1 /*
  2  * Copyright (c) 2008, 2018, Oracle and/or its affiliates. All rights reserved.
  3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  4  *
  5  * This code is free software; you can redistribute it and/or modify it
  6  * under the terms of the GNU General Public License version 2 only, as
  7  * published by the Free Software Foundation.
  8  *
  9  * This code is distributed in the hope that it will be useful, but WITHOUT
 10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 12  * version 2 for more details (a copy is included in the LICENSE file that
 13  * accompanied this code).
 14  *
 15  * You should have received a copy of the GNU General Public License version
 16  * 2 along with this work; if not, write to the Free Software Foundation,
 17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 18  *
 19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 20  * or visit www.oracle.com if you need additional information or have any
 21  * questions.
 22  *
 23  */
 24 
 25 // no precompiled headers
 26 #include "jvm.h"
 27 #include "assembler_arm.inline.hpp"
 28 #include "classfile/classLoader.hpp"
 29 #include "classfile/systemDictionary.hpp"
 30 #include "classfile/vmSymbols.hpp"
 31 #include "code/icBuffer.hpp"
 32 #include "code/vtableStubs.hpp"
 33 #include "interpreter/interpreter.hpp"
 34 #include "memory/allocation.inline.hpp"
 35 #include "nativeInst_arm.hpp"
 36 #include "os_share_linux.hpp"
 37 #include "prims/jniFastGetField.hpp"
 38 #include "prims/jvm_misc.hpp"
 39 #include "runtime/arguments.hpp"
 40 #include "runtime/extendedPC.hpp"
 41 #include "runtime/frame.inline.hpp"
 42 #include "runtime/interfaceSupport.inline.hpp"
 43 #include "runtime/java.hpp"
 44 #include "runtime/javaCalls.hpp"
 45 #include "runtime/mutexLocker.hpp"
 46 #include "runtime/osThread.hpp"
 47 #include "runtime/sharedRuntime.hpp"
 48 #include "runtime/stubRoutines.hpp"
 49 #include "runtime/timer.hpp"
 50 #include "utilities/debug.hpp"
 51 #include "utilities/events.hpp"
 52 #include "utilities/vmError.hpp"
 53 
 54 // put OS-includes here
 55 # include <sys/types.h>
 56 # include <sys/mman.h>
 57 # include <pthread.h>
 58 # include <signal.h>
 59 # include <errno.h>
 60 # include <dlfcn.h>
 61 # include <stdlib.h>
 62 # include <stdio.h>
 63 # include <unistd.h>
 64 # include <sys/resource.h>
 65 # include <pthread.h>
 66 # include <sys/stat.h>
 67 # include <sys/time.h>
 68 # include <sys/utsname.h>
 69 # include <sys/socket.h>
 70 # include <sys/wait.h>
 71 # include <pwd.h>
 72 # include <poll.h>
 73 # include <ucontext.h>
 74 # include <fpu_control.h>
 75 # include <asm/ptrace.h>
 76 
 77 #define SPELL_REG_SP  "sp"
 78 
 79 // Don't #define SPELL_REG_FP for thumb because it is not safe to use, so this makes sure we never fetch it.
 80 #ifndef __thumb__
 81 #define SPELL_REG_FP "fp"
 82 #endif
 83 
 84 address os::current_stack_pointer() {
 85   register address sp __asm__ (SPELL_REG_SP);
 86   return sp;
 87 }
 88 
 89 char* os::non_memory_address_word() {
 90   // Must never look like an address returned by reserve_memory
 91   return (char*) -1;
 92 }
 93 
 94 
 95 #if NGREG == 16
 96 // These definitions are based on the observation that until
 97 // the certain version of GCC mcontext_t was defined as
 98 // a structure containing gregs[NGREG] array with 16 elements.
 99 // In later GCC versions mcontext_t was redefined as struct sigcontext,
100 // along with NGREG constant changed to 18.
101 #define arm_pc gregs[15]
102 #define arm_sp gregs[13]
103 #define arm_fp gregs[11]
104 #define arm_r0 gregs[0]
105 #endif
106 
107 #define ARM_REGS_IN_CONTEXT  16
108 
109 
110 address os::Linux::ucontext_get_pc(const ucontext_t* uc) {
111   return (address)uc->uc_mcontext.arm_pc;
112 }
113 
114 void os::Linux::ucontext_set_pc(ucontext_t* uc, address pc) {
115   uc->uc_mcontext.arm_pc = (uintx)pc;
116 }
117 
118 intptr_t* os::Linux::ucontext_get_sp(const ucontext_t* uc) {
119   return (intptr_t*)uc->uc_mcontext.arm_sp;
120 }
121 
122 intptr_t* os::Linux::ucontext_get_fp(const ucontext_t* uc) {
123   return (intptr_t*)uc->uc_mcontext.arm_fp;
124 }
125 
126 bool is_safe_for_fp(address pc) {
127 #ifdef __thumb__
128   if (CodeCache::find_blob(pc) != NULL) {
129     return true;
130   }
131   // For thumb C frames, given an fp we have no idea how to access the frame contents.
132   return false;
133 #else
134   // Calling os::address_is_in_vm() here leads to a dladdr call. Calling any libc
135   // function during os::get_native_stack() can result in a deadlock if JFR is
136   // enabled. For now, be more lenient and allow all pc's. There are other
137   // frame sanity checks in shared code, and to date they have been sufficient
138   // for other platforms.
139   //return os::address_is_in_vm(pc);
140   return true;
141 #endif
142 }
143 
144 // For Forte Analyzer AsyncGetCallTrace profiling support - thread
145 // is currently interrupted by SIGPROF.
146 // os::Solaris::fetch_frame_from_ucontext() tries to skip nested signal
147 // frames. Currently we don't do that on Linux, so it's the same as
148 // os::fetch_frame_from_context().
149 ExtendedPC os::Linux::fetch_frame_from_ucontext(Thread* thread,
150   const ucontext_t* uc, intptr_t** ret_sp, intptr_t** ret_fp) {
151 
152   assert(thread != NULL, "just checking");
153   assert(ret_sp != NULL, "just checking");
154   assert(ret_fp != NULL, "just checking");
155 
156   return os::fetch_frame_from_context(uc, ret_sp, ret_fp);
157 }
158 
159 ExtendedPC os::fetch_frame_from_context(const void* ucVoid,
160                     intptr_t** ret_sp, intptr_t** ret_fp) {
161 
162   ExtendedPC  epc;
163   const ucontext_t* uc = (const ucontext_t*)ucVoid;
164 
165   if (uc != NULL) {
166     epc = ExtendedPC(os::Linux::ucontext_get_pc(uc));
167     if (ret_sp) *ret_sp = os::Linux::ucontext_get_sp(uc);
168     if (ret_fp) {
169       intptr_t* fp = os::Linux::ucontext_get_fp(uc);
170 #ifndef __thumb__
171       if (CodeCache::find_blob(epc.pc()) == NULL) {
172         // It's a C frame. We need to adjust the fp.
173         fp += os::C_frame_offset;
174       }
175 #endif
176       // Clear FP when stack walking is dangerous so that
177       // the frame created will not be walked.
178       // However, ensure FP is set correctly when reliable and
179       // potentially necessary.
180       if (!is_safe_for_fp(epc.pc())) {
181         // FP unreliable
182         fp = (intptr_t *)NULL;
183       }
184       *ret_fp = fp;
185     }
186   } else {
187     // construct empty ExtendedPC for return value checking
188     epc = ExtendedPC(NULL);
189     if (ret_sp) *ret_sp = (intptr_t *)NULL;
190     if (ret_fp) *ret_fp = (intptr_t *)NULL;
191   }
192 
193   return epc;
194 }
195 
196 frame os::fetch_frame_from_context(const void* ucVoid) {
197   intptr_t* sp;
198   intptr_t* fp;
199   ExtendedPC epc = fetch_frame_from_context(ucVoid, &sp, &fp);
200   return frame(sp, fp, epc.pc());
201 }
202 
203 frame os::get_sender_for_C_frame(frame* fr) {
204 #ifdef __thumb__
205   // We can't reliably get anything from a thumb C frame.
206   return frame();
207 #else
208   address pc = fr->sender_pc();
209   if (! is_safe_for_fp(pc)) {
210     return frame(fr->sender_sp(), (intptr_t *)NULL, pc);
211   } else {
212     return frame(fr->sender_sp(), fr->link() + os::C_frame_offset, pc);
213   }
214 #endif
215 }
216 
217 //
218 // This actually returns two frames up. It does not return os::current_frame(),
219 // which is the actual current frame. Nor does it return os::get_native_stack(),
220 // which is the caller. It returns whoever called os::get_native_stack(). Not
221 // very intuitive, but consistent with how this API is implemented on other
222 // platforms.
223 //
224 frame os::current_frame() {
225 #ifdef __thumb__
226   // We can't reliably get anything from a thumb C frame.
227   return frame();
228 #else
229   register intptr_t* fp __asm__ (SPELL_REG_FP);
230   // fp is for os::current_frame. We want the fp for our caller.
231   frame myframe((intptr_t*)os::current_stack_pointer(), fp + os::C_frame_offset,
232                  CAST_FROM_FN_PTR(address, os::current_frame));
233   frame caller_frame = os::get_sender_for_C_frame(&myframe);
234 
235   if (os::is_first_C_frame(&caller_frame)) {
236     // stack is not walkable
237     // Assert below was added because it does not seem like this can ever happen.
238     // How can this frame ever be the first C frame since it is called from C code?
239     // If it does ever happen, undo the assert and comment here on when/why it happens.
240     assert(false, "this should never happen");
241     return frame();
242   }
243 
244   // return frame for our caller's caller
245   return os::get_sender_for_C_frame(&caller_frame);
246 #endif
247 }
248 
249 extern "C" address check_vfp_fault_instr;
250 extern "C" address check_vfp3_32_fault_instr;
251 
252 address check_vfp_fault_instr = NULL;
253 address check_vfp3_32_fault_instr = NULL;
254 extern "C" address check_simd_fault_instr;
255 address check_simd_fault_instr = NULL;
256 
257 // Utility functions
258 
259 extern "C" int JVM_handle_linux_signal(int sig, siginfo_t* info,
260                                        void* ucVoid, int abort_if_unrecognized) {
261   ucontext_t* uc = (ucontext_t*) ucVoid;
262 
263   Thread* t = Thread::current_or_null_safe();
264 
265   // Must do this before SignalHandlerMark, if crash protection installed we will longjmp away
266   // (no destructors can be run)
267   os::ThreadCrashProtection::check_crash_protection(sig, t);
268 
269   SignalHandlerMark shm(t);
270 
271   if (sig == SIGILL &&
272       ((info->si_addr == (caddr_t)check_simd_fault_instr)
273        || info->si_addr == (caddr_t)check_vfp_fault_instr
274        || info->si_addr == (caddr_t)check_vfp3_32_fault_instr)) {
275     // skip faulty instruction + instruction that sets return value to
276     // success and set return value to failure.
277     os::Linux::ucontext_set_pc(uc, (address)info->si_addr + 8);
278     uc->uc_mcontext.arm_r0 = 0;
279     return true;
280   }
281 
282   // Note: it's not uncommon that JNI code uses signal/sigset to install
283   // then restore certain signal handler (e.g. to temporarily block SIGPIPE,
284   // or have a SIGILL handler when detecting CPU type). When that happens,
285   // JVM_handle_linux_signal() might be invoked with junk info/ucVoid. To
286   // avoid unnecessary crash when libjsig is not preloaded, try handle signals
287   // that do not require siginfo/ucontext first.
288 
289   if (sig == SIGPIPE || sig == SIGXFSZ) {
290     // allow chained handler to go first
291     if (os::Linux::chained_handler(sig, info, ucVoid)) {
292       return true;
293     } else {
294       // Ignoring SIGPIPE/SIGXFSZ - see bugs 4229104 or 6499219
295       return true;
296     }
297   }
298 
299 #ifdef CAN_SHOW_REGISTERS_ON_ASSERT
300   if ((sig == SIGSEGV || sig == SIGBUS) && info != NULL && info->si_addr == g_assert_poison) {
301     handle_assert_poison_fault(ucVoid, info->si_addr);
302     return 1;
303   }
304 #endif
305 
306   JavaThread* thread = NULL;
307   VMThread* vmthread = NULL;
308   if (os::Linux::signal_handlers_are_installed) {
309     if (t != NULL ){
310       if(t->is_Java_thread()) {
311         thread = (JavaThread*)t;
312       }
313       else if(t->is_VM_thread()){
314         vmthread = (VMThread *)t;
315       }
316     }
317   }
318 
319   address stub = NULL;
320   address pc = NULL;
321   bool unsafe_access = false;
322 
323   if (info != NULL && uc != NULL && thread != NULL) {
324     pc = (address) os::Linux::ucontext_get_pc(uc);
325 
326     // Handle ALL stack overflow variations here
327     if (sig == SIGSEGV) {
328       address addr = (address) info->si_addr;
329 
330       if (StubRoutines::is_safefetch_fault(pc)) {
331         os::Linux::ucontext_set_pc(uc, StubRoutines::continuation_for_safefetch_fault(pc));
332         return 1;
333       }
334       // check if fault address is within thread stack
335       if (addr < thread->stack_base() &&
336           addr >= thread->stack_base() - thread->stack_size()) {
337         // stack overflow
338         if (thread->in_stack_yellow_reserved_zone(addr)) {
339           thread->disable_stack_yellow_reserved_zone();
340           if (thread->thread_state() == _thread_in_Java) {
341             // Throw a stack overflow exception.  Guard pages will be reenabled
342             // while unwinding the stack.
343             stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::STACK_OVERFLOW);
344           } else {
345             // Thread was in the vm or native code.  Return and try to finish.
346             return 1;
347           }
348         } else if (thread->in_stack_red_zone(addr)) {
349           // Fatal red zone violation.  Disable the guard pages and fall through
350           // to handle_unexpected_exception way down below.
351           thread->disable_stack_red_zone();
352           tty->print_raw_cr("An irrecoverable stack overflow has occurred.");
353         } else {
354           // Accessing stack address below sp may cause SEGV if current
355           // thread has MAP_GROWSDOWN stack. This should only happen when
356           // current thread was created by user code with MAP_GROWSDOWN flag
357           // and then attached to VM. See notes in os_linux.cpp.
358           if (thread->osthread()->expanding_stack() == 0) {
359              thread->osthread()->set_expanding_stack();
360              if (os::Linux::manually_expand_stack(thread, addr)) {
361                thread->osthread()->clear_expanding_stack();
362                return 1;
363              }
364              thread->osthread()->clear_expanding_stack();
365           } else {
366              fatal("recursive segv. expanding stack.");
367           }
368         }
369       }
370     }
371 
372     if (thread->thread_state() == _thread_in_Java) {
373       // Java thread running in Java code => find exception handler if any
374       // a fault inside compiled code, the interpreter, or a stub
375 
376       if (sig == SIGSEGV && os::is_poll_address((address)info->si_addr)) {
377         stub = SharedRuntime::get_poll_stub(pc);
378       } else if (sig == SIGBUS) {
379         // BugId 4454115: A read from a MappedByteBuffer can fault
380         // here if the underlying file has been truncated.
381         // Do not crash the VM in such a case.
382         CodeBlob* cb = CodeCache::find_blob_unsafe(pc);
383         CompiledMethod* nm = (cb != NULL) ? cb->as_compiled_method_or_null() : NULL;
384         if (nm != NULL && nm->has_unsafe_access()) {
385           unsafe_access = true;
386         }
387       } else if (sig == SIGSEGV &&
388                  MacroAssembler::uses_implicit_null_check(info->si_addr)) {
389           // Determination of interpreter/vtable stub/compiled code null exception
390           CodeBlob* cb = CodeCache::find_blob_unsafe(pc);
391           if (cb != NULL) {
392             stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_NULL);
393           }
394       } else if (sig == SIGILL && *(int *)pc == NativeInstruction::zombie_illegal_instruction) {
395         // Zombie
396         stub = SharedRuntime::get_handle_wrong_method_stub();
397       }
398     } else if (thread->thread_state() == _thread_in_vm &&
399                sig == SIGBUS && thread->doing_unsafe_access()) {
400         unsafe_access = true;
401     }
402 
403     // jni_fast_Get<Primitive>Field can trap at certain pc's if a GC kicks in
404     // and the heap gets shrunk before the field access.
405     if (sig == SIGSEGV || sig == SIGBUS) {
406       address addr = JNI_FastGetField::find_slowcase_pc(pc);
407       if (addr != (address)-1) {
408         stub = addr;
409       }
410     }
411 
412     // Check to see if we caught the safepoint code in the
413     // process of write protecting the memory serialization page.
414     // It write enables the page immediately after protecting it
415     // so we can just return to retry the write.
416     if (sig == SIGSEGV && os::is_memory_serialize_page(thread, (address) info->si_addr)) {
417       // Block current thread until the memory serialize page permission restored.
418       os::block_on_serialize_page_trap();
419       return true;
420     }
421   }
422 
423   if (unsafe_access && stub == NULL) {
424     // it can be an unsafe access and we haven't found
425     // any other suitable exception reason,
426     // so assume it is an unsafe access.
427     address next_pc = pc + Assembler::InstructionSize;
428 #ifdef __thumb__
429     if (uc->uc_mcontext.arm_cpsr & PSR_T_BIT) {
430       next_pc = (address)((intptr_t)next_pc | 0x1);
431     }
432 #endif
433 
434     stub = SharedRuntime::handle_unsafe_access(thread, next_pc);
435   }
436 
437   if (stub != NULL) {
438 #ifdef __thumb__
439     if (uc->uc_mcontext.arm_cpsr & PSR_T_BIT) {
440       intptr_t p = (intptr_t)pc | 0x1;
441       pc = (address)p;
442 
443       // Clear Thumb mode bit if we're redirected into the ARM ISA based code
444       if (((intptr_t)stub & 0x1) == 0) {
445         uc->uc_mcontext.arm_cpsr &= ~PSR_T_BIT;
446       }
447     } else {
448       // No Thumb2 compiled stubs are triggered from ARM ISA compiled JIT'd code today.
449       // The support needs to be added if that changes
450       assert((((intptr_t)stub & 0x1) == 0), "can't return to Thumb code");
451     }
452 #endif
453 
454     // save all thread context in case we need to restore it
455     if (thread != NULL) thread->set_saved_exception_pc(pc);
456 
457     os::Linux::ucontext_set_pc(uc, stub);
458     return true;
459   }
460 
461   // signal-chaining
462   if (os::Linux::chained_handler(sig, info, ucVoid)) {
463      return true;
464   }
465 
466   if (!abort_if_unrecognized) {
467     // caller wants another chance, so give it to him
468     return false;
469   }
470 
471   if (pc == NULL && uc != NULL) {
472     pc = os::Linux::ucontext_get_pc(uc);
473   }
474 
475   // unmask current signal
476   sigset_t newset;
477   sigemptyset(&newset);
478   sigaddset(&newset, sig);
479   sigprocmask(SIG_UNBLOCK, &newset, NULL);
480 
481   VMError::report_and_die(t, sig, pc, info, ucVoid);
482 
483   ShouldNotReachHere();
484   return false;
485 }
486 
487 void os::Linux::init_thread_fpu_state(void) {
488   os::setup_fpu();
489 }
490 
491 int os::Linux::get_fpu_control_word(void) {
492   return 0;
493 }
494 
495 void os::Linux::set_fpu_control_word(int fpu_control) {
496   // Nothing to do
497 }
498 
499 void os::setup_fpu() {
500 #if !defined(__SOFTFP__) && defined(__VFP_FP__)
501   // Turn on IEEE-754 compliant VFP mode
502   __asm__ volatile (
503     "mov %%r0, #0;"
504     "fmxr fpscr, %%r0"
505     : /* no output */ : /* no input */ : "r0"
506   );
507 #endif
508 }
509 
510 bool os::is_allocatable(size_t bytes) {
511   return true;
512 }
513 
514 ////////////////////////////////////////////////////////////////////////////////
515 // thread stack
516 
517 // Minimum usable stack sizes required to get to user code. Space for
518 // HotSpot guard pages is added later.
519 size_t os::Posix::_compiler_thread_min_stack_allowed = (32 DEBUG_ONLY(+ 4)) * K;
520 size_t os::Posix::_java_thread_min_stack_allowed = (32 DEBUG_ONLY(+ 4)) * K;
521 size_t os::Posix::_vm_internal_thread_min_stack_allowed = (48 DEBUG_ONLY(+ 4)) * K;
522 
523 // return default stack size for thr_type
524 size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
525   // default stack size (compiler thread needs larger stack)
526   size_t s = (thr_type == os::compiler_thread ? 2 * M : 512 * K);
527   return s;
528 }
529 
530 /////////////////////////////////////////////////////////////////////////////
531 // helper functions for fatal error handler
532 
533 void os::print_context(outputStream *st, const void *context) {
534   if (context == NULL) return;
535   const ucontext_t *uc = (const ucontext_t*)context;
536 
537   st->print_cr("Registers:");
538   intx* reg_area = (intx*)&uc->uc_mcontext.arm_r0;
539   for (int r = 0; r < ARM_REGS_IN_CONTEXT; r++) {
540     st->print_cr("  %-3s = " INTPTR_FORMAT, as_Register(r)->name(), reg_area[r]);
541   }
542 #define U64_FORMAT "0x%016llx"
543   // now print flag register
544   st->print_cr("  %-4s = 0x%08lx", "cpsr",uc->uc_mcontext.arm_cpsr);
545   st->cr();
546 
547   intptr_t *sp = (intptr_t *)os::Linux::ucontext_get_sp(uc);
548   st->print_cr("Top of Stack: (sp=" INTPTR_FORMAT ")", p2i(sp));
549   print_hex_dump(st, (address)sp, (address)(sp + 8*sizeof(intptr_t)), sizeof(intptr_t));
550   st->cr();
551 
552   // Note: it may be unsafe to inspect memory near pc. For example, pc may
553   // point to garbage if entry point in an nmethod is corrupted. Leave
554   // this at the end, and hope for the best.
555   address pc = os::Linux::ucontext_get_pc(uc);
556   st->print_cr("Instructions: (pc=" INTPTR_FORMAT ")", p2i(pc));
557   print_hex_dump(st, pc - 32, pc + 32, Assembler::InstructionSize);
558 }
559 
560 void os::print_register_info(outputStream *st, const void *context) {
561   if (context == NULL) return;
562 
563   const ucontext_t *uc = (const ucontext_t*)context;
564   intx* reg_area = (intx*)&uc->uc_mcontext.arm_r0;
565 
566   st->print_cr("Register to memory mapping:");
567   st->cr();
568   for (int r = 0; r < ARM_REGS_IN_CONTEXT; r++) {
569     st->print_cr("  %-3s = " INTPTR_FORMAT, as_Register(r)->name(), reg_area[r]);
570     print_location(st, reg_area[r]);
571     st->cr();
572   }
573   st->cr();
574 }
575 
576 
577 
578 typedef int64_t cmpxchg_long_func_t(int64_t, int64_t, volatile int64_t*);
579 
580 cmpxchg_long_func_t* os::atomic_cmpxchg_long_func = os::atomic_cmpxchg_long_bootstrap;
581 
582 int64_t os::atomic_cmpxchg_long_bootstrap(int64_t compare_value, int64_t exchange_value, volatile int64_t* dest) {
583   // try to use the stub:
584   cmpxchg_long_func_t* func = CAST_TO_FN_PTR(cmpxchg_long_func_t*, StubRoutines::atomic_cmpxchg_long_entry());
585 
586   if (func != NULL) {
587     os::atomic_cmpxchg_long_func = func;
588     return (*func)(compare_value, exchange_value, dest);
589   }
590   assert(Threads::number_of_threads() == 0, "for bootstrap only");
591 
592   int64_t old_value = *dest;
593   if (old_value == compare_value)
594     *dest = exchange_value;
595   return old_value;
596 }
597 typedef int64_t load_long_func_t(const volatile int64_t*);
598 
599 load_long_func_t* os::atomic_load_long_func = os::atomic_load_long_bootstrap;
600 
601 int64_t os::atomic_load_long_bootstrap(const volatile int64_t* src) {
602   // try to use the stub:
603   load_long_func_t* func = CAST_TO_FN_PTR(load_long_func_t*, StubRoutines::atomic_load_long_entry());
604 
605   if (func != NULL) {
606     os::atomic_load_long_func = func;
607     return (*func)(src);
608   }
609   assert(Threads::number_of_threads() == 0, "for bootstrap only");
610 
611   int64_t old_value = *src;
612   return old_value;
613 }
614 
615 typedef void store_long_func_t(int64_t, volatile int64_t*);
616 
617 store_long_func_t* os::atomic_store_long_func = os::atomic_store_long_bootstrap;
618 
619 void os::atomic_store_long_bootstrap(int64_t val, volatile int64_t* dest) {
620   // try to use the stub:
621   store_long_func_t* func = CAST_TO_FN_PTR(store_long_func_t*, StubRoutines::atomic_store_long_entry());
622 
623   if (func != NULL) {
624     os::atomic_store_long_func = func;
625     return (*func)(val, dest);
626   }
627   assert(Threads::number_of_threads() == 0, "for bootstrap only");
628 
629   *dest = val;
630 }
631 
632 typedef int32_t  atomic_add_func_t(int32_t add_value, volatile int32_t *dest);
633 
634 atomic_add_func_t * os::atomic_add_func = os::atomic_add_bootstrap;
635 
636 int32_t  os::atomic_add_bootstrap(int32_t add_value, volatile int32_t *dest) {
637   atomic_add_func_t * func = CAST_TO_FN_PTR(atomic_add_func_t*,
638                                             StubRoutines::atomic_add_entry());
639   if (func != NULL) {
640     os::atomic_add_func = func;
641     return (*func)(add_value, dest);
642   }
643 
644   int32_t old_value = *dest;
645   *dest = old_value + add_value;
646   return (old_value + add_value);
647 }
648 
649 typedef int32_t  atomic_xchg_func_t(int32_t exchange_value, volatile int32_t *dest);
650 
651 atomic_xchg_func_t * os::atomic_xchg_func = os::atomic_xchg_bootstrap;
652 
653 int32_t  os::atomic_xchg_bootstrap(int32_t exchange_value, volatile int32_t *dest) {
654   atomic_xchg_func_t * func = CAST_TO_FN_PTR(atomic_xchg_func_t*,
655                                             StubRoutines::atomic_xchg_entry());
656   if (func != NULL) {
657     os::atomic_xchg_func = func;
658     return (*func)(exchange_value, dest);
659   }
660 
661   int32_t old_value = *dest;
662   *dest = exchange_value;
663   return (old_value);
664 }
665 
666 typedef int32_t cmpxchg_func_t(int32_t, int32_t, volatile int32_t*);
667 
668 cmpxchg_func_t* os::atomic_cmpxchg_func = os::atomic_cmpxchg_bootstrap;
669 
670 int32_t os::atomic_cmpxchg_bootstrap(int32_t compare_value, int32_t exchange_value, volatile int32_t* dest) {
671   // try to use the stub:
672   cmpxchg_func_t* func = CAST_TO_FN_PTR(cmpxchg_func_t*, StubRoutines::atomic_cmpxchg_entry());
673 
674   if (func != NULL) {
675     os::atomic_cmpxchg_func = func;
676     return (*func)(compare_value, exchange_value, dest);
677   }
678   assert(Threads::number_of_threads() == 0, "for bootstrap only");
679 
680   int32_t old_value = *dest;
681   if (old_value == compare_value)
682     *dest = exchange_value;
683   return old_value;
684 }
685 
686 
687 #ifndef PRODUCT
688 void os::verify_stack_alignment() {
689 }
690 #endif
691 
692 int os::extra_bang_size_in_bytes() {
693   // ARM does not require an additional stack bang.
694   return 0;
695 }