snprintf(jrelib_p, buflen-len, "/jre/lib");
2677 if (0 != access(buf, F_OK)) {
2678 snprintf(jrelib_p, buflen-len, "/lib");
2679 }
2680
2681 if (0 == access(buf, F_OK)) {
2682 // Use current module name "libjvm.so"
2683 len = strlen(buf);
2684 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2685 } else {
2686 // Go back to path of .so
2687 rp = os::Posix::realpath(dli_fname, buf, buflen);
2688 if (rp == NULL) {
2689 return;
2690 }
2691 }
2692 }
2693 }
2694 }
2695
2696 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2697 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2698 }
2699
2700 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2701 // no prefix required, not even "_"
2702 }
2703
2704 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2705 // no suffix required
2706 }
2707
2708 ////////////////////////////////////////////////////////////////////////////////
2709 // sun.misc.Signal support
2710
2711 static void UserHandler(int sig, void *siginfo, void *context) {
2712 // Ctrl-C is pressed during error reporting, likely because the error
2713 // handler fails to abort. Let VM die immediately.
2714 if (sig == SIGINT && VMError::is_error_reported()) {
2715 os::die();
2716 }
2717
2718 os::signal_notify(sig);
2719 }
2720
2721 void* os::user_handler() {
2722 return CAST_FROM_FN_PTR(void*, UserHandler);
2723 }
2724
2725 extern "C" {
2726 typedef void (*sa_handler_t)(int);
2727 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2728 }
2729
2730 void* os::signal(int signal_number, void* handler) {
2731 struct sigaction sigAct, oldSigAct;
2732
2733 sigfillset(&(sigAct.sa_mask));
2734 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2735 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2736
2737 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2738 // -1 means registration failed
2739 return (void *)-1;
2740 }
2741
2742 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2743 }
2744
2745 void os::signal_raise(int signal_number) {
2746 ::raise(signal_number);
2747 }
2748
2749 // The following code is moved from os.cpp for making this
2750 // code platform specific, which it is by its very nature.
2751
2752 // Will be modified when max signal is changed to be dynamic
2753 int os::sigexitnum_pd() {
2754 return NSIG;
2755 }
2756
2757 // a counter for each possible signal value
2758 static volatile jint pending_signals[NSIG+1] = { 0 };
2759
2760 // Linux(POSIX) specific hand shaking semaphore.
2761 static Semaphore* sig_sem = NULL;
2762 static PosixSemaphore sr_semaphore;
2763
2764 static void jdk_misc_signal_init() {
2765 // Initialize signal structures
2766 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2767
2768 // Initialize signal semaphore
2769 sig_sem = new Semaphore();
2770 }
2771
2772 void os::signal_notify(int sig) {
2773 if (sig_sem != NULL) {
2774 Atomic::inc(&pending_signals[sig]);
2775 sig_sem->signal();
2776 } else {
2777 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2778 // initialization isn't called.
2779 assert(ReduceSignalUsage, "signal semaphore should be created");
2780 }
2781 }
2782
2783 static int check_pending_signals() {
2784 for (;;) {
2785 for (int i = 0; i < NSIG + 1; i++) {
2786 jint n = pending_signals[i];
2787 if (n > 0 && n == Atomic::cmpxchg(&pending_signals[i], n, n - 1)) {
2788 return i;
2789 }
2790 }
2791 JavaThread *thread = JavaThread::current();
2792 ThreadBlockInVM tbivm(thread);
2793
2794 bool threadIsSuspended;
2795 do {
2796 thread->set_suspend_equivalent();
2797 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2798 sig_sem->wait();
2799
2800 // were we externally suspended while we were waiting?
2801 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2802 if (threadIsSuspended) {
2803 // The semaphore has been incremented, but while we were waiting
2804 // another thread suspended us. We don't want to continue running
2805 // while suspended because that would surprise the thread that
2806 // suspended us.
2807 sig_sem->signal();
2808
2809 thread->java_suspend_self();
2810 }
2811 } while (threadIsSuspended);
2812 }
2813 }
2814
2815 int os::signal_wait() {
2816 return check_pending_signals();
2817 }
2818
2819 ////////////////////////////////////////////////////////////////////////////////
2820 // Virtual Memory
2821
2822 int os::vm_page_size() {
2823 // Seems redundant as all get out
2824 assert(os::Linux::page_size() != -1, "must call os::init");
2825 return os::Linux::page_size();
2826 }
2827
2828 // Solaris allocates memory by pages.
2829 int os::vm_allocation_granularity() {
2830 assert(os::Linux::page_size() != -1, "must call os::init");
2831 return os::Linux::page_size();
2832 }
2833
2834 // Rationale behind this function:
2835 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2836 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2837 // samples for JITted code. Here we create private executable mapping over the code cache
2838 // and then we can use standard (well, almost, as mapping can change) way to provide
2839 // info for the reporting script by storing timestamp and location of symbol
2840 void linux_wrap_code(char* base, size_t size) {
2841 static volatile jint cnt = 0;
2842
2843 if (!UseOprofile) {
2844 return;
2845 }
2846
2847 char buf[PATH_MAX+1];
2848 int num = Atomic::add(&cnt, 1);
2849
2850 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2851 os::get_temp_directory(), os::current_process_id(), num);
2852 unlink(buf);
2853
2854 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2855
2856 if (fd != -1) {
2857 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2858 if (rv != (off_t)-1) {
2859 if (::write(fd, "", 1) == 1) {
2860 mmap(base, size,
2861 PROT_READ|PROT_WRITE|PROT_EXEC,
2862 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2863 }
2864 }
2865 ::close(fd);
2866 unlink(buf);
2867 }
2868 }
2869
2870 static bool recoverable_mmap_error(int err) {
2871 // See if the error is one we can let the caller handle. This
2872 // list of errno values comes from JBS-6843484. I can't find a
2873 // Linux man page that documents this specific set of errno
2874 // values so while this list currently matches Solaris, it may
2875 // change as we gain experience with this failure mode.
2876 switch (err) {
2877 case EBADF:
2878 case EINVAL:
2879 case ENOTSUP:
2880 // let the caller deal with these errors
2881 return true;
2882
2883 default:
2884 // Any remaining errors on this OS can cause our reserved mapping
2885 // to be lost. That can cause confusion where different data
2886 // structures think they have the same memory mapped. The worst
2887 // scenario is if both the VM and a library think they have the
2888 // same memory mapped.
2889 return false;
2890 }
2891 }
2892
2893 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2894 int err) {
2895 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2896 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2897 os::strerror(err), err);
2898 }
2899
2900 static void warn_fail_commit_memory(char* addr, size_t size,
2901 size_t alignment_hint, bool exec,
2902 int err) {
2903 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2904 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2905 alignment_hint, exec, os::strerror(err), err);
2906 }
2907
2908 // NOTE: Linux kernel does not really reserve the pages for us.
2909 // All it does is to check if there are enough free pages
2910 // left at the time of mmap(). This could be a potential
2911 // problem.
2912 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2913 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2914 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2915 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2916 if (res != (uintptr_t) MAP_FAILED) {
2917 if (UseNUMAInterleaving) {
2918 numa_make_global(addr, size);
2919 }
2920 return 0;
2921 }
2922
2923 int err = errno; // save errno from mmap() call above
2924
2925 if (!recoverable_mmap_error(err)) {
2926 warn_fail_commit_memory(addr, size, exec, err);
2927 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2928 }
2929
2930 return err;
2931 }
2932
2933 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2934 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2935 }
2936
2937 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2938 const char* mesg) {
2939 assert(mesg != NULL, "mesg must be specified");
2940 int err = os::Linux::commit_memory_impl(addr, size, exec);
2941 if (err != 0) {
2942 // the caller wants all commit errors to exit with the specified mesg:
2943 warn_fail_commit_memory(addr, size, exec, err);
2944 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2945 }
2946 }
2947
2948 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2949 #ifndef MAP_HUGETLB
2950 #define MAP_HUGETLB 0x40000
2951 #endif
2952
2953 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2954 #ifndef MADV_HUGEPAGE
2955 #define MADV_HUGEPAGE 14
2956 #endif
2957
2958 int os::Linux::commit_memory_impl(char* addr, size_t size,
2959 size_t alignment_hint, bool exec) {
2960 int err = os::Linux::commit_memory_impl(addr, size, exec);
2961 if (err == 0) {
2962 realign_memory(addr, size, alignment_hint);
2963 }
2964 return err;
2965 }
2966
2967 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2968 bool exec) {
2969 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2970 }
2971
2972 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2973 size_t alignment_hint, bool exec,
2974 const char* mesg) {
2975 assert(mesg != NULL, "mesg must be specified");
2976 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2977 if (err != 0) {
2978 // the caller wants all commit errors to exit with the specified mesg:
2979 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2980 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2981 }
2982 }
2983
2984 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2985 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2986 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2987 // be supported or the memory may already be backed by huge pages.
2988 ::madvise(addr, bytes, MADV_HUGEPAGE);
2989 }
2990 }
2991
2992 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2993 // This method works by doing an mmap over an existing mmaping and effectively discarding
2994 // the existing pages. However it won't work for SHM-based large pages that cannot be
2995 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2996 // small pages on top of the SHM segment. This method always works for small pages, so we
2997 // allow that in any case.
2998 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2999 commit_memory(addr, bytes, alignment_hint, !ExecMem);
3000 }
3001 }
3002
3003 void os::numa_make_global(char *addr, size_t bytes) {
3004 Linux::numa_interleave_memory(addr, bytes);
3005 }
3006
3007 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
3008 // bind policy to MPOL_PREFERRED for the current thread.
3009 #define USE_MPOL_PREFERRED 0
3010
3011 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
3012 // To make NUMA and large pages more robust when both enabled, we need to ease
3013 // the requirements on where the memory should be allocated. MPOL_BIND is the
3014 // default policy and it will force memory to be allocated on the specified
3015 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
3016 // the specified node, but will not force it. Using this policy will prevent
3017 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
3018 // free large pages.
3019 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
3020 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
3021 }
3022
3023 bool os::numa_topology_changed() { return false; }
3024
3025 size_t os::numa_get_groups_num() {
3026 // Return just the number of nodes in which it's possible to allocate memory
3027 // (in numa terminology, configured nodes).
3028 return Linux::numa_num_configured_nodes();
3029 }
3030
3031 int os::numa_get_group_id() {
3032 int cpu_id = Linux::sched_getcpu();
3033 if (cpu_id != -1) {
3034 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3035 if (lgrp_id != -1) {
3036 return lgrp_id;
3037 }
3038 }
3039 return 0;
3040 }
3041
3042 int os::numa_get_group_id_for_address(const void* address) {
3043 void** pages = const_cast<void**>(&address);
3044 int id = -1;
3045
3046 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) {
3047 return -1;
3048 }
3049 if (id < 0) {
3050 return -1;
3051 }
3052 return id;
3053 }
3054
3055 int os::Linux::get_existing_num_nodes() {
3056 int node;
3057 int highest_node_number = Linux::numa_max_node();
3058 int num_nodes = 0;
3059
3060 // Get the total number of nodes in the system including nodes without memory.
3061 for (node = 0; node <= highest_node_number; node++) {
3062 if (is_node_in_existing_nodes(node)) {
3063 num_nodes++;
3064 }
3065 }
3066 return num_nodes;
3067 }
3068
3069 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3070 int highest_node_number = Linux::numa_max_node();
3071 size_t i = 0;
3072
3073 // Map all node ids in which it is possible to allocate memory. Also nodes are
3074 // not always consecutively available, i.e. available from 0 to the highest
3075 // node number. If the nodes have been bound explicitly using numactl membind,
3076 // then allocate memory from those nodes only.
3077 for (int node = 0; node <= highest_node_number; node++) {
3078 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3079 ids[i++] = node;
3080 }
3081 }
3082 return i;
3083 }
3084
3085 bool os::get_page_info(char *start, page_info* info) {
3086 return false;
3087 }
3088
3089 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3090 page_info* page_found) {
3091 return end;
3092 }
3093
3094
3095 int os::Linux::sched_getcpu_syscall(void) {
3096 unsigned int cpu = 0;
3097 int retval = -1;
3098
3099 #if defined(IA32)
3100 #ifndef SYS_getcpu
3101 #define SYS_getcpu 318
3102 #endif
3103 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3104 #elif defined(AMD64)
3105 // Unfortunately we have to bring all these macros here from vsyscall.h
3106 // to be able to compile on old linuxes.
3107 #define __NR_vgetcpu 2
3108 #define VSYSCALL_START (-10UL << 20)
3109 #define VSYSCALL_SIZE 1024
3110 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3111 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3112 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3113 retval = vgetcpu(&cpu, NULL, NULL);
3114 #endif
3115
3116 return (retval == -1) ? retval : cpu;
3117 }
3118
3119 void os::Linux::sched_getcpu_init() {
3120 // sched_getcpu() should be in libc.
3121 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3122 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3123
3124 // If it's not, try a direct syscall.
3125 if (sched_getcpu() == -1) {
3126 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3127 (void*)&sched_getcpu_syscall));
3128 }
3129
3130 if (sched_getcpu() == -1) {
3131 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3132 }
3133 }
3134
3135 // Something to do with the numa-aware allocator needs these symbols
3136 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3137 extern "C" JNIEXPORT void numa_error(char *where) { }
3138
3139 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3140 // load symbol from base version instead.
3141 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3142 void *f = dlvsym(handle, name, "libnuma_1.1");
3143 if (f == NULL) {
3144 f = dlsym(handle, name);
3145 }
3146 return f;
3147 }
3148
3149 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3150 // Return NULL if the symbol is not defined in this particular version.
3151 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3152 return dlvsym(handle, name, "libnuma_1.2");
3153 }
3154
3155 bool os::Linux::libnuma_init() {
3156 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3157 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3158 if (handle != NULL) {
3159 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3160 libnuma_dlsym(handle, "numa_node_to_cpus")));
3161 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3162 libnuma_dlsym(handle, "numa_max_node")));
3163 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3164 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3165 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3166 libnuma_dlsym(handle, "numa_available")));
3167 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3168 libnuma_dlsym(handle, "numa_tonode_memory")));
3169 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3170 libnuma_dlsym(handle, "numa_interleave_memory")));
3171 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3172 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3173 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3174 libnuma_dlsym(handle, "numa_set_bind_policy")));
3175 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3176 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3177 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3178 libnuma_dlsym(handle, "numa_distance")));
3179 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3180 libnuma_v2_dlsym(handle, "numa_get_membind")));
3181 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3182 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3183 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t,
3184 libnuma_dlsym(handle, "numa_move_pages")));
3185 set_numa_set_preferred(CAST_TO_FN_PTR(numa_set_preferred_func_t,
3186 libnuma_dlsym(handle, "numa_set_preferred")));
3187
3188 if (numa_available() != -1) {
3189 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3190 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3191 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3192 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3193 set_numa_membind_bitmask(_numa_get_membind());
3194 // Create an index -> node mapping, since nodes are not always consecutive
3195 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3196 rebuild_nindex_to_node_map();
3197 // Create a cpu -> node mapping
3198 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3199 rebuild_cpu_to_node_map();
3200 return true;
3201 }
3202 }
3203 }
3204 return false;
3205 }
3206
3207 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3208 // Creating guard page is very expensive. Java thread has HotSpot
3209 // guard pages, only enable glibc guard page for non-Java threads.
3210 // (Remember: compiler thread is a Java thread, too!)
3211 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3212 }
3213
3214 void os::Linux::rebuild_nindex_to_node_map() {
3215 int highest_node_number = Linux::numa_max_node();
3216
3217 nindex_to_node()->clear();
3218 for (int node = 0; node <= highest_node_number; node++) {
3219 if (Linux::is_node_in_existing_nodes(node)) {
3220 nindex_to_node()->append(node);
3221 }
3222 }
3223 }
3224
3225 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3226 // The table is later used in get_node_by_cpu().
3227 void os::Linux::rebuild_cpu_to_node_map() {
3228 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3229 // in libnuma (possible values are starting from 16,
3230 // and continuing up with every other power of 2, but less
3231 // than the maximum number of CPUs supported by kernel), and
3232 // is a subject to change (in libnuma version 2 the requirements
3233 // are more reasonable) we'll just hardcode the number they use
3234 // in the library.
3235 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3236
3237 size_t cpu_num = processor_count();
3238 size_t cpu_map_size = NCPUS / BitsPerCLong;
3239 size_t cpu_map_valid_size =
3240 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3241
3242 cpu_to_node()->clear();
3243 cpu_to_node()->at_grow(cpu_num - 1);
3244
3245 size_t node_num = get_existing_num_nodes();
3246
3247 int distance = 0;
3248 int closest_distance = INT_MAX;
3249 int closest_node = 0;
3250 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3251 for (size_t i = 0; i < node_num; i++) {
3252 // Check if node is configured (not a memory-less node). If it is not, find
3253 // the closest configured node. Check also if node is bound, i.e. it's allowed
3254 // to allocate memory from the node. If it's not allowed, map cpus in that node
3255 // to the closest node from which memory allocation is allowed.
3256 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3257 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3258 closest_distance = INT_MAX;
3259 // Check distance from all remaining nodes in the system. Ignore distance
3260 // from itself, from another non-configured node, and from another non-bound
3261 // node.
3262 for (size_t m = 0; m < node_num; m++) {
3263 if (m != i &&
3264 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3265 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3266 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3267 // If a closest node is found, update. There is always at least one
3268 // configured and bound node in the system so there is always at least
3269 // one node close.
3270 if (distance != 0 && distance < closest_distance) {
3271 closest_distance = distance;
3272 closest_node = nindex_to_node()->at(m);
3273 }
3274 }
3275 }
3276 } else {
3277 // Current node is already a configured node.
3278 closest_node = nindex_to_node()->at(i);
3279 }
3280
3281 // Get cpus from the original node and map them to the closest node. If node
3282 // is a configured node (not a memory-less node), then original node and
3283 // closest node are the same.
3284 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3285 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3286 if (cpu_map[j] != 0) {
3287 for (size_t k = 0; k < BitsPerCLong; k++) {
3288 if (cpu_map[j] & (1UL << k)) {
3289 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3290 }
3291 }
3292 }
3293 }
3294 }
3295 }
3296 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3297 }
3298
3299 int os::Linux::get_node_by_cpu(int cpu_id) {
3300 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3301 return cpu_to_node()->at(cpu_id);
3302 }
3303 return -1;
3304 }
3305
3306 GrowableArray<int>* os::Linux::_cpu_to_node;
3307 GrowableArray<int>* os::Linux::_nindex_to_node;
3308 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3309 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3310 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3311 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3312 os::Linux::numa_available_func_t os::Linux::_numa_available;
3313 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3314 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3315 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3316 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3317 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3318 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3319 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3320 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3321 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages;
3322 os::Linux::numa_set_preferred_func_t os::Linux::_numa_set_preferred;
3323 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3324 unsigned long* os::Linux::_numa_all_nodes;
3325 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3326 struct bitmask* os::Linux::_numa_nodes_ptr;
3327 struct bitmask* os::Linux::_numa_interleave_bitmask;
3328 struct bitmask* os::Linux::_numa_membind_bitmask;
3329
3330 bool os::pd_uncommit_memory(char* addr, size_t size) {
3331 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3332 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3333 return res != (uintptr_t) MAP_FAILED;
3334 }
3335
3336 static address get_stack_commited_bottom(address bottom, size_t size) {
3337 address nbot = bottom;
3338 address ntop = bottom + size;
3339
3340 size_t page_sz = os::vm_page_size();
3341 unsigned pages = size / page_sz;
3342
3343 unsigned char vec[1];
3344 unsigned imin = 1, imax = pages + 1, imid;
3345 int mincore_return_value = 0;
3346
3347 assert(imin <= imax, "Unexpected page size");
3348
3349 while (imin < imax) {
3350 imid = (imax + imin) / 2;
3351 nbot = ntop - (imid * page_sz);
3352
3353 // Use a trick with mincore to check whether the page is mapped or not.
3354 // mincore sets vec to 1 if page resides in memory and to 0 if page
3355 // is swapped output but if page we are asking for is unmapped
3356 // it returns -1,ENOMEM
3357 mincore_return_value = mincore(nbot, page_sz, vec);
3358
3359 if (mincore_return_value == -1) {
3360 // Page is not mapped go up
3361 // to find first mapped page
3362 if (errno != EAGAIN) {
3363 assert(errno == ENOMEM, "Unexpected mincore errno");
3364 imax = imid;
3365 }
3366 } else {
3367 // Page is mapped go down
3368 // to find first not mapped page
3369 imin = imid + 1;
3370 }
3371 }
3372
3373 nbot = nbot + page_sz;
3374
3375 // Adjust stack bottom one page up if last checked page is not mapped
3376 if (mincore_return_value == -1) {
3377 nbot = nbot + page_sz;
3378 }
3379
3380 return nbot;
3381 }
3382
3383 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3384 int mincore_return_value;
3385 const size_t stripe = 1024; // query this many pages each time
3386 unsigned char vec[stripe + 1];
3387 // set a guard
3388 vec[stripe] = 'X';
3389
3390 const size_t page_sz = os::vm_page_size();
3391 size_t pages = size / page_sz;
3392
3393 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3394 assert(is_aligned(size, page_sz), "Size must be page aligned");
3395
3396 committed_start = NULL;
3397
3398 int loops = (pages + stripe - 1) / stripe;
3399 int committed_pages = 0;
3400 address loop_base = start;
3401 bool found_range = false;
3402
3403 for (int index = 0; index < loops && !found_range; index ++) {
3404 assert(pages > 0, "Nothing to do");
3405 int pages_to_query = (pages >= stripe) ? stripe : pages;
3406 pages -= pages_to_query;
3407
3408 // Get stable read
3409 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3410
3411 // During shutdown, some memory goes away without properly notifying NMT,
3412 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3413 // Bailout and return as not committed for now.
3414 if (mincore_return_value == -1 && errno == ENOMEM) {
3415 return false;
3416 }
3417
3418 assert(vec[stripe] == 'X', "overflow guard");
3419 assert(mincore_return_value == 0, "Range must be valid");
3420 // Process this stripe
3421 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3422 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3423 // End of current contiguous region
3424 if (committed_start != NULL) {
3425 found_range = true;
3426 break;
3427 }
3428 } else { // committed
3429 // Start of region
3430 if (committed_start == NULL) {
3431 committed_start = loop_base + page_sz * vecIdx;
3432 }
3433 committed_pages ++;
3434 }
3435 }
3436
3437 loop_base += pages_to_query * page_sz;
3438 }
3439
3440 if (committed_start != NULL) {
3441 assert(committed_pages > 0, "Must have committed region");
3442 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3443 assert(committed_start >= start && committed_start < start + size, "Out of range");
3444 committed_size = page_sz * committed_pages;
3445 return true;
3446 } else {
3447 assert(committed_pages == 0, "Should not have committed region");
3448 return false;
3449 }
3450 }
3451
3452
3453 // Linux uses a growable mapping for the stack, and if the mapping for
3454 // the stack guard pages is not removed when we detach a thread the
3455 // stack cannot grow beyond the pages where the stack guard was
3456 // mapped. If at some point later in the process the stack expands to
3457 // that point, the Linux kernel cannot expand the stack any further
3458 // because the guard pages are in the way, and a segfault occurs.
3459 //
3460 // However, it's essential not to split the stack region by unmapping
3461 // a region (leaving a hole) that's already part of the stack mapping,
3462 // so if the stack mapping has already grown beyond the guard pages at
3463 // the time we create them, we have to truncate the stack mapping.
3464 // So, we need to know the extent of the stack mapping when
3465 // create_stack_guard_pages() is called.
3466
3467 // We only need this for stacks that are growable: at the time of
3468 // writing thread stacks don't use growable mappings (i.e. those
3469 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3470 // only applies to the main thread.
3471
3472 // If the (growable) stack mapping already extends beyond the point
3473 // where we're going to put our guard pages, truncate the mapping at
3474 // that point by munmap()ping it. This ensures that when we later
3475 // munmap() the guard pages we don't leave a hole in the stack
3476 // mapping. This only affects the main/primordial thread
3477
3478 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3479 if (os::is_primordial_thread()) {
3480 // As we manually grow stack up to bottom inside create_attached_thread(),
3481 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3482 // we don't need to do anything special.
3483 // Check it first, before calling heavy function.
3484 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3485 unsigned char vec[1];
3486
3487 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3488 // Fallback to slow path on all errors, including EAGAIN
3489 stack_extent = (uintptr_t) get_stack_commited_bottom(
3490 os::Linux::initial_thread_stack_bottom(),
3491 (size_t)addr - stack_extent);
3492 }
3493
3494 if (stack_extent < (uintptr_t)addr) {
3495 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3496 }
3497 }
3498
3499 return os::commit_memory(addr, size, !ExecMem);
3500 }
3501
3502 // If this is a growable mapping, remove the guard pages entirely by
3503 // munmap()ping them. If not, just call uncommit_memory(). This only
3504 // affects the main/primordial thread, but guard against future OS changes.
3505 // It's safe to always unmap guard pages for primordial thread because we
3506 // always place it right after end of the mapped region.
3507
3508 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3509 uintptr_t stack_extent, stack_base;
3510
3511 if (os::is_primordial_thread()) {
3512 return ::munmap(addr, size) == 0;
3513 }
3514
3515 return os::uncommit_memory(addr, size);
3516 }
3517
3518 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3519 // at 'requested_addr'. If there are existing memory mappings at the same
3520 // location, however, they will be overwritten. If 'fixed' is false,
3521 // 'requested_addr' is only treated as a hint, the return value may or
3522 // may not start from the requested address. Unlike Linux mmap(), this
3523 // function returns NULL to indicate failure.
3524 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3525 char * addr;
3526 int flags;
3527
3528 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3529 if (fixed) {
3530 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3531 flags |= MAP_FIXED;
3532 }
3533
3534 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3535 // touch an uncommitted page. Otherwise, the read/write might
3536 // succeed if we have enough swap space to back the physical page.
3537 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3538 flags, -1, 0);
3539
3540 return addr == MAP_FAILED ? NULL : addr;
3541 }
3542
3543 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3544 // (req_addr != NULL) or with a given alignment.
3545 // - bytes shall be a multiple of alignment.
3546 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3547 // - alignment sets the alignment at which memory shall be allocated.
3548 // It must be a multiple of allocation granularity.
3549 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3550 // req_addr or NULL.
3551 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3552
3553 size_t extra_size = bytes;
3554 if (req_addr == NULL && alignment > 0) {
3555 extra_size += alignment;
3556 }
3557
3558 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3559 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3560 -1, 0);
3561 if (start == MAP_FAILED) {
3562 start = NULL;
3563 } else {
3564 if (req_addr != NULL) {
3565 if (start != req_addr) {
3566 ::munmap(start, extra_size);
3567 start = NULL;
3568 }
3569 } else {
3570 char* const start_aligned = align_up(start, alignment);
3571 char* const end_aligned = start_aligned + bytes;
3572 char* const end = start + extra_size;
3573 if (start_aligned > start) {
3574 ::munmap(start, start_aligned - start);
3575 }
3576 if (end_aligned < end) {
3577 ::munmap(end_aligned, end - end_aligned);
3578 }
3579 start = start_aligned;
3580 }
3581 }
3582 return start;
3583 }
3584
3585 static int anon_munmap(char * addr, size_t size) {
3586 return ::munmap(addr, size) == 0;
3587 }
3588
3589 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3590 size_t alignment_hint) {
3591 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3592 }
3593
3594 bool os::pd_release_memory(char* addr, size_t size) {
3595 return anon_munmap(addr, size);
3596 }
3597
3598 static bool linux_mprotect(char* addr, size_t size, int prot) {
3599 // Linux wants the mprotect address argument to be page aligned.
3600 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3601
3602 // According to SUSv3, mprotect() should only be used with mappings
3603 // established by mmap(), and mmap() always maps whole pages. Unaligned
3604 // 'addr' likely indicates problem in the VM (e.g. trying to change
3605 // protection of malloc'ed or statically allocated memory). Check the
3606 // caller if you hit this assert.
3607 assert(addr == bottom, "sanity check");
3608
3609 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3610 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3611 return ::mprotect(bottom, size, prot) == 0;
3612 }
3613
3614 // Set protections specified
3615 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3616 bool is_committed) {
3617 unsigned int p = 0;
3618 switch (prot) {
3619 case MEM_PROT_NONE: p = PROT_NONE; break;
3620 case MEM_PROT_READ: p = PROT_READ; break;
3621 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3622 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3623 default:
3624 ShouldNotReachHere();
3625 }
3626 // is_committed is unused.
3627 return linux_mprotect(addr, bytes, p);
3628 }
3629
3630 bool os::guard_memory(char* addr, size_t size) {
3631 return linux_mprotect(addr, size, PROT_NONE);
3632 }
3633
3634 bool os::unguard_memory(char* addr, size_t size) {
3635 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3636 }
3637
3638 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3639 size_t page_size) {
3640 bool result = false;
3641 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3642 MAP_ANONYMOUS|MAP_PRIVATE,
3643 -1, 0);
3644 if (p != MAP_FAILED) {
3645 void *aligned_p = align_up(p, page_size);
3646
3647 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3648
3649 munmap(p, page_size * 2);
3650 }
3651
3652 if (warn && !result) {
3653 warning("TransparentHugePages is not supported by the operating system.");
3654 }
3655
3656 return result;
3657 }
3658
3659 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3660 bool result = false;
3661 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3662 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3663 -1, 0);
3664
3665 if (p != MAP_FAILED) {
3666 // We don't know if this really is a huge page or not.
3667 FILE *fp = fopen("/proc/self/maps", "r");
3668 if (fp) {
3669 while (!feof(fp)) {
3670 char chars[257];
3671 long x = 0;
3672 if (fgets(chars, sizeof(chars), fp)) {
3673 if (sscanf(chars, "%lx-%*x", &x) == 1
3674 && x == (long)p) {
3675 if (strstr (chars, "hugepage")) {
3676 result = true;
3677 break;
3678 }
3679 }
3680 }
3681 }
3682 fclose(fp);
3683 }
3684 munmap(p, page_size);
3685 }
3686
3687 if (warn && !result) {
3688 warning("HugeTLBFS is not supported by the operating system.");
3689 }
3690
3691 return result;
3692 }
3693
3694 // From the coredump_filter documentation:
3695 //
3696 // - (bit 0) anonymous private memory
3697 // - (bit 1) anonymous shared memory
3698 // - (bit 2) file-backed private memory
3699 // - (bit 3) file-backed shared memory
3700 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3701 // effective only if the bit 2 is cleared)
3702 // - (bit 5) hugetlb private memory
3703 // - (bit 6) hugetlb shared memory
3704 // - (bit 7) dax private memory
3705 // - (bit 8) dax shared memory
3706 //
3707 static void set_coredump_filter(CoredumpFilterBit bit) {
3708 FILE *f;
3709 long cdm;
3710
3711 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3712 return;
3713 }
3714
3715 if (fscanf(f, "%lx", &cdm) != 1) {
3716 fclose(f);
3717 return;
3718 }
3719
3720 long saved_cdm = cdm;
3721 rewind(f);
3722 cdm |= bit;
3723
3724 if (cdm != saved_cdm) {
3725 fprintf(f, "%#lx", cdm);
3726 }
3727
3728 fclose(f);
3729 }
3730
3731 // Large page support
3732
3733 static size_t _large_page_size = 0;
3734
3735 size_t os::Linux::find_large_page_size() {
3736 size_t large_page_size = 0;
3737
3738 // large_page_size on Linux is used to round up heap size. x86 uses either
3739 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3740 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3741 // page as large as 256M.
3742 //
3743 // Here we try to figure out page size by parsing /proc/meminfo and looking
3744 // for a line with the following format:
3745 // Hugepagesize: 2048 kB
3746 //
3747 // If we can't determine the value (e.g. /proc is not mounted, or the text
3748 // format has been changed), we'll use the largest page size supported by
3749 // the processor.
3750
3751 #ifndef ZERO
3752 large_page_size =
3753 AARCH64_ONLY(2 * M)
3754 AMD64_ONLY(2 * M)
3755 ARM32_ONLY(2 * M)
3756 IA32_ONLY(4 * M)
3757 IA64_ONLY(256 * M)
3758 PPC_ONLY(4 * M)
3759 S390_ONLY(1 * M)
3760 SPARC_ONLY(4 * M);
3761 #endif // ZERO
3762
3763 FILE *fp = fopen("/proc/meminfo", "r");
3764 if (fp) {
3765 while (!feof(fp)) {
3766 int x = 0;
3767 char buf[16];
3768 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3769 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3770 large_page_size = x * K;
3771 break;
3772 }
3773 } else {
3774 // skip to next line
3775 for (;;) {
3776 int ch = fgetc(fp);
3777 if (ch == EOF || ch == (int)'\n') break;
3778 }
3779 }
3780 }
3781 fclose(fp);
3782 }
3783
3784 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3785 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3786 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3787 proper_unit_for_byte_size(large_page_size));
3788 }
3789
3790 return large_page_size;
3791 }
3792
3793 size_t os::Linux::setup_large_page_size() {
3794 _large_page_size = Linux::find_large_page_size();
3795 const size_t default_page_size = (size_t)Linux::page_size();
3796 if (_large_page_size > default_page_size) {
3797 _page_sizes[0] = _large_page_size;
3798 _page_sizes[1] = default_page_size;
3799 _page_sizes[2] = 0;
3800 }
3801
3802 return _large_page_size;
3803 }
3804
3805 bool os::Linux::setup_large_page_type(size_t page_size) {
3806 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3807 FLAG_IS_DEFAULT(UseSHM) &&
3808 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3809
3810 // The type of large pages has not been specified by the user.
3811
3812 // Try UseHugeTLBFS and then UseSHM.
3813 UseHugeTLBFS = UseSHM = true;
3814
3815 // Don't try UseTransparentHugePages since there are known
3816 // performance issues with it turned on. This might change in the future.
3817 UseTransparentHugePages = false;
3818 }
3819
3820 if (UseTransparentHugePages) {
3821 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3822 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3823 UseHugeTLBFS = false;
3824 UseSHM = false;
3825 return true;
3826 }
3827 UseTransparentHugePages = false;
3828 }
3829
3830 if (UseHugeTLBFS) {
3831 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3832 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3833 UseSHM = false;
3834 return true;
3835 }
3836 UseHugeTLBFS = false;
3837 }
3838
3839 return UseSHM;
3840 }
3841
3842 void os::large_page_init() {
3843 if (!UseLargePages &&
3844 !UseTransparentHugePages &&
3845 !UseHugeTLBFS &&
3846 !UseSHM) {
3847 // Not using large pages.
3848 return;
3849 }
3850
3851 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3852 // The user explicitly turned off large pages.
3853 // Ignore the rest of the large pages flags.
3854 UseTransparentHugePages = false;
3855 UseHugeTLBFS = false;
3856 UseSHM = false;
3857 return;
3858 }
3859
3860 size_t large_page_size = Linux::setup_large_page_size();
3861 UseLargePages = Linux::setup_large_page_type(large_page_size);
3862
3863 set_coredump_filter(LARGEPAGES_BIT);
3864 }
3865
3866 #ifndef SHM_HUGETLB
3867 #define SHM_HUGETLB 04000
3868 #endif
3869
3870 #define shm_warning_format(format, ...) \
3871 do { \
3872 if (UseLargePages && \
3873 (!FLAG_IS_DEFAULT(UseLargePages) || \
3874 !FLAG_IS_DEFAULT(UseSHM) || \
3875 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3876 warning(format, __VA_ARGS__); \
3877 } \
3878 } while (0)
3879
3880 #define shm_warning(str) shm_warning_format("%s", str)
3881
3882 #define shm_warning_with_errno(str) \
3883 do { \
3884 int err = errno; \
3885 shm_warning_format(str " (error = %d)", err); \
3886 } while (0)
3887
3888 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3889 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3890
3891 if (!is_aligned(alignment, SHMLBA)) {
3892 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3893 return NULL;
3894 }
3895
3896 // To ensure that we get 'alignment' aligned memory from shmat,
3897 // we pre-reserve aligned virtual memory and then attach to that.
3898
3899 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3900 if (pre_reserved_addr == NULL) {
3901 // Couldn't pre-reserve aligned memory.
3902 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3903 return NULL;
3904 }
3905
3906 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3907 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3908
3909 if ((intptr_t)addr == -1) {
3910 int err = errno;
3911 shm_warning_with_errno("Failed to attach shared memory.");
3912
3913 assert(err != EACCES, "Unexpected error");
3914 assert(err != EIDRM, "Unexpected error");
3915 assert(err != EINVAL, "Unexpected error");
3916
3917 // Since we don't know if the kernel unmapped the pre-reserved memory area
3918 // we can't unmap it, since that would potentially unmap memory that was
3919 // mapped from other threads.
3920 return NULL;
3921 }
3922
3923 return addr;
3924 }
3925
3926 static char* shmat_at_address(int shmid, char* req_addr) {
3927 if (!is_aligned(req_addr, SHMLBA)) {
3928 assert(false, "Requested address needs to be SHMLBA aligned");
3929 return NULL;
3930 }
3931
3932 char* addr = (char*)shmat(shmid, req_addr, 0);
3933
3934 if ((intptr_t)addr == -1) {
3935 shm_warning_with_errno("Failed to attach shared memory.");
3936 return NULL;
3937 }
3938
3939 return addr;
3940 }
3941
3942 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3943 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3944 if (req_addr != NULL) {
3945 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3946 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3947 return shmat_at_address(shmid, req_addr);
3948 }
3949
3950 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3951 // return large page size aligned memory addresses when req_addr == NULL.
3952 // However, if the alignment is larger than the large page size, we have
3953 // to manually ensure that the memory returned is 'alignment' aligned.
3954 if (alignment > os::large_page_size()) {
3955 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3956 return shmat_with_alignment(shmid, bytes, alignment);
3957 } else {
3958 return shmat_at_address(shmid, NULL);
3959 }
3960 }
3961
3962 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3963 char* req_addr, bool exec) {
3964 // "exec" is passed in but not used. Creating the shared image for
3965 // the code cache doesn't have an SHM_X executable permission to check.
3966 assert(UseLargePages && UseSHM, "only for SHM large pages");
3967 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3968 assert(is_aligned(req_addr, alignment), "Unaligned address");
3969
3970 if (!is_aligned(bytes, os::large_page_size())) {
3971 return NULL; // Fallback to small pages.
3972 }
3973
3974 // Create a large shared memory region to attach to based on size.
3975 // Currently, size is the total size of the heap.
3976 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3977 if (shmid == -1) {
3978 // Possible reasons for shmget failure:
3979 // 1. shmmax is too small for Java heap.
3980 // > check shmmax value: cat /proc/sys/kernel/shmmax
3981 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3982 // 2. not enough large page memory.
3983 // > check available large pages: cat /proc/meminfo
3984 // > increase amount of large pages:
3985 // echo new_value > /proc/sys/vm/nr_hugepages
3986 // Note 1: different Linux may use different name for this property,
3987 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3988 // Note 2: it's possible there's enough physical memory available but
3989 // they are so fragmented after a long run that they can't
3990 // coalesce into large pages. Try to reserve large pages when
3991 // the system is still "fresh".
3992 shm_warning_with_errno("Failed to reserve shared memory.");
3993 return NULL;
3994 }
3995
3996 // Attach to the region.
3997 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3998
3999 // Remove shmid. If shmat() is successful, the actual shared memory segment
4000 // will be deleted when it's detached by shmdt() or when the process
4001 // terminates. If shmat() is not successful this will remove the shared
4002 // segment immediately.
4003 shmctl(shmid, IPC_RMID, NULL);
4004
4005 return addr;
4006 }
4007
4008 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
4009 int error) {
4010 assert(error == ENOMEM, "Only expect to fail if no memory is available");
4011
4012 bool warn_on_failure = UseLargePages &&
4013 (!FLAG_IS_DEFAULT(UseLargePages) ||
4014 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
4015 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
4016
4017 if (warn_on_failure) {
4018 char msg[128];
4019 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
4020 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
4021 warning("%s", msg);
4022 }
4023 }
4024
4025 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
4026 char* req_addr,
4027 bool exec) {
4028 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4029 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
4030 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4031
4032 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4033 char* addr = (char*)::mmap(req_addr, bytes, prot,
4034 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
4035 -1, 0);
4036
4037 if (addr == MAP_FAILED) {
4038 warn_on_large_pages_failure(req_addr, bytes, errno);
4039 return NULL;
4040 }
4041
4042 assert(is_aligned(addr, os::large_page_size()), "Must be");
4043
4044 return addr;
4045 }
4046
4047 // Reserve memory using mmap(MAP_HUGETLB).
4048 // - bytes shall be a multiple of alignment.
4049 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4050 // - alignment sets the alignment at which memory shall be allocated.
4051 // It must be a multiple of allocation granularity.
4052 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4053 // req_addr or NULL.
4054 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4055 size_t alignment,
4056 char* req_addr,
4057 bool exec) {
4058 size_t large_page_size = os::large_page_size();
4059 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4060
4061 assert(is_aligned(req_addr, alignment), "Must be");
4062 assert(is_aligned(bytes, alignment), "Must be");
4063
4064 // First reserve - but not commit - the address range in small pages.
4065 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4066
4067 if (start == NULL) {
4068 return NULL;
4069 }
4070
4071 assert(is_aligned(start, alignment), "Must be");
4072
4073 char* end = start + bytes;
4074
4075 // Find the regions of the allocated chunk that can be promoted to large pages.
4076 char* lp_start = align_up(start, large_page_size);
4077 char* lp_end = align_down(end, large_page_size);
4078
4079 size_t lp_bytes = lp_end - lp_start;
4080
4081 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4082
4083 if (lp_bytes == 0) {
4084 // The mapped region doesn't even span the start and the end of a large page.
4085 // Fall back to allocate a non-special area.
4086 ::munmap(start, end - start);
4087 return NULL;
4088 }
4089
4090 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4091
4092 void* result;
4093
4094 // Commit small-paged leading area.
4095 if (start != lp_start) {
4096 result = ::mmap(start, lp_start - start, prot,
4097 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4098 -1, 0);
4099 if (result == MAP_FAILED) {
4100 ::munmap(lp_start, end - lp_start);
4101 return NULL;
4102 }
4103 }
4104
4105 // Commit large-paged area.
4106 result = ::mmap(lp_start, lp_bytes, prot,
4107 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4108 -1, 0);
4109 if (result == MAP_FAILED) {
4110 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4111 // If the mmap above fails, the large pages region will be unmapped and we
4112 // have regions before and after with small pages. Release these regions.
4113 //
4114 // | mapped | unmapped | mapped |
4115 // ^ ^ ^ ^
4116 // start lp_start lp_end end
4117 //
4118 ::munmap(start, lp_start - start);
4119 ::munmap(lp_end, end - lp_end);
4120 return NULL;
4121 }
4122
4123 // Commit small-paged trailing area.
4124 if (lp_end != end) {
4125 result = ::mmap(lp_end, end - lp_end, prot,
4126 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4127 -1, 0);
4128 if (result == MAP_FAILED) {
4129 ::munmap(start, lp_end - start);
4130 return NULL;
4131 }
4132 }
4133
4134 return start;
4135 }
4136
4137 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4138 size_t alignment,
4139 char* req_addr,
4140 bool exec) {
4141 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4142 assert(is_aligned(req_addr, alignment), "Must be");
4143 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4144 assert(is_power_of_2(os::large_page_size()), "Must be");
4145 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4146
4147 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4148 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4149 } else {
4150 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4151 }
4152 }
4153
4154 char* os::pd_reserve_memory_special(size_t bytes, size_t alignment,
4155 char* req_addr, bool exec) {
4156 assert(UseLargePages, "only for large pages");
4157
4158 char* addr;
4159 if (UseSHM) {
4160 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4161 } else {
4162 assert(UseHugeTLBFS, "must be");
4163 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4164 }
4165
4166 if (addr != NULL) {
4167 if (UseNUMAInterleaving) {
4168 numa_make_global(addr, bytes);
4169 }
4170 }
4171
4172 return addr;
4173 }
4174
4175 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4176 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4177 return shmdt(base) == 0;
4178 }
4179
4180 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4181 return pd_release_memory(base, bytes);
4182 }
4183
4184 bool os::pd_release_memory_special(char* base, size_t bytes) {
4185 assert(UseLargePages, "only for large pages");
4186 bool res;
4187
4188 if (UseSHM) {
4189 res = os::Linux::release_memory_special_shm(base, bytes);
4190 } else {
4191 assert(UseHugeTLBFS, "must be");
4192 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4193 }
4194 return res;
4195 }
4196
4197 size_t os::large_page_size() {
4198 return _large_page_size;
4199 }
4200
4201 // With SysV SHM the entire memory region must be allocated as shared
4202 // memory.
4203 // HugeTLBFS allows application to commit large page memory on demand.
4204 // However, when committing memory with HugeTLBFS fails, the region
4205 // that was supposed to be committed will lose the old reservation
4206 // and allow other threads to steal that memory region. Because of this
4207 // behavior we can't commit HugeTLBFS memory.
4208 bool os::can_commit_large_page_memory() {
4209 return UseTransparentHugePages;
4210 }
4211
4212 bool os::can_execute_large_page_memory() {
4213 return UseTransparentHugePages || UseHugeTLBFS;
4214 }
4215
4216 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4217 assert(file_desc >= 0, "file_desc is not valid");
4218 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4219 if (result != NULL) {
4220 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4221 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4222 }
4223 }
4224 return result;
4225 }
4226
4227 // Reserve memory at an arbitrary address, only if that area is
4228 // available (and not reserved for something else).
4229
4230 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4231 // Assert only that the size is a multiple of the page size, since
4232 // that's all that mmap requires, and since that's all we really know
4233 // about at this low abstraction level. If we need higher alignment,
4234 // we can either pass an alignment to this method or verify alignment
4235 // in one of the methods further up the call chain. See bug 5044738.
4236 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4237
4238 // Repeatedly allocate blocks until the block is allocated at the
4239 // right spot.
4240
4241 // Linux mmap allows caller to pass an address as hint; give it a try first,
4242 // if kernel honors the hint then we can return immediately.
4243 char * addr = anon_mmap(requested_addr, bytes, false);
4244 if (addr == requested_addr) {
4245 return requested_addr;
4246 }
4247
4248 if (addr != NULL) {
4249 // mmap() is successful but it fails to reserve at the requested address
4250 anon_munmap(addr, bytes);
4251 }
4252
4253 return NULL;
4254 }
4255
4256 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4257 void os::infinite_sleep() {
4258 while (true) { // sleep forever ...
4259 ::sleep(100); // ... 100 seconds at a time
4260 }
4261 }
4262
4263 // Used to convert frequent JVM_Yield() to nops
4264 bool os::dont_yield() {
4265 return DontYieldALot;
4266 }
4267
4268 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4269 // actually give up the CPU. Since skip buddy (v2.6.28):
4270 //
4271 // * Sets the yielding task as skip buddy for current CPU's run queue.
4272 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4273 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4274 //
4275 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4276 // getting re-scheduled immediately.
4277 //
4278 void os::naked_yield() {
4279 sched_yield();
4280 }
4281
4282 ////////////////////////////////////////////////////////////////////////////////
4283 // thread priority support
4284
4285 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4286 // only supports dynamic priority, static priority must be zero. For real-time
4287 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4288 // However, for large multi-threaded applications, SCHED_RR is not only slower
4289 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4290 // of 5 runs - Sep 2005).
4291 //
4292 // The following code actually changes the niceness of kernel-thread/LWP. It
4293 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4294 // not the entire user process, and user level threads are 1:1 mapped to kernel
4295 // threads. It has always been the case, but could change in the future. For
4296 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4297 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4298 // (e.g., root privilege or CAP_SYS_NICE capability).
4299
4300 int os::java_to_os_priority[CriticalPriority + 1] = {
4301 19, // 0 Entry should never be used
4302
4303 4, // 1 MinPriority
4304 3, // 2
4305 2, // 3
4306
4307 1, // 4
4308 0, // 5 NormPriority
4309 -1, // 6
4310
4311 -2, // 7
4312 -3, // 8
4313 -4, // 9 NearMaxPriority
4314
4315 -5, // 10 MaxPriority
4316
4317 -5 // 11 CriticalPriority
4318 };
4319
4320 static int prio_init() {
4321 if (ThreadPriorityPolicy == 1) {
4322 if (geteuid() != 0) {
4323 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy) && !FLAG_IS_JIMAGE_RESOURCE(ThreadPriorityPolicy)) {
4324 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4325 "e.g., being the root user. If the necessary permission is not " \
4326 "possessed, changes to priority will be silently ignored.");
4327 }
4328 }
4329 }
4330 if (UseCriticalJavaThreadPriority) {
4331 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4332 }
4333 return 0;
4334 }
4335
4336 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4337 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4338
4339 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4340 return (ret == 0) ? OS_OK : OS_ERR;
4341 }
4342
4343 OSReturn os::get_native_priority(const Thread* const thread,
4344 int *priority_ptr) {
4345 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4346 *priority_ptr = java_to_os_priority[NormPriority];
4347 return OS_OK;
4348 }
4349
4350 errno = 0;
4351 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4352 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4353 }
4354
4355 ////////////////////////////////////////////////////////////////////////////////
4356 // suspend/resume support
4357
4358 // The low-level signal-based suspend/resume support is a remnant from the
4359 // old VM-suspension that used to be for java-suspension, safepoints etc,
4360 // within hotspot. Currently used by JFR's OSThreadSampler
4361 //
4362 // The remaining code is greatly simplified from the more general suspension
4363 // code that used to be used.
4364 //
4365 // The protocol is quite simple:
4366 // - suspend:
4367 // - sends a signal to the target thread
4368 // - polls the suspend state of the osthread using a yield loop
4369 // - target thread signal handler (SR_handler) sets suspend state
4370 // and blocks in sigsuspend until continued
4371 // - resume:
4372 // - sets target osthread state to continue
4373 // - sends signal to end the sigsuspend loop in the SR_handler
4374 //
4375 // Note that the SR_lock plays no role in this suspend/resume protocol,
4376 // but is checked for NULL in SR_handler as a thread termination indicator.
4377 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4378 //
4379 // Note that resume_clear_context() and suspend_save_context() are needed
4380 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4381 // which in part is used by:
4382 // - Forte Analyzer: AsyncGetCallTrace()
4383 // - StackBanging: get_frame_at_stack_banging_point()
4384
4385 static void resume_clear_context(OSThread *osthread) {
4386 osthread->set_ucontext(NULL);
4387 osthread->set_siginfo(NULL);
4388 }
4389
4390 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4391 ucontext_t* context) {
4392 osthread->set_ucontext(context);
4393 osthread->set_siginfo(siginfo);
4394 }
4395
4396 // Handler function invoked when a thread's execution is suspended or
4397 // resumed. We have to be careful that only async-safe functions are
4398 // called here (Note: most pthread functions are not async safe and
4399 // should be avoided.)
4400 //
4401 // Note: sigwait() is a more natural fit than sigsuspend() from an
4402 // interface point of view, but sigwait() prevents the signal hander
4403 // from being run. libpthread would get very confused by not having
4404 // its signal handlers run and prevents sigwait()'s use with the
4405 // mutex granting granting signal.
4406 //
4407 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4408 //
4409 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4410 // Save and restore errno to avoid confusing native code with EINTR
4411 // after sigsuspend.
4412 int old_errno = errno;
4413
4414 Thread* thread = Thread::current_or_null_safe();
4415 assert(thread != NULL, "Missing current thread in SR_handler");
4416
4417 // On some systems we have seen signal delivery get "stuck" until the signal
4418 // mask is changed as part of thread termination. Check that the current thread
4419 // has not already terminated (via SR_lock()) - else the following assertion
4420 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4421 // destructor has completed.
4422
4423 if (thread->SR_lock() == NULL) {
4424 return;
4425 }
4426
4427 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4428
4429 OSThread* osthread = thread->osthread();
4430
4431 os::SuspendResume::State current = osthread->sr.state();
4432 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4433 suspend_save_context(osthread, siginfo, context);
4434
4435 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4436 os::SuspendResume::State state = osthread->sr.suspended();
4437 if (state == os::SuspendResume::SR_SUSPENDED) {
4438 sigset_t suspend_set; // signals for sigsuspend()
4439 sigemptyset(&suspend_set);
4440 // get current set of blocked signals and unblock resume signal
4441 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4442 sigdelset(&suspend_set, SR_signum);
4443
4444 sr_semaphore.signal();
4445 // wait here until we are resumed
4446 while (1) {
4447 sigsuspend(&suspend_set);
4448
4449 os::SuspendResume::State result = osthread->sr.running();
4450 if (result == os::SuspendResume::SR_RUNNING) {
4451 sr_semaphore.signal();
4452 break;
4453 }
4454 }
4455
4456 } else if (state == os::SuspendResume::SR_RUNNING) {
4457 // request was cancelled, continue
4458 } else {
4459 ShouldNotReachHere();
4460 }
4461
4462 resume_clear_context(osthread);
4463 } else if (current == os::SuspendResume::SR_RUNNING) {
4464 // request was cancelled, continue
4465 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4466 // ignore
4467 } else {
4468 // ignore
4469 }
4470
4471 errno = old_errno;
4472 }
4473
4474 static int SR_initialize() {
4475 struct sigaction act;
4476 char *s;
4477
4478 // Get signal number to use for suspend/resume
4479 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4480 int sig = ::strtol(s, 0, 10);
4481 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4482 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4483 SR_signum = sig;
4484 } else {
4485 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4486 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4487 }
4488 }
4489
4490 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4491 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4492
4493 sigemptyset(&SR_sigset);
4494 sigaddset(&SR_sigset, SR_signum);
4495
4496 // Set up signal handler for suspend/resume
4497 act.sa_flags = SA_RESTART|SA_SIGINFO;
4498 act.sa_handler = (void (*)(int)) SR_handler;
4499
4500 // SR_signum is blocked by default.
4501 // 4528190 - We also need to block pthread restart signal (32 on all
4502 // supported Linux platforms). Note that LinuxThreads need to block
4503 // this signal for all threads to work properly. So we don't have
4504 // to use hard-coded signal number when setting up the mask.
4505 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4506
4507 if (sigaction(SR_signum, &act, 0) == -1) {
4508 return -1;
4509 }
4510
4511 // Save signal flag
4512 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4513 return 0;
4514 }
4515
4516 static int sr_notify(OSThread* osthread) {
4517 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4518 assert_status(status == 0, status, "pthread_kill");
4519 return status;
4520 }
4521
4522 // "Randomly" selected value for how long we want to spin
4523 // before bailing out on suspending a thread, also how often
4524 // we send a signal to a thread we want to resume
4525 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4526 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4527
4528 // returns true on success and false on error - really an error is fatal
4529 // but this seems the normal response to library errors
4530 static bool do_suspend(OSThread* osthread) {
4531 assert(osthread->sr.is_running(), "thread should be running");
4532 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4533
4534 // mark as suspended and send signal
4535 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4536 // failed to switch, state wasn't running?
4537 ShouldNotReachHere();
4538 return false;
4539 }
4540
4541 if (sr_notify(osthread) != 0) {
4542 ShouldNotReachHere();
4543 }
4544
4545 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4546 while (true) {
4547 if (sr_semaphore.timedwait(2)) {
4548 break;
4549 } else {
4550 // timeout
4551 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4552 if (cancelled == os::SuspendResume::SR_RUNNING) {
4553 return false;
4554 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4555 // make sure that we consume the signal on the semaphore as well
4556 sr_semaphore.wait();
4557 break;
4558 } else {
4559 ShouldNotReachHere();
4560 return false;
4561 }
4562 }
4563 }
4564
4565 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4566 return true;
4567 }
4568
4569 static void do_resume(OSThread* osthread) {
4570 assert(osthread->sr.is_suspended(), "thread should be suspended");
4571 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4572
4573 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4574 // failed to switch to WAKEUP_REQUEST
4575 ShouldNotReachHere();
4576 return;
4577 }
4578
4579 while (true) {
4580 if (sr_notify(osthread) == 0) {
4581 if (sr_semaphore.timedwait(2)) {
4582 if (osthread->sr.is_running()) {
4583 return;
4584 }
4585 }
4586 } else {
4587 ShouldNotReachHere();
4588 }
4589 }
4590
4591 guarantee(osthread->sr.is_running(), "Must be running!");
4592 }
4593
4594 ///////////////////////////////////////////////////////////////////////////////////
4595 // signal handling (except suspend/resume)
4596
4597 // This routine may be used by user applications as a "hook" to catch signals.
4598 // The user-defined signal handler must pass unrecognized signals to this
4599 // routine, and if it returns true (non-zero), then the signal handler must
4600 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4601 // routine will never retun false (zero), but instead will execute a VM panic
4602 // routine kill the process.
4603 //
4604 // If this routine returns false, it is OK to call it again. This allows
4605 // the user-defined signal handler to perform checks either before or after
4606 // the VM performs its own checks. Naturally, the user code would be making
4607 // a serious error if it tried to handle an exception (such as a null check
4608 // or breakpoint) that the VM was generating for its own correct operation.
4609 //
4610 // This routine may recognize any of the following kinds of signals:
4611 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4612 // It should be consulted by handlers for any of those signals.
4613 //
4614 // The caller of this routine must pass in the three arguments supplied
4615 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4616 // field of the structure passed to sigaction(). This routine assumes that
4617 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4618 //
4619 // Note that the VM will print warnings if it detects conflicting signal
4620 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4621 //
4622 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4623 siginfo_t* siginfo,
4624 void* ucontext,
4625 int abort_if_unrecognized);
4626
4627 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4628 assert(info != NULL && uc != NULL, "it must be old kernel");
4629 int orig_errno = errno; // Preserve errno value over signal handler.
4630 JVM_handle_linux_signal(sig, info, uc, true);
4631 errno = orig_errno;
4632 }
4633
4634
4635 // This boolean allows users to forward their own non-matching signals
4636 // to JVM_handle_linux_signal, harmlessly.
4637 bool os::Linux::signal_handlers_are_installed = false;
4638
4639 // For signal-chaining
4640 bool os::Linux::libjsig_is_loaded = false;
4641 typedef struct sigaction *(*get_signal_t)(int);
4642 get_signal_t os::Linux::get_signal_action = NULL;
4643
4644 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4645 struct sigaction *actp = NULL;
4646
4647 if (libjsig_is_loaded) {
4648 // Retrieve the old signal handler from libjsig
4649 actp = (*get_signal_action)(sig);
4650 }
4651 if (actp == NULL) {
4652 // Retrieve the preinstalled signal handler from jvm
4653 actp = os::Posix::get_preinstalled_handler(sig);
4654 }
4655
4656 return actp;
4657 }
4658
4659 static bool call_chained_handler(struct sigaction *actp, int sig,
4660 siginfo_t *siginfo, void *context) {
4661 // Call the old signal handler
4662 if (actp->sa_handler == SIG_DFL) {
4663 // It's more reasonable to let jvm treat it as an unexpected exception
4664 // instead of taking the default action.
4665 return false;
4666 } else if (actp->sa_handler != SIG_IGN) {
4667 if ((actp->sa_flags & SA_NODEFER) == 0) {
4668 // automaticlly block the signal
4669 sigaddset(&(actp->sa_mask), sig);
4670 }
4671
4672 sa_handler_t hand = NULL;
4673 sa_sigaction_t sa = NULL;
4674 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4675 // retrieve the chained handler
4676 if (siginfo_flag_set) {
4677 sa = actp->sa_sigaction;
4678 } else {
4679 hand = actp->sa_handler;
4680 }
4681
4682 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4683 actp->sa_handler = SIG_DFL;
4684 }
4685
4686 // try to honor the signal mask
4687 sigset_t oset;
4688 sigemptyset(&oset);
4689 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4690
4691 // call into the chained handler
4692 if (siginfo_flag_set) {
4693 (*sa)(sig, siginfo, context);
4694 } else {
4695 (*hand)(sig);
4696 }
4697
4698 // restore the signal mask
4699 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4700 }
4701 // Tell jvm's signal handler the signal is taken care of.
4702 return true;
4703 }
4704
4705 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4706 bool chained = false;
4707 // signal-chaining
4708 if (UseSignalChaining) {
4709 struct sigaction *actp = get_chained_signal_action(sig);
4710 if (actp != NULL) {
4711 chained = call_chained_handler(actp, sig, siginfo, context);
4712 }
4713 }
4714 return chained;
4715 }
4716
4717 // for diagnostic
4718 int sigflags[NSIG];
4719
4720 int os::Linux::get_our_sigflags(int sig) {
4721 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4722 return sigflags[sig];
4723 }
4724
4725 void os::Linux::set_our_sigflags(int sig, int flags) {
4726 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4727 if (sig > 0 && sig < NSIG) {
4728 sigflags[sig] = flags;
4729 }
4730 }
4731
4732 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4733 // Check for overwrite.
4734 struct sigaction oldAct;
4735 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4736
4737 void* oldhand = oldAct.sa_sigaction
4738 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4739 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4740 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4741 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4742 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4743 if (AllowUserSignalHandlers || !set_installed) {
4744 // Do not overwrite; user takes responsibility to forward to us.
4745 return;
4746 } else if (UseSignalChaining) {
4747 // save the old handler in jvm
4748 os::Posix::save_preinstalled_handler(sig, oldAct);
4749 // libjsig also interposes the sigaction() call below and saves the
4750 // old sigaction on it own.
4751 } else {
4752 fatal("Encountered unexpected pre-existing sigaction handler "
4753 "%#lx for signal %d.", (long)oldhand, sig);
4754 }
4755 }
4756
4757 struct sigaction sigAct;
4758 sigfillset(&(sigAct.sa_mask));
4759 sigAct.sa_handler = SIG_DFL;
4760 if (!set_installed) {
4761 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4762 } else {
4763 sigAct.sa_sigaction = signalHandler;
4764 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4765 }
4766 // Save flags, which are set by ours
4767 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4768 sigflags[sig] = sigAct.sa_flags;
4769
4770 int ret = sigaction(sig, &sigAct, &oldAct);
4771 assert(ret == 0, "check");
4772
4773 void* oldhand2 = oldAct.sa_sigaction
4774 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4775 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4776 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4777 }
4778
4779 // install signal handlers for signals that HotSpot needs to
4780 // handle in order to support Java-level exception handling.
4781
4782 void os::Linux::install_signal_handlers() {
4783 if (!signal_handlers_are_installed) {
4784 signal_handlers_are_installed = true;
4785
4786 // signal-chaining
4787 typedef void (*signal_setting_t)();
4788 signal_setting_t begin_signal_setting = NULL;
4789 signal_setting_t end_signal_setting = NULL;
4790 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4791 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4792 if (begin_signal_setting != NULL) {
4793 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4794 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4795 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4796 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4797 libjsig_is_loaded = true;
4798 assert(UseSignalChaining, "should enable signal-chaining");
4799 }
4800 if (libjsig_is_loaded) {
4801 // Tell libjsig jvm is setting signal handlers
4802 (*begin_signal_setting)();
4803 }
4804
4805 set_signal_handler(SIGSEGV, true);
4806 set_signal_handler(SIGPIPE, true);
4807 set_signal_handler(SIGBUS, true);
4808 set_signal_handler(SIGILL, true);
4809 set_signal_handler(SIGFPE, true);
4810 #if defined(PPC64)
4811 set_signal_handler(SIGTRAP, true);
4812 #endif
4813 set_signal_handler(SIGXFSZ, true);
4814
4815 if (libjsig_is_loaded) {
4816 // Tell libjsig jvm finishes setting signal handlers
4817 (*end_signal_setting)();
4818 }
4819
4820 // We don't activate signal checker if libjsig is in place, we trust ourselves
4821 // and if UserSignalHandler is installed all bets are off.
4822 // Log that signal checking is off only if -verbose:jni is specified.
4823 if (CheckJNICalls) {
4824 if (libjsig_is_loaded) {
4825 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled");
4826 check_signals = false;
4827 }
4828 if (AllowUserSignalHandlers) {
4829 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4830 check_signals = false;
4831 }
4832 }
4833 }
4834 }
4835
4836 // This is the fastest way to get thread cpu time on Linux.
4837 // Returns cpu time (user+sys) for any thread, not only for current.
4838 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4839 // It might work on 2.6.10+ with a special kernel/glibc patch.
4840 // For reference, please, see IEEE Std 1003.1-2004:
4841 // http://www.unix.org/single_unix_specification
4842
4843 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4844 struct timespec tp;
4845 int rc = os::Posix::clock_gettime(clockid, &tp);
4846 assert(rc == 0, "clock_gettime is expected to return 0 code");
4847
4848 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4849 }
4850
4851 /////
4852 // glibc on Linux platform uses non-documented flag
4853 // to indicate, that some special sort of signal
4854 // trampoline is used.
4855 // We will never set this flag, and we should
4856 // ignore this flag in our diagnostic
4857 #ifdef SIGNIFICANT_SIGNAL_MASK
4858 #undef SIGNIFICANT_SIGNAL_MASK
4859 #endif
4860 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4861
4862 static const char* get_signal_handler_name(address handler,
4863 char* buf, int buflen) {
4864 int offset = 0;
4865 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4866 if (found) {
4867 // skip directory names
4868 const char *p1, *p2;
4869 p1 = buf;
4870 size_t len = strlen(os::file_separator());
4871 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4872 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4873 } else {
4874 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4875 }
4876 return buf;
4877 }
4878
4879 static void print_signal_handler(outputStream* st, int sig,
4880 char* buf, size_t buflen) {
4881 struct sigaction sa;
4882
4883 sigaction(sig, NULL, &sa);
4884
4885 // See comment for SIGNIFICANT_SIGNAL_MASK define
4886 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4887
4888 st->print("%s: ", os::exception_name(sig, buf, buflen));
4889
4890 address handler = (sa.sa_flags & SA_SIGINFO)
4891 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4892 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4893
4894 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4895 st->print("SIG_DFL");
4896 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4897 st->print("SIG_IGN");
4898 } else {
4899 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4900 }
4901
4902 st->print(", sa_mask[0]=");
4903 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4904
4905 address rh = VMError::get_resetted_sighandler(sig);
4906 // May be, handler was resetted by VMError?
4907 if (rh != NULL) {
4908 handler = rh;
4909 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4910 }
4911
4912 st->print(", sa_flags=");
4913 os::Posix::print_sa_flags(st, sa.sa_flags);
4914
4915 // Check: is it our handler?
4916 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4917 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4918 // It is our signal handler
4919 // check for flags, reset system-used one!
4920 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4921 st->print(
4922 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4923 os::Linux::get_our_sigflags(sig));
4924 }
4925 }
4926 st->cr();
4927 }
4928
4929
4930 #define DO_SIGNAL_CHECK(sig) \
4931 do { \
4932 if (!sigismember(&check_signal_done, sig)) { \
4933 os::Linux::check_signal_handler(sig); \
4934 } \
4935 } while (0)
4936
4937 // This method is a periodic task to check for misbehaving JNI applications
4938 // under CheckJNI, we can add any periodic checks here
4939
4940 void os::run_periodic_checks() {
4941 if (check_signals == false) return;
4942
4943 // SEGV and BUS if overridden could potentially prevent
4944 // generation of hs*.log in the event of a crash, debugging
4945 // such a case can be very challenging, so we absolutely
4946 // check the following for a good measure:
4947 DO_SIGNAL_CHECK(SIGSEGV);
4948 DO_SIGNAL_CHECK(SIGILL);
4949 DO_SIGNAL_CHECK(SIGFPE);
4950 DO_SIGNAL_CHECK(SIGBUS);
4951 DO_SIGNAL_CHECK(SIGPIPE);
4952 DO_SIGNAL_CHECK(SIGXFSZ);
4953 #if defined(PPC64)
4954 DO_SIGNAL_CHECK(SIGTRAP);
4955 #endif
4956
4957 // ReduceSignalUsage allows the user to override these handlers
4958 // see comments at the very top and jvm_md.h
4959 if (!ReduceSignalUsage) {
4960 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4961 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4962 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4963 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4964 }
4965
4966 DO_SIGNAL_CHECK(SR_signum);
4967 }
4968
4969 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4970
4971 static os_sigaction_t os_sigaction = NULL;
4972
4973 void os::Linux::check_signal_handler(int sig) {
4974 char buf[O_BUFLEN];
4975 address jvmHandler = NULL;
4976
4977
4978 struct sigaction act;
4979 if (os_sigaction == NULL) {
4980 // only trust the default sigaction, in case it has been interposed
4981 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4982 if (os_sigaction == NULL) return;
4983 }
4984
4985 os_sigaction(sig, (struct sigaction*)NULL, &act);
4986
4987
4988 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4989
4990 address thisHandler = (act.sa_flags & SA_SIGINFO)
4991 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4992 : CAST_FROM_FN_PTR(address, act.sa_handler);
4993
4994
4995 switch (sig) {
4996 case SIGSEGV:
4997 case SIGBUS:
4998 case SIGFPE:
4999 case SIGPIPE:
5000 case SIGILL:
5001 case SIGXFSZ:
5002 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5003 break;
5004
5005 case SHUTDOWN1_SIGNAL:
5006 case SHUTDOWN2_SIGNAL:
5007 case SHUTDOWN3_SIGNAL:
5008 case BREAK_SIGNAL:
5009 jvmHandler = (address)user_handler();
5010 break;
5011
5012 default:
5013 if (sig == SR_signum) {
5014 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5015 } else {
5016 return;
5017 }
5018 break;
5019 }
5020
5021 if (thisHandler != jvmHandler) {
5022 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5023 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5024 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5025 // No need to check this sig any longer
5026 sigaddset(&check_signal_done, sig);
5027 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5028 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5029 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5030 exception_name(sig, buf, O_BUFLEN));
5031 }
5032 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5033 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5034 tty->print("expected:");
5035 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5036 tty->cr();
5037 tty->print(" found:");
5038 os::Posix::print_sa_flags(tty, act.sa_flags);
5039 tty->cr();
5040 // No need to check this sig any longer
5041 sigaddset(&check_signal_done, sig);
5042 }
5043
5044 // Dump all the signal
5045 if (sigismember(&check_signal_done, sig)) {
5046 print_signal_handlers(tty, buf, O_BUFLEN);
5047 }
5048 }
5049
5050 extern void report_error(char* file_name, int line_no, char* title,
5051 char* format, ...);
5052
5053 // this is called _before_ most of the global arguments have been parsed
5054 void os::init(void) {
5055 char dummy; // used to get a guess on initial stack address
5056
5057 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5058
5059 init_random(1234567);
5060
5061 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5062 if (Linux::page_size() == -1) {
5063 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5064 os::strerror(errno));
5065 }
5066 init_page_sizes((size_t) Linux::page_size());
5067
5068 Linux::initialize_system_info();
5069
5070 os::Linux::CPUPerfTicks pticks;
5071 bool res = os::Linux::get_tick_information(&pticks, -1);
5072
5073 if (res && pticks.has_steal_ticks) {
5074 has_initial_tick_info = true;
5075 initial_total_ticks = pticks.total;
5076 initial_steal_ticks = pticks.steal;
5077 }
5078
5079 // _main_thread points to the thread that created/loaded the JVM.
5080 Linux::_main_thread = pthread_self();
5081
5082 // retrieve entry point for pthread_setname_np
5083 Linux::_pthread_setname_np =
5084 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5085
5086 os::Posix::init();
5087
5088 initial_time_count = javaTimeNanos();
5089
5090 // Always warn if no monotonic clock available
5091 if (!os::Posix::supports_monotonic_clock()) {
5092 warning("No monotonic clock was available - timed services may " \
5093 "be adversely affected if the time-of-day clock changes");
5094 }
5095 }
5096
5097 // To install functions for atexit system call
5098 extern "C" {
5099 static void perfMemory_exit_helper() {
5100 perfMemory_exit();
5101 }
5102 }
5103
5104 void os::pd_init_container_support() {
5105 OSContainer::init();
5106 }
5107
5108 void os::Linux::numa_init() {
5109
5110 // Java can be invoked as
5111 // 1. Without numactl and heap will be allocated/configured on all nodes as
5112 // per the system policy.
5113 // 2. With numactl --interleave:
5114 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5115 // API for membind case bitmask is reset.
5116 // Interleave is only hint and Kernel can fallback to other nodes if
5117 // no memory is available on the target nodes.
5118 // 3. With numactl --membind:
5119 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5120 // interleave case returns bitmask of all nodes.
5121 // numa_all_nodes_ptr holds bitmask of all nodes.
5122 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5123 // bitmask when externally configured to run on all or fewer nodes.
5124
5125 if (!Linux::libnuma_init()) {
5126 UseNUMA = false;
5127 } else {
5128 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5129 // If there's only one node (they start from 0) or if the process
5130 // is bound explicitly to a single node using membind, disable NUMA unless
5131 // user explicilty forces NUMA optimizations on single-node/UMA systems
5132 UseNUMA = ForceNUMA;
5133 } else {
5134
5135 LogTarget(Info,os) log;
5136 LogStream ls(log);
5137
5138 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5139
5140 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5141 const char* numa_mode = "membind";
5142
5143 if (Linux::is_running_in_interleave_mode()) {
5144 bmp = Linux::_numa_interleave_bitmask;
5145 numa_mode = "interleave";
5146 }
5147
5148 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5149 " Heap will be configured using NUMA memory nodes:", numa_mode);
5150
5151 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5152 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5153 ls.print(" %d", node);
5154 }
5155 }
5156 }
5157 }
5158
5159 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5160 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5161 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5162 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5163 // and disable adaptive resizing.
5164 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5165 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5166 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5167 UseAdaptiveSizePolicy = false;
5168 UseAdaptiveNUMAChunkSizing = false;
5169 }
5170 }
5171 }
5172
5173 // this is called _after_ the global arguments have been parsed
5174 jint os::init_2(void) {
5175
5176 // This could be set after os::Posix::init() but all platforms
5177 // have to set it the same so we have to mirror Solaris.
5178 DEBUG_ONLY(os::set_mutex_init_done();)
5179
5180 os::Posix::init_2();
5181
5182 Linux::fast_thread_clock_init();
5183
5184 // initialize suspend/resume support - must do this before signal_sets_init()
5185 if (SR_initialize() != 0) {
5186 perror("SR_initialize failed");
5187 return JNI_ERR;
5188 }
5189
5190 Linux::signal_sets_init();
5191 Linux::install_signal_handlers();
5192 // Initialize data for jdk.internal.misc.Signal
5193 if (!ReduceSignalUsage) {
5194 jdk_misc_signal_init();
5195 }
5196
5197 if (AdjustStackSizeForTLS) {
5198 get_minstack_init();
5199 }
5200
5201 // Check and sets minimum stack sizes against command line options
5202 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5203 return JNI_ERR;
5204 }
5205
5206 #if defined(IA32)
5207 // Need to ensure we've determined the process's initial stack to
5208 // perform the workaround
5209 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5210 workaround_expand_exec_shield_cs_limit();
5211 #else
5212 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5213 if (!suppress_primordial_thread_resolution) {
5214 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5215 }
5216 #endif
5217
5218 Linux::libpthread_init();
5219 Linux::sched_getcpu_init();
5220 log_info(os)("HotSpot is running with %s, %s",
5221 Linux::glibc_version(), Linux::libpthread_version());
5222
5223 if (UseNUMA) {
5224 Linux::numa_init();
5225 }
5226
5227 if (MaxFDLimit) {
5228 // set the number of file descriptors to max. print out error
5229 // if getrlimit/setrlimit fails but continue regardless.
5230 struct rlimit nbr_files;
5231 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5232 if (status != 0) {
5233 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5234 } else {
5235 nbr_files.rlim_cur = nbr_files.rlim_max;
5236 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5237 if (status != 0) {
5238 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5239 }
5240 }
5241 }
5242
5243 // at-exit methods are called in the reverse order of their registration.
5244 // atexit functions are called on return from main or as a result of a
5245 // call to exit(3C). There can be only 32 of these functions registered
5246 // and atexit() does not set errno.
5247
5248 if (PerfAllowAtExitRegistration) {
5249 // only register atexit functions if PerfAllowAtExitRegistration is set.
5250 // atexit functions can be delayed until process exit time, which
5251 // can be problematic for embedded VM situations. Embedded VMs should
5252 // call DestroyJavaVM() to assure that VM resources are released.
5253
5254 // note: perfMemory_exit_helper atexit function may be removed in
5255 // the future if the appropriate cleanup code can be added to the
5256 // VM_Exit VMOperation's doit method.
5257 if (atexit(perfMemory_exit_helper) != 0) {
5258 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5259 }
5260 }
5261
5262 // initialize thread priority policy
5263 prio_init();
5264
5265 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5266 set_coredump_filter(DAX_SHARED_BIT);
5267 }
5268
5269 if (DumpPrivateMappingsInCore) {
5270 set_coredump_filter(FILE_BACKED_PVT_BIT);
5271 }
5272
5273 if (DumpSharedMappingsInCore) {
5274 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5275 }
5276
5277 return JNI_OK;
5278 }
5279
5280 // older glibc versions don't have this macro (which expands to
5281 // an optimized bit-counting function) so we have to roll our own
5282 #ifndef CPU_COUNT
5283
5284 static int _cpu_count(const cpu_set_t* cpus) {
5285 int count = 0;
5286 // only look up to the number of configured processors
5287 for (int i = 0; i < os::processor_count(); i++) {
5288 if (CPU_ISSET(i, cpus)) {
5289 count++;
5290 }
5291 }
5292 return count;
5293 }
5294
5295 #define CPU_COUNT(cpus) _cpu_count(cpus)
5296
5297 #endif // CPU_COUNT
5298
5299 // Get the current number of available processors for this process.
5300 // This value can change at any time during a process's lifetime.
5301 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5302 // If it appears there may be more than 1024 processors then we do a
5303 // dynamic check - see 6515172 for details.
5304 // If anything goes wrong we fallback to returning the number of online
5305 // processors - which can be greater than the number available to the process.
5306 int os::Linux::active_processor_count() {
5307 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5308 cpu_set_t* cpus_p = &cpus;
5309 int cpus_size = sizeof(cpu_set_t);
5310
5311 int configured_cpus = os::processor_count(); // upper bound on available cpus
5312 int cpu_count = 0;
5313
5314 // old build platforms may not support dynamic cpu sets
5315 #ifdef CPU_ALLOC
5316
5317 // To enable easy testing of the dynamic path on different platforms we
5318 // introduce a diagnostic flag: UseCpuAllocPath
5319 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5320 // kernel may use a mask bigger than cpu_set_t
5321 log_trace(os)("active_processor_count: using dynamic path %s"
5322 "- configured processors: %d",
5323 UseCpuAllocPath ? "(forced) " : "",
5324 configured_cpus);
5325 cpus_p = CPU_ALLOC(configured_cpus);
5326 if (cpus_p != NULL) {
5327 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5328 // zero it just to be safe
5329 CPU_ZERO_S(cpus_size, cpus_p);
5330 }
5331 else {
5332 // failed to allocate so fallback to online cpus
5333 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5334 log_trace(os)("active_processor_count: "
5335 "CPU_ALLOC failed (%s) - using "
5336 "online processor count: %d",
5337 os::strerror(errno), online_cpus);
5338 return online_cpus;
5339 }
5340 }
5341 else {
5342 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5343 configured_cpus);
5344 }
5345 #else // CPU_ALLOC
5346 // these stubs won't be executed
5347 #define CPU_COUNT_S(size, cpus) -1
5348 #define CPU_FREE(cpus)
5349
5350 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5351 configured_cpus);
5352 #endif // CPU_ALLOC
5353
5354 // pid 0 means the current thread - which we have to assume represents the process
5355 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5356 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5357 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5358 }
5359 else {
5360 cpu_count = CPU_COUNT(cpus_p);
5361 }
5362 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5363 }
5364 else {
5365 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5366 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5367 "which may exceed available processors", os::strerror(errno), cpu_count);
5368 }
5369
5370 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5371 CPU_FREE(cpus_p);
5372 }
5373
5374 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5375 return cpu_count;
5376 }
5377
5378 // Determine the active processor count from one of
5379 // three different sources:
5380 //
5381 // 1. User option -XX:ActiveProcessorCount
5382 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5383 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5384 //
5385 // Option 1, if specified, will always override.
5386 // If the cgroup subsystem is active and configured, we
5387 // will return the min of the cgroup and option 2 results.
5388 // This is required since tools, such as numactl, that
5389 // alter cpu affinity do not update cgroup subsystem
5390 // cpuset configuration files.
5391 int os::active_processor_count() {
5392 // User has overridden the number of active processors
5393 if (ActiveProcessorCount > 0) {
5394 log_trace(os)("active_processor_count: "
5395 "active processor count set by user : %d",
5396 ActiveProcessorCount);
5397 return ActiveProcessorCount;
5398 }
5399
5400 int active_cpus;
5401 if (OSContainer::is_containerized()) {
5402 active_cpus = OSContainer::active_processor_count();
5403 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5404 active_cpus);
5405 } else {
5406 active_cpus = os::Linux::active_processor_count();
5407 }
5408
5409 return active_cpus;
5410 }
5411
5412 uint os::processor_id() {
5413 const int id = Linux::sched_getcpu();
5414 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5415 return (uint)id;
5416 }
5417
5418 void os::set_native_thread_name(const char *name) {
5419 if (Linux::_pthread_setname_np) {
5420 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5421 snprintf(buf, sizeof(buf), "%s", name);
5422 buf[sizeof(buf) - 1] = '\0';
5423 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5424 // ERANGE should not happen; all other errors should just be ignored.
5425 assert(rc != ERANGE, "pthread_setname_np failed");
5426 }
5427 }
5428
5429 bool os::bind_to_processor(uint processor_id) {
5430 // Not yet implemented.
5431 return false;
5432 }
5433
5434 ///
5435
5436 void os::SuspendedThreadTask::internal_do_task() {
5437 if (do_suspend(_thread->osthread())) {
5438 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5439 do_task(context);
5440 do_resume(_thread->osthread());
5441 }
5442 }
5443
5444 ////////////////////////////////////////////////////////////////////////////////
5445 // debug support
5446
5447 bool os::find(address addr, outputStream* st) {
5448 Dl_info dlinfo;
5449 memset(&dlinfo, 0, sizeof(dlinfo));
5450 if (dladdr(addr, &dlinfo) != 0) {
5451 st->print(PTR_FORMAT ": ", p2i(addr));
5452 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5453 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5454 p2i(addr) - p2i(dlinfo.dli_saddr));
5455 } else if (dlinfo.dli_fbase != NULL) {
5456 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5457 } else {
5458 st->print("<absolute address>");
5459 }
5460 if (dlinfo.dli_fname != NULL) {
5461 st->print(" in %s", dlinfo.dli_fname);
5462 }
5463 if (dlinfo.dli_fbase != NULL) {
5464 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5465 }
5466 st->cr();
5467
5468 if (Verbose) {
5469 // decode some bytes around the PC
5470 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5471 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5472 address lowest = (address) dlinfo.dli_sname;
5473 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5474 if (begin < lowest) begin = lowest;
5475 Dl_info dlinfo2;
5476 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5477 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5478 end = (address) dlinfo2.dli_saddr;
5479 }
5480 Disassembler::decode(begin, end, st);
5481 }
5482 return true;
5483 }
5484 return false;
5485 }
5486
5487 ////////////////////////////////////////////////////////////////////////////////
5488 // misc
5489
5490 // This does not do anything on Linux. This is basically a hook for being
5491 // able to use structured exception handling (thread-local exception filters)
5492 // on, e.g., Win32.
5493 void
5494 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5495 JavaCallArguments* args, Thread* thread) {
5496 f(value, method, args, thread);
5497 }
5498
5499 void os::print_statistics() {
5500 }
5501
5502 bool os::message_box(const char* title, const char* message) {
5503 int i;
5504 fdStream err(defaultStream::error_fd());
5505 for (i = 0; i < 78; i++) err.print_raw("=");
5506 err.cr();
5507 err.print_raw_cr(title);
5508 for (i = 0; i < 78; i++) err.print_raw("-");
5509 err.cr();
5510 err.print_raw_cr(message);
5511 for (i = 0; i < 78; i++) err.print_raw("=");
5512 err.cr();
5513
5514 char buf[16];
5515 // Prevent process from exiting upon "read error" without consuming all CPU
5516 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5517
5518 return buf[0] == 'y' || buf[0] == 'Y';
5519 }
5520
5521 // Is a (classpath) directory empty?
5522 bool os::dir_is_empty(const char* path) {
5523 DIR *dir = NULL;
5524 struct dirent *ptr;
5525
5526 dir = opendir(path);
5527 if (dir == NULL) return true;
5528
5529 // Scan the directory
5530 bool result = true;
5531 while (result && (ptr = readdir(dir)) != NULL) {
5532 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5533 result = false;
5534 }
5535 }
5536 closedir(dir);
5537 return result;
5538 }
5539
5540 // This code originates from JDK's sysOpen and open64_w
5541 // from src/solaris/hpi/src/system_md.c
5542
5543 int os::open(const char *path, int oflag, int mode) {
5544 if (strlen(path) > MAX_PATH - 1) {
5545 errno = ENAMETOOLONG;
5546 return -1;
5547 }
5548
5549 // All file descriptors that are opened in the Java process and not
5550 // specifically destined for a subprocess should have the close-on-exec
5551 // flag set. If we don't set it, then careless 3rd party native code
5552 // might fork and exec without closing all appropriate file descriptors
5553 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5554 // turn might:
5555 //
5556 // - cause end-of-file to fail to be detected on some file
5557 // descriptors, resulting in mysterious hangs, or
5558 //
5559 // - might cause an fopen in the subprocess to fail on a system
5560 // suffering from bug 1085341.
5561 //
5562 // (Yes, the default setting of the close-on-exec flag is a Unix
5563 // design flaw)
5564 //
5565 // See:
5566 // 1085341: 32-bit stdio routines should support file descriptors >255
5567 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5568 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5569 //
5570 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5571 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5572 // because it saves a system call and removes a small window where the flag
5573 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5574 // and we fall back to using FD_CLOEXEC (see below).
5575 #ifdef O_CLOEXEC
5576 oflag |= O_CLOEXEC;
5577 #endif
5578
5579 int fd = ::open64(path, oflag, mode);
5580 if (fd == -1) return -1;
5581
5582 //If the open succeeded, the file might still be a directory
5583 {
5584 struct stat64 buf64;
5585 int ret = ::fstat64(fd, &buf64);
5586 int st_mode = buf64.st_mode;
5587
5588 if (ret != -1) {
5589 if ((st_mode & S_IFMT) == S_IFDIR) {
5590 errno = EISDIR;
5591 ::close(fd);
5592 return -1;
5593 }
5594 } else {
5595 ::close(fd);
5596 return -1;
5597 }
5598 }
5599
5600 #ifdef FD_CLOEXEC
5601 // Validate that the use of the O_CLOEXEC flag on open above worked.
5602 // With recent kernels, we will perform this check exactly once.
5603 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5604 if (!O_CLOEXEC_is_known_to_work) {
5605 int flags = ::fcntl(fd, F_GETFD);
5606 if (flags != -1) {
5607 if ((flags & FD_CLOEXEC) != 0)
5608 O_CLOEXEC_is_known_to_work = 1;
5609 else
5610 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5611 }
5612 }
5613 #endif
5614
5615 return fd;
5616 }
5617
5618
5619 // create binary file, rewriting existing file if required
5620 int os::create_binary_file(const char* path, bool rewrite_existing) {
5621 int oflags = O_WRONLY | O_CREAT;
5622 if (!rewrite_existing) {
5623 oflags |= O_EXCL;
5624 }
5625 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5626 }
5627
5628 // return current position of file pointer
5629 jlong os::current_file_offset(int fd) {
5630 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5631 }
5632
5633 // move file pointer to the specified offset
5634 jlong os::seek_to_file_offset(int fd, jlong offset) {
5635 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5636 }
5637
5638 // This code originates from JDK's sysAvailable
5639 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5640
5641 int os::available(int fd, jlong *bytes) {
5642 jlong cur, end;
5643 int mode;
5644 struct stat64 buf64;
5645
5646 if (::fstat64(fd, &buf64) >= 0) {
5647 mode = buf64.st_mode;
5648 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5649 int n;
5650 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5651 *bytes = n;
5652 return 1;
5653 }
5654 }
5655 }
5656 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5657 return 0;
5658 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5659 return 0;
5660 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5661 return 0;
5662 }
5663 *bytes = end - cur;
5664 return 1;
5665 }
5666
5667 // Map a block of memory.
5668 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5669 char *addr, size_t bytes, bool read_only,
5670 bool allow_exec) {
5671 int prot;
5672 int flags = MAP_PRIVATE;
5673
5674 if (read_only) {
5675 prot = PROT_READ;
5676 } else {
5677 prot = PROT_READ | PROT_WRITE;
5678 }
5679
5680 if (allow_exec) {
5681 prot |= PROT_EXEC;
5682 }
5683
5684 if (addr != NULL) {
5685 flags |= MAP_FIXED;
5686 }
5687
5688 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5689 fd, file_offset);
5690 if (mapped_address == MAP_FAILED) {
5691 return NULL;
5692 }
5693 return mapped_address;
5694 }
5695
5696
5697 // Remap a block of memory.
5698 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5699 char *addr, size_t bytes, bool read_only,
5700 bool allow_exec) {
5701 // same as map_memory() on this OS
5702 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5703 allow_exec);
5704 }
5705
5706
5707 // Unmap a block of memory.
5708 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5709 return munmap(addr, bytes) == 0;
5710 }
5711
5712 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5713
5714 static jlong fast_cpu_time(Thread *thread) {
5715 clockid_t clockid;
5716 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5717 &clockid);
5718 if (rc == 0) {
5719 return os::Linux::fast_thread_cpu_time(clockid);
5720 } else {
5721 // It's possible to encounter a terminated native thread that failed
5722 // to detach itself from the VM - which should result in ESRCH.
5723 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5724 return -1;
5725 }
5726 }
5727
5728 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5729 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5730 // of a thread.
5731 //
5732 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5733 // the fast estimate available on the platform.
5734
5735 jlong os::current_thread_cpu_time() {
5736 if (os::Linux::supports_fast_thread_cpu_time()) {
5737 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5738 } else {
5739 // return user + sys since the cost is the same
5740 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5741 }
5742 }
5743
5744 jlong os::thread_cpu_time(Thread* thread) {
5745 // consistent with what current_thread_cpu_time() returns
5746 if (os::Linux::supports_fast_thread_cpu_time()) {
5747 return fast_cpu_time(thread);
5748 } else {
5749 return slow_thread_cpu_time(thread, true /* user + sys */);
5750 }
5751 }
5752
5753 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5754 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5755 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5756 } else {
5757 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5758 }
5759 }
5760
5761 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5762 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5763 return fast_cpu_time(thread);
5764 } else {
5765 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5766 }
5767 }
5768
5769 // -1 on error.
5770 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5771 pid_t tid = thread->osthread()->thread_id();
5772 char *s;
5773 char stat[2048];
5774 int statlen;
5775 char proc_name[64];
5776 int count;
5777 long sys_time, user_time;
5778 char cdummy;
5779 int idummy;
5780 long ldummy;
5781 FILE *fp;
5782
5783 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5784 fp = fopen(proc_name, "r");
5785 if (fp == NULL) return -1;
5786 statlen = fread(stat, 1, 2047, fp);
5787 stat[statlen] = '\0';
5788 fclose(fp);
5789
5790 // Skip pid and the command string. Note that we could be dealing with
5791 // weird command names, e.g. user could decide to rename java launcher
5792 // to "java 1.4.2 :)", then the stat file would look like
5793 // 1234 (java 1.4.2 :)) R ... ...
5794 // We don't really need to know the command string, just find the last
5795 // occurrence of ")" and then start parsing from there. See bug 4726580.
5796 s = strrchr(stat, ')');
5797 if (s == NULL) return -1;
5798
5799 // Skip blank chars
5800 do { s++; } while (s && isspace(*s));
5801
5802 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5803 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5804 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5805 &user_time, &sys_time);
5806 if (count != 13) return -1;
5807 if (user_sys_cpu_time) {
5808 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5809 } else {
5810 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5811 }
5812 }
5813
5814 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5815 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5816 info_ptr->may_skip_backward = false; // elapsed time not wall time
5817 info_ptr->may_skip_forward = false; // elapsed time not wall time
5818 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5819 }
5820
5821 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5822 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5823 info_ptr->may_skip_backward = false; // elapsed time not wall time
5824 info_ptr->may_skip_forward = false; // elapsed time not wall time
5825 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5826 }
5827
5828 bool os::is_thread_cpu_time_supported() {
5829 return true;
5830 }
5831
5832 // System loadavg support. Returns -1 if load average cannot be obtained.
5833 // Linux doesn't yet have a (official) notion of processor sets,
5834 // so just return the system wide load average.
5835 int os::loadavg(double loadavg[], int nelem) {
5836 return ::getloadavg(loadavg, nelem);
5837 }
5838
5839 void os::pause() {
5840 char filename[MAX_PATH];
5841 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5842 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5843 } else {
5844 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5845 }
5846
5847 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5848 if (fd != -1) {
5849 struct stat buf;
5850 ::close(fd);
5851 while (::stat(filename, &buf) == 0) {
5852 (void)::poll(NULL, 0, 100);
5853 }
5854 } else {
5855 jio_fprintf(stderr,
5856 "Could not open pause file '%s', continuing immediately.\n", filename);
5857 }
5858 }
5859
5860 extern char** environ;
5861
5862 // Run the specified command in a separate process. Return its exit value,
5863 // or -1 on failure (e.g. can't fork a new process).
5864 // Unlike system(), this function can be called from signal handler. It
5865 // doesn't block SIGINT et al.
5866 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5867 const char * argv[4] = {"sh", "-c", cmd, NULL};
5868
5869 pid_t pid ;
5870
5871 if (use_vfork_if_available) {
5872 pid = vfork();
5873 } else {
5874 pid = fork();
5875 }
5876
5877 if (pid < 0) {
5878 // fork failed
5879 return -1;
5880
5881 } else if (pid == 0) {
5882 // child process
5883
5884 execve("/bin/sh", (char* const*)argv, environ);
5885
5886 // execve failed
5887 _exit(-1);
5888
5889 } else {
5890 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5891 // care about the actual exit code, for now.
5892
5893 int status;
5894
5895 // Wait for the child process to exit. This returns immediately if
5896 // the child has already exited. */
5897 while (waitpid(pid, &status, 0) < 0) {
5898 switch (errno) {
5899 case ECHILD: return 0;
5900 case EINTR: break;
5901 default: return -1;
5902 }
5903 }
5904
5905 if (WIFEXITED(status)) {
5906 // The child exited normally; get its exit code.
5907 return WEXITSTATUS(status);
5908 } else if (WIFSIGNALED(status)) {
5909 // The child exited because of a signal
5910 // The best value to return is 0x80 + signal number,
5911 // because that is what all Unix shells do, and because
5912 // it allows callers to distinguish between process exit and
5913 // process death by signal.
5914 return 0x80 + WTERMSIG(status);
5915 } else {
5916 // Unknown exit code; pass it through
5917 return status;
5918 }
5919 }
5920 }
5921
5922 // Get the default path to the core file
5923 // Returns the length of the string
5924 int os::get_core_path(char* buffer, size_t bufferSize) {
5925 /*
5926 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5927 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5928 */
5929 const int core_pattern_len = 129;
5930 char core_pattern[core_pattern_len] = {0};
5931
5932 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5933 if (core_pattern_file == -1) {
5934 return -1;
5935 }
5936
5937 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5938 ::close(core_pattern_file);
5939 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5940 return -1;
5941 }
5942 if (core_pattern[ret-1] == '\n') {
5943 core_pattern[ret-1] = '\0';
5944 } else {
5945 core_pattern[ret] = '\0';
5946 }
5947
5948 // Replace the %p in the core pattern with the process id. NOTE: we do this
5949 // only if the pattern doesn't start with "|", and we support only one %p in
5950 // the pattern.
5951 char *pid_pos = strstr(core_pattern, "%p");
5952 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5953 int written;
5954
5955 if (core_pattern[0] == '/') {
5956 if (pid_pos != NULL) {
5957 *pid_pos = '\0';
5958 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5959 current_process_id(), tail);
5960 } else {
5961 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5962 }
5963 } else {
5964 char cwd[PATH_MAX];
5965
5966 const char* p = get_current_directory(cwd, PATH_MAX);
5967 if (p == NULL) {
5968 return -1;
5969 }
5970
5971 if (core_pattern[0] == '|') {
5972 written = jio_snprintf(buffer, bufferSize,
5973 "\"%s\" (or dumping to %s/core.%d)",
5974 &core_pattern[1], p, current_process_id());
5975 } else if (pid_pos != NULL) {
5976 *pid_pos = '\0';
5977 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5978 current_process_id(), tail);
5979 } else {
5980 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5981 }
5982 }
5983
5984 if (written < 0) {
5985 return -1;
5986 }
5987
5988 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5989 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5990
5991 if (core_uses_pid_file != -1) {
5992 char core_uses_pid = 0;
5993 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5994 ::close(core_uses_pid_file);
5995
5996 if (core_uses_pid == '1') {
5997 jio_snprintf(buffer + written, bufferSize - written,
5998 ".%d", current_process_id());
5999 }
6000 }
6001 }
6002
6003 return strlen(buffer);
6004 }
6005
6006 bool os::start_debugging(char *buf, int buflen) {
6007 int len = (int)strlen(buf);
6008 char *p = &buf[len];
6009
6010 jio_snprintf(p, buflen-len,
6011 "\n\n"
6012 "Do you want to debug the problem?\n\n"
6013 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6014 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6015 "Otherwise, press RETURN to abort...",
6016 os::current_process_id(), os::current_process_id(),
6017 os::current_thread_id(), os::current_thread_id());
6018
6019 bool yes = os::message_box("Unexpected Error", buf);
6020
6021 if (yes) {
6022 // yes, user asked VM to launch debugger
6023 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6024 os::current_process_id(), os::current_process_id());
6025
6026 os::fork_and_exec(buf);
6027 yes = false;
6028 }
6029 return yes;
6030 }
6031
6032
6033 // Java/Compiler thread:
6034 //
6035 // Low memory addresses
6036 // P0 +------------------------+
6037 // | |\ Java thread created by VM does not have glibc
6038 // | glibc guard page | - guard page, attached Java thread usually has
6039 // | |/ 1 glibc guard page.
6040 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6041 // | |\
6042 // | HotSpot Guard Pages | - red, yellow and reserved pages
6043 // | |/
6044 // +------------------------+ JavaThread::stack_reserved_zone_base()
6045 // | |\
6046 // | Normal Stack | -
6047 // | |/
6048 // P2 +------------------------+ Thread::stack_base()
6049 //
6050 // Non-Java thread:
6051 //
6052 // Low memory addresses
6053 // P0 +------------------------+
6054 // | |\
6055 // | glibc guard page | - usually 1 page
6056 // | |/
6057 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6058 // | |\
6059 // | Normal Stack | -
6060 // | |/
6061 // P2 +------------------------+ Thread::stack_base()
6062 //
6063 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6064 // returned from pthread_attr_getstack().
6065 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6066 // of the stack size given in pthread_attr. We work around this for
6067 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6068 //
6069 #ifndef ZERO
6070 static void current_stack_region(address * bottom, size_t * size) {
6071 if (os::is_primordial_thread()) {
6072 // primordial thread needs special handling because pthread_getattr_np()
6073 // may return bogus value.
6074 *bottom = os::Linux::initial_thread_stack_bottom();
6075 *size = os::Linux::initial_thread_stack_size();
6076 } else {
6077 pthread_attr_t attr;
6078
6079 int rslt = pthread_getattr_np(pthread_self(), &attr);
6080
6081 // JVM needs to know exact stack location, abort if it fails
6082 if (rslt != 0) {
6083 if (rslt == ENOMEM) {
6084 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6085 } else {
6086 fatal("pthread_getattr_np failed with error = %d", rslt);
6087 }
6088 }
6089
6090 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6091 fatal("Cannot locate current stack attributes!");
6092 }
6093
6094 // Work around NPTL stack guard error.
6095 size_t guard_size = 0;
6096 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6097 if (rslt != 0) {
6098 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6099 }
6100 *bottom += guard_size;
6101 *size -= guard_size;
6102
6103 pthread_attr_destroy(&attr);
6104
6105 }
6106 assert(os::current_stack_pointer() >= *bottom &&
6107 os::current_stack_pointer() < *bottom + *size, "just checking");
6108 }
6109
6110 address os::current_stack_base() {
6111 address bottom;
6112 size_t size;
6113 current_stack_region(&bottom, &size);
6114 return (bottom + size);
6115 }
6116
6117 size_t os::current_stack_size() {
6118 // This stack size includes the usable stack and HotSpot guard pages
6119 // (for the threads that have Hotspot guard pages).
6120 address bottom;
6121 size_t size;
6122 current_stack_region(&bottom, &size);
6123 return size;
6124 }
6125 #endif
6126
6127 static inline struct timespec get_mtime(const char* filename) {
6128 struct stat st;
6129 int ret = os::stat(filename, &st);
6130 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6131 return st.st_mtim;
6132 }
6133
6134 int os::compare_file_modified_times(const char* file1, const char* file2) {
6135 struct timespec filetime1 = get_mtime(file1);
6136 struct timespec filetime2 = get_mtime(file2);
6137 int diff = filetime1.tv_sec - filetime2.tv_sec;
6138 if (diff == 0) {
6139 return filetime1.tv_nsec - filetime2.tv_nsec;
6140 }
6141 return diff;
6142 }
6143
6144 bool os::supports_map_sync() {
6145 return true;
6146 }
6147
6148 /////////////// Unit tests ///////////////
6149
6150 #ifndef PRODUCT
6151
6152 class TestReserveMemorySpecial : AllStatic {
6153 public:
6154 static void small_page_write(void* addr, size_t size) {
6155 size_t page_size = os::vm_page_size();
6156
6157 char* end = (char*)addr + size;
6158 for (char* p = (char*)addr; p < end; p += page_size) {
6159 *p = 1;
6160 }
6161 }
6162
6163 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6164 if (!UseHugeTLBFS) {
6165 return;
6166 }
6167
6168 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6169
6170 if (addr != NULL) {
6171 small_page_write(addr, size);
6172
6173 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6174 }
6175 }
6176
6177 static void test_reserve_memory_special_huge_tlbfs_only() {
6178 if (!UseHugeTLBFS) {
6179 return;
6180 }
6181
6182 size_t lp = os::large_page_size();
6183
6184 for (size_t size = lp; size <= lp * 10; size += lp) {
6185 test_reserve_memory_special_huge_tlbfs_only(size);
6186 }
6187 }
6188
6189 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6190 size_t lp = os::large_page_size();
6191 size_t ag = os::vm_allocation_granularity();
6192
6193 // sizes to test
6194 const size_t sizes[] = {
6195 lp, lp + ag, lp + lp / 2, lp * 2,
6196 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6197 lp * 10, lp * 10 + lp / 2
6198 };
6199 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6200
6201 // For each size/alignment combination, we test three scenarios:
6202 // 1) with req_addr == NULL
6203 // 2) with a non-null req_addr at which we expect to successfully allocate
6204 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6205 // expect the allocation to either fail or to ignore req_addr
6206
6207 // Pre-allocate two areas; they shall be as large as the largest allocation
6208 // and aligned to the largest alignment we will be testing.
6209 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6210 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6211 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6212 -1, 0);
6213 assert(mapping1 != MAP_FAILED, "should work");
6214
6215 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6216 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6217 -1, 0);
6218 assert(mapping2 != MAP_FAILED, "should work");
6219
6220 // Unmap the first mapping, but leave the second mapping intact: the first
6221 // mapping will serve as a value for a "good" req_addr (case 2). The second
6222 // mapping, still intact, as "bad" req_addr (case 3).
6223 ::munmap(mapping1, mapping_size);
6224
6225 // Case 1
6226 for (int i = 0; i < num_sizes; i++) {
6227 const size_t size = sizes[i];
6228 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6229 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6230 if (p != NULL) {
6231 assert(is_aligned(p, alignment), "must be");
6232 small_page_write(p, size);
6233 os::Linux::release_memory_special_huge_tlbfs(p, size);
6234 }
6235 }
6236 }
6237
6238 // Case 2
6239 for (int i = 0; i < num_sizes; i++) {
6240 const size_t size = sizes[i];
6241 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6242 char* const req_addr = align_up(mapping1, alignment);
6243 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6244 if (p != NULL) {
6245 assert(p == req_addr, "must be");
6246 small_page_write(p, size);
6247 os::Linux::release_memory_special_huge_tlbfs(p, size);
6248 }
6249 }
6250 }
6251
6252 // Case 3
6253 for (int i = 0; i < num_sizes; i++) {
6254 const size_t size = sizes[i];
6255 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6256 char* const req_addr = align_up(mapping2, alignment);
6257 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6258 // as the area around req_addr contains already existing mappings, the API should always
6259 // return NULL (as per contract, it cannot return another address)
6260 assert(p == NULL, "must be");
6261 }
6262 }
6263
6264 ::munmap(mapping2, mapping_size);
6265
6266 }
6267
6268 static void test_reserve_memory_special_huge_tlbfs() {
6269 if (!UseHugeTLBFS) {
6270 return;
6271 }
6272
6273 test_reserve_memory_special_huge_tlbfs_only();
6274 test_reserve_memory_special_huge_tlbfs_mixed();
6275 }
6276
6277 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6278 if (!UseSHM) {
6279 return;
6280 }
6281
6282 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6283
6284 if (addr != NULL) {
6285 assert(is_aligned(addr, alignment), "Check");
6286 assert(is_aligned(addr, os::large_page_size()), "Check");
6287
6288 small_page_write(addr, size);
6289
6290 os::Linux::release_memory_special_shm(addr, size);
6291 }
6292 }
6293
6294 static void test_reserve_memory_special_shm() {
6295 size_t lp = os::large_page_size();
6296 size_t ag = os::vm_allocation_granularity();
6297
6298 for (size_t size = ag; size < lp * 3; size += ag) {
6299 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6300 test_reserve_memory_special_shm(size, alignment);
6301 }
6302 }
6303 }
6304
6305 static void test() {
6306 test_reserve_memory_special_huge_tlbfs();
6307 test_reserve_memory_special_shm();
6308 }
6309 };
6310
6311 void TestReserveMemorySpecial_test() {
6312 TestReserveMemorySpecial::test();
6313 }
6314
6315 #endif