os_linux_x86.cpp revision 7426:0f6100dde08e
1/* 2 * Copyright (c) 1999, 2014, 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 "asm/macroAssembler.hpp" 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 "interpreter/interpreter.hpp" 33#include "jvm_linux.h" 34#include "memory/allocation.inline.hpp" 35#include "mutex_linux.inline.hpp" 36#include "os_share_linux.hpp" 37#include "prims/jniFastGetField.hpp" 38#include "prims/jvm.h" 39#include "prims/jvm_misc.hpp" 40#include "runtime/arguments.hpp" 41#include "runtime/extendedPC.hpp" 42#include "runtime/frame.inline.hpp" 43#include "runtime/interfaceSupport.hpp" 44#include "runtime/java.hpp" 45#include "runtime/javaCalls.hpp" 46#include "runtime/mutexLocker.hpp" 47#include "runtime/osThread.hpp" 48#include "runtime/sharedRuntime.hpp" 49#include "runtime/stubRoutines.hpp" 50#include "runtime/thread.inline.hpp" 51#include "runtime/timer.hpp" 52#include "services/memTracker.hpp" 53#include "utilities/events.hpp" 54#include "utilities/vmError.hpp" 55 56// put OS-includes here 57# include <sys/types.h> 58# include <sys/mman.h> 59# include <pthread.h> 60# include <signal.h> 61# include <errno.h> 62# include <dlfcn.h> 63# include <stdlib.h> 64# include <stdio.h> 65# include <unistd.h> 66# include <sys/resource.h> 67# include <pthread.h> 68# include <sys/stat.h> 69# include <sys/time.h> 70# include <sys/utsname.h> 71# include <sys/socket.h> 72# include <sys/wait.h> 73# include <pwd.h> 74# include <poll.h> 75# include <ucontext.h> 76# include <fpu_control.h> 77 78#ifdef AMD64 79#define REG_SP REG_RSP 80#define REG_PC REG_RIP 81#define REG_FP REG_RBP 82#define SPELL_REG_SP "rsp" 83#define SPELL_REG_FP "rbp" 84#else 85#define REG_SP REG_UESP 86#define REG_PC REG_EIP 87#define REG_FP REG_EBP 88#define SPELL_REG_SP "esp" 89#define SPELL_REG_FP "ebp" 90#endif // AMD64 91 92PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 93 94address os::current_stack_pointer() { 95#ifdef SPARC_WORKS 96 register void *esp; 97 __asm__("mov %%"SPELL_REG_SP", %0":"=r"(esp)); 98 return (address) ((char*)esp + sizeof(long)*2); 99#elif defined(__clang__) 100 intptr_t* esp; 101 __asm__ __volatile__ ("mov %%"SPELL_REG_SP", %0":"=r"(esp):); 102 return (address) esp; 103#else 104 register void *esp __asm__ (SPELL_REG_SP); 105 return (address) esp; 106#endif 107} 108 109char* os::non_memory_address_word() { 110 // Must never look like an address returned by reserve_memory, 111 // even in its subfields (as defined by the CPU immediate fields, 112 // if the CPU splits constants across multiple instructions). 113 114 return (char*) -1; 115} 116 117void os::initialize_thread(Thread* thr) { 118// Nothing to do. 119} 120 121address os::Linux::ucontext_get_pc(ucontext_t * uc) { 122 return (address)uc->uc_mcontext.gregs[REG_PC]; 123} 124 125intptr_t* os::Linux::ucontext_get_sp(ucontext_t * uc) { 126 return (intptr_t*)uc->uc_mcontext.gregs[REG_SP]; 127} 128 129intptr_t* os::Linux::ucontext_get_fp(ucontext_t * uc) { 130 return (intptr_t*)uc->uc_mcontext.gregs[REG_FP]; 131} 132 133// For Forte Analyzer AsyncGetCallTrace profiling support - thread 134// is currently interrupted by SIGPROF. 135// os::Solaris::fetch_frame_from_ucontext() tries to skip nested signal 136// frames. Currently we don't do that on Linux, so it's the same as 137// os::fetch_frame_from_context(). 138ExtendedPC os::Linux::fetch_frame_from_ucontext(Thread* thread, 139 ucontext_t* uc, intptr_t** ret_sp, intptr_t** ret_fp) { 140 141 assert(thread != NULL, "just checking"); 142 assert(ret_sp != NULL, "just checking"); 143 assert(ret_fp != NULL, "just checking"); 144 145 return os::fetch_frame_from_context(uc, ret_sp, ret_fp); 146} 147 148ExtendedPC os::fetch_frame_from_context(void* ucVoid, 149 intptr_t** ret_sp, intptr_t** ret_fp) { 150 151 ExtendedPC epc; 152 ucontext_t* uc = (ucontext_t*)ucVoid; 153 154 if (uc != NULL) { 155 epc = ExtendedPC(os::Linux::ucontext_get_pc(uc)); 156 if (ret_sp) *ret_sp = os::Linux::ucontext_get_sp(uc); 157 if (ret_fp) *ret_fp = os::Linux::ucontext_get_fp(uc); 158 } else { 159 // construct empty ExtendedPC for return value checking 160 epc = ExtendedPC(NULL); 161 if (ret_sp) *ret_sp = (intptr_t *)NULL; 162 if (ret_fp) *ret_fp = (intptr_t *)NULL; 163 } 164 165 return epc; 166} 167 168frame os::fetch_frame_from_context(void* ucVoid) { 169 intptr_t* sp; 170 intptr_t* fp; 171 ExtendedPC epc = fetch_frame_from_context(ucVoid, &sp, &fp); 172 return frame(sp, fp, epc.pc()); 173} 174 175// By default, gcc always save frame pointer (%ebp/%rbp) on stack. It may get 176// turned off by -fomit-frame-pointer, 177frame os::get_sender_for_C_frame(frame* fr) { 178 return frame(fr->sender_sp(), fr->link(), fr->sender_pc()); 179} 180 181intptr_t* _get_previous_fp() { 182#ifdef SPARC_WORKS 183 register intptr_t **ebp; 184 __asm__("mov %%"SPELL_REG_FP", %0":"=r"(ebp)); 185#elif defined(__clang__) 186 intptr_t **ebp; 187 __asm__ __volatile__ ("mov %%"SPELL_REG_FP", %0":"=r"(ebp):); 188#else 189 register intptr_t **ebp __asm__ (SPELL_REG_FP); 190#endif 191 return (intptr_t*) *ebp; // we want what it points to. 192} 193 194 195frame os::current_frame() { 196 intptr_t* fp = _get_previous_fp(); 197 frame myframe((intptr_t*)os::current_stack_pointer(), 198 (intptr_t*)fp, 199 CAST_FROM_FN_PTR(address, os::current_frame)); 200 if (os::is_first_C_frame(&myframe)) { 201 // stack is not walkable 202 return frame(); 203 } else { 204 return os::get_sender_for_C_frame(&myframe); 205 } 206} 207 208// Utility functions 209 210// From IA32 System Programming Guide 211enum { 212 trap_page_fault = 0xE 213}; 214 215extern "C" JNIEXPORT int 216JVM_handle_linux_signal(int sig, 217 siginfo_t* info, 218 void* ucVoid, 219 int abort_if_unrecognized) { 220 ucontext_t* uc = (ucontext_t*) ucVoid; 221 222 Thread* t = ThreadLocalStorage::get_thread_slow(); 223 224 // Must do this before SignalHandlerMark, if crash protection installed we will longjmp away 225 // (no destructors can be run) 226 os::WatcherThreadCrashProtection::check_crash_protection(sig, t); 227 228 SignalHandlerMark shm(t); 229 230 // Note: it's not uncommon that JNI code uses signal/sigset to install 231 // then restore certain signal handler (e.g. to temporarily block SIGPIPE, 232 // or have a SIGILL handler when detecting CPU type). When that happens, 233 // JVM_handle_linux_signal() might be invoked with junk info/ucVoid. To 234 // avoid unnecessary crash when libjsig is not preloaded, try handle signals 235 // that do not require siginfo/ucontext first. 236 237 if (sig == SIGPIPE || sig == SIGXFSZ) { 238 // allow chained handler to go first 239 if (os::Linux::chained_handler(sig, info, ucVoid)) { 240 return true; 241 } else { 242 if (PrintMiscellaneous && (WizardMode || Verbose)) { 243 char buf[64]; 244 warning("Ignoring %s - see bugs 4229104 or 646499219", 245 os::exception_name(sig, buf, sizeof(buf))); 246 } 247 return true; 248 } 249 } 250 251 JavaThread* thread = NULL; 252 VMThread* vmthread = NULL; 253 if (os::Linux::signal_handlers_are_installed) { 254 if (t != NULL ){ 255 if(t->is_Java_thread()) { 256 thread = (JavaThread*)t; 257 } 258 else if(t->is_VM_thread()){ 259 vmthread = (VMThread *)t; 260 } 261 } 262 } 263/* 264 NOTE: does not seem to work on linux. 265 if (info == NULL || info->si_code <= 0 || info->si_code == SI_NOINFO) { 266 // can't decode this kind of signal 267 info = NULL; 268 } else { 269 assert(sig == info->si_signo, "bad siginfo"); 270 } 271*/ 272 // decide if this trap can be handled by a stub 273 address stub = NULL; 274 275 address pc = NULL; 276 277 //%note os_trap_1 278 if (info != NULL && uc != NULL && thread != NULL) { 279 pc = (address) os::Linux::ucontext_get_pc(uc); 280 281 if (StubRoutines::is_safefetch_fault(pc)) { 282 uc->uc_mcontext.gregs[REG_PC] = intptr_t(StubRoutines::continuation_for_safefetch_fault(pc)); 283 return 1; 284 } 285 286#ifndef AMD64 287 // Halt if SI_KERNEL before more crashes get misdiagnosed as Java bugs 288 // This can happen in any running code (currently more frequently in 289 // interpreter code but has been seen in compiled code) 290 if (sig == SIGSEGV && info->si_addr == 0 && info->si_code == SI_KERNEL) { 291 fatal("An irrecoverable SI_KERNEL SIGSEGV has occurred due " 292 "to unstable signal handling in this distribution."); 293 } 294#endif // AMD64 295 296 // Handle ALL stack overflow variations here 297 if (sig == SIGSEGV) { 298 address addr = (address) info->si_addr; 299 300 // check if fault address is within thread stack 301 if (addr < thread->stack_base() && 302 addr >= thread->stack_base() - thread->stack_size()) { 303 // stack overflow 304 if (thread->in_stack_yellow_zone(addr)) { 305 thread->disable_stack_yellow_zone(); 306 if (thread->thread_state() == _thread_in_Java) { 307 // Throw a stack overflow exception. Guard pages will be reenabled 308 // while unwinding the stack. 309 stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::STACK_OVERFLOW); 310 } else { 311 // Thread was in the vm or native code. Return and try to finish. 312 return 1; 313 } 314 } else if (thread->in_stack_red_zone(addr)) { 315 // Fatal red zone violation. Disable the guard pages and fall through 316 // to handle_unexpected_exception way down below. 317 thread->disable_stack_red_zone(); 318 tty->print_raw_cr("An irrecoverable stack overflow has occurred."); 319 320 // This is a likely cause, but hard to verify. Let's just print 321 // it as a hint. 322 tty->print_raw_cr("Please check if any of your loaded .so files has " 323 "enabled executable stack (see man page execstack(8))"); 324 } else { 325 // Accessing stack address below sp may cause SEGV if current 326 // thread has MAP_GROWSDOWN stack. This should only happen when 327 // current thread was created by user code with MAP_GROWSDOWN flag 328 // and then attached to VM. See notes in os_linux.cpp. 329 if (thread->osthread()->expanding_stack() == 0) { 330 thread->osthread()->set_expanding_stack(); 331 if (os::Linux::manually_expand_stack(thread, addr)) { 332 thread->osthread()->clear_expanding_stack(); 333 return 1; 334 } 335 thread->osthread()->clear_expanding_stack(); 336 } else { 337 fatal("recursive segv. expanding stack."); 338 } 339 } 340 } 341 } 342 343 if ((sig == SIGSEGV) && VM_Version::is_cpuinfo_segv_addr(pc)) { 344 // Verify that OS save/restore AVX registers. 345 stub = VM_Version::cpuinfo_cont_addr(); 346 } 347 348 if (thread->thread_state() == _thread_in_Java) { 349 // Java thread running in Java code => find exception handler if any 350 // a fault inside compiled code, the interpreter, or a stub 351 352 if (sig == SIGSEGV && os::is_poll_address((address)info->si_addr)) { 353 stub = SharedRuntime::get_poll_stub(pc); 354 } else if (sig == SIGBUS /* && info->si_code == BUS_OBJERR */) { 355 // BugId 4454115: A read from a MappedByteBuffer can fault 356 // here if the underlying file has been truncated. 357 // Do not crash the VM in such a case. 358 CodeBlob* cb = CodeCache::find_blob_unsafe(pc); 359 nmethod* nm = (cb != NULL && cb->is_nmethod()) ? (nmethod*)cb : NULL; 360 if (nm != NULL && nm->has_unsafe_access()) { 361 stub = StubRoutines::handler_for_unsafe_access(); 362 } 363 } 364 else 365 366#ifdef AMD64 367 if (sig == SIGFPE && 368 (info->si_code == FPE_INTDIV || info->si_code == FPE_FLTDIV)) { 369 stub = 370 SharedRuntime:: 371 continuation_for_implicit_exception(thread, 372 pc, 373 SharedRuntime:: 374 IMPLICIT_DIVIDE_BY_ZERO); 375#else 376 if (sig == SIGFPE /* && info->si_code == FPE_INTDIV */) { 377 // HACK: si_code does not work on linux 2.2.12-20!!! 378 int op = pc[0]; 379 if (op == 0xDB) { 380 // FIST 381 // TODO: The encoding of D2I in i486.ad can cause an exception 382 // prior to the fist instruction if there was an invalid operation 383 // pending. We want to dismiss that exception. From the win_32 384 // side it also seems that if it really was the fist causing 385 // the exception that we do the d2i by hand with different 386 // rounding. Seems kind of weird. 387 // NOTE: that we take the exception at the NEXT floating point instruction. 388 assert(pc[0] == 0xDB, "not a FIST opcode"); 389 assert(pc[1] == 0x14, "not a FIST opcode"); 390 assert(pc[2] == 0x24, "not a FIST opcode"); 391 return true; 392 } else if (op == 0xF7) { 393 // IDIV 394 stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_DIVIDE_BY_ZERO); 395 } else { 396 // TODO: handle more cases if we are using other x86 instructions 397 // that can generate SIGFPE signal on linux. 398 tty->print_cr("unknown opcode 0x%X with SIGFPE.", op); 399 fatal("please update this code."); 400 } 401#endif // AMD64 402 } else if (sig == SIGSEGV && 403 !MacroAssembler::needs_explicit_null_check((intptr_t)info->si_addr)) { 404 // Determination of interpreter/vtable stub/compiled code null exception 405 stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_NULL); 406 } 407 } else if (thread->thread_state() == _thread_in_vm && 408 sig == SIGBUS && /* info->si_code == BUS_OBJERR && */ 409 thread->doing_unsafe_access()) { 410 stub = StubRoutines::handler_for_unsafe_access(); 411 } 412 413 // jni_fast_Get<Primitive>Field can trap at certain pc's if a GC kicks in 414 // and the heap gets shrunk before the field access. 415 if ((sig == SIGSEGV) || (sig == SIGBUS)) { 416 address addr = JNI_FastGetField::find_slowcase_pc(pc); 417 if (addr != (address)-1) { 418 stub = addr; 419 } 420 } 421 422 // Check to see if we caught the safepoint code in the 423 // process of write protecting the memory serialization page. 424 // It write enables the page immediately after protecting it 425 // so we can just return to retry the write. 426 if ((sig == SIGSEGV) && 427 os::is_memory_serialize_page(thread, (address) info->si_addr)) { 428 // Block current thread until the memory serialize page permission restored. 429 os::block_on_serialize_page_trap(); 430 return true; 431 } 432 } 433 434#ifndef AMD64 435 // Execution protection violation 436 // 437 // This should be kept as the last step in the triage. We don't 438 // have a dedicated trap number for a no-execute fault, so be 439 // conservative and allow other handlers the first shot. 440 // 441 // Note: We don't test that info->si_code == SEGV_ACCERR here. 442 // this si_code is so generic that it is almost meaningless; and 443 // the si_code for this condition may change in the future. 444 // Furthermore, a false-positive should be harmless. 445 if (UnguardOnExecutionViolation > 0 && 446 (sig == SIGSEGV || sig == SIGBUS) && 447 uc->uc_mcontext.gregs[REG_TRAPNO] == trap_page_fault) { 448 int page_size = os::vm_page_size(); 449 address addr = (address) info->si_addr; 450 address pc = os::Linux::ucontext_get_pc(uc); 451 // Make sure the pc and the faulting address are sane. 452 // 453 // If an instruction spans a page boundary, and the page containing 454 // the beginning of the instruction is executable but the following 455 // page is not, the pc and the faulting address might be slightly 456 // different - we still want to unguard the 2nd page in this case. 457 // 458 // 15 bytes seems to be a (very) safe value for max instruction size. 459 bool pc_is_near_addr = 460 (pointer_delta((void*) addr, (void*) pc, sizeof(char)) < 15); 461 bool instr_spans_page_boundary = 462 (align_size_down((intptr_t) pc ^ (intptr_t) addr, 463 (intptr_t) page_size) > 0); 464 465 if (pc == addr || (pc_is_near_addr && instr_spans_page_boundary)) { 466 static volatile address last_addr = 467 (address) os::non_memory_address_word(); 468 469 // In conservative mode, don't unguard unless the address is in the VM 470 if (addr != last_addr && 471 (UnguardOnExecutionViolation > 1 || os::address_is_in_vm(addr))) { 472 473 // Set memory to RWX and retry 474 address page_start = 475 (address) align_size_down((intptr_t) addr, (intptr_t) page_size); 476 bool res = os::protect_memory((char*) page_start, page_size, 477 os::MEM_PROT_RWX); 478 479 if (PrintMiscellaneous && Verbose) { 480 char buf[256]; 481 jio_snprintf(buf, sizeof(buf), "Execution protection violation " 482 "at " INTPTR_FORMAT 483 ", unguarding " INTPTR_FORMAT ": %s, errno=%d", addr, 484 page_start, (res ? "success" : "failed"), errno); 485 tty->print_raw_cr(buf); 486 } 487 stub = pc; 488 489 // Set last_addr so if we fault again at the same address, we don't end 490 // up in an endless loop. 491 // 492 // There are two potential complications here. Two threads trapping at 493 // the same address at the same time could cause one of the threads to 494 // think it already unguarded, and abort the VM. Likely very rare. 495 // 496 // The other race involves two threads alternately trapping at 497 // different addresses and failing to unguard the page, resulting in 498 // an endless loop. This condition is probably even more unlikely than 499 // the first. 500 // 501 // Although both cases could be avoided by using locks or thread local 502 // last_addr, these solutions are unnecessary complication: this 503 // handler is a best-effort safety net, not a complete solution. It is 504 // disabled by default and should only be used as a workaround in case 505 // we missed any no-execute-unsafe VM code. 506 507 last_addr = addr; 508 } 509 } 510 } 511#endif // !AMD64 512 513 if (stub != NULL) { 514 // save all thread context in case we need to restore it 515 if (thread != NULL) thread->set_saved_exception_pc(pc); 516 517 uc->uc_mcontext.gregs[REG_PC] = (greg_t)stub; 518 return true; 519 } 520 521 // signal-chaining 522 if (os::Linux::chained_handler(sig, info, ucVoid)) { 523 return true; 524 } 525 526 if (!abort_if_unrecognized) { 527 // caller wants another chance, so give it to him 528 return false; 529 } 530 531 if (pc == NULL && uc != NULL) { 532 pc = os::Linux::ucontext_get_pc(uc); 533 } 534 535 // unmask current signal 536 sigset_t newset; 537 sigemptyset(&newset); 538 sigaddset(&newset, sig); 539 sigprocmask(SIG_UNBLOCK, &newset, NULL); 540 541 VMError err(t, sig, pc, info, ucVoid); 542 err.report_and_die(); 543 544 ShouldNotReachHere(); 545 return true; // Mute compiler 546} 547 548void os::Linux::init_thread_fpu_state(void) { 549#ifndef AMD64 550 // set fpu to 53 bit precision 551 set_fpu_control_word(0x27f); 552#endif // !AMD64 553} 554 555int os::Linux::get_fpu_control_word(void) { 556#ifdef AMD64 557 return 0; 558#else 559 int fpu_control; 560 _FPU_GETCW(fpu_control); 561 return fpu_control & 0xffff; 562#endif // AMD64 563} 564 565void os::Linux::set_fpu_control_word(int fpu_control) { 566#ifndef AMD64 567 _FPU_SETCW(fpu_control); 568#endif // !AMD64 569} 570 571// Check that the linux kernel version is 2.4 or higher since earlier 572// versions do not support SSE without patches. 573bool os::supports_sse() { 574#ifdef AMD64 575 return true; 576#else 577 struct utsname uts; 578 if( uname(&uts) != 0 ) return false; // uname fails? 579 char *minor_string; 580 int major = strtol(uts.release,&minor_string,10); 581 int minor = strtol(minor_string+1,NULL,10); 582 bool result = (major > 2 || (major==2 && minor >= 4)); 583#ifndef PRODUCT 584 if (PrintMiscellaneous && Verbose) { 585 tty->print("OS version is %d.%d, which %s support SSE/SSE2\n", 586 major,minor, result ? "DOES" : "does NOT"); 587 } 588#endif 589 return result; 590#endif // AMD64 591} 592 593bool os::is_allocatable(size_t bytes) { 594#ifdef AMD64 595 // unused on amd64? 596 return true; 597#else 598 599 if (bytes < 2 * G) { 600 return true; 601 } 602 603 char* addr = reserve_memory(bytes, NULL); 604 605 if (addr != NULL) { 606 release_memory(addr, bytes); 607 } 608 609 return addr != NULL; 610#endif // AMD64 611} 612 613//////////////////////////////////////////////////////////////////////////////// 614// thread stack 615 616#ifdef AMD64 617size_t os::Linux::min_stack_allowed = 64 * K; 618 619// amd64: pthread on amd64 is always in floating stack mode 620bool os::Linux::supports_variable_stack_size() { return true; } 621#else 622size_t os::Linux::min_stack_allowed = (48 DEBUG_ONLY(+4))*K; 623 624#ifdef __GNUC__ 625#define GET_GS() ({int gs; __asm__ volatile("movw %%gs, %w0":"=q"(gs)); gs&0xffff;}) 626#endif 627 628// Test if pthread library can support variable thread stack size. LinuxThreads 629// in fixed stack mode allocates 2M fixed slot for each thread. LinuxThreads 630// in floating stack mode and NPTL support variable stack size. 631bool os::Linux::supports_variable_stack_size() { 632 if (os::Linux::is_NPTL()) { 633 // NPTL, yes 634 return true; 635 636 } else { 637 // Note: We can't control default stack size when creating a thread. 638 // If we use non-default stack size (pthread_attr_setstacksize), both 639 // floating stack and non-floating stack LinuxThreads will return the 640 // same value. This makes it impossible to implement this function by 641 // detecting thread stack size directly. 642 // 643 // An alternative approach is to check %gs. Fixed-stack LinuxThreads 644 // do not use %gs, so its value is 0. Floating-stack LinuxThreads use 645 // %gs (either as LDT selector or GDT selector, depending on kernel) 646 // to access thread specific data. 647 // 648 // Note that %gs is a reserved glibc register since early 2001, so 649 // applications are not allowed to change its value (Ulrich Drepper from 650 // Redhat confirmed that all known offenders have been modified to use 651 // either %fs or TSD). In the worst case scenario, when VM is embedded in 652 // a native application that plays with %gs, we might see non-zero %gs 653 // even LinuxThreads is running in fixed stack mode. As the result, we'll 654 // return true and skip _thread_safety_check(), so we may not be able to 655 // detect stack-heap collisions. But otherwise it's harmless. 656 // 657#ifdef __GNUC__ 658 return (GET_GS() != 0); 659#else 660 return false; 661#endif 662 } 663} 664#endif // AMD64 665 666// return default stack size for thr_type 667size_t os::Linux::default_stack_size(os::ThreadType thr_type) { 668 // default stack size (compiler thread needs larger stack) 669#ifdef AMD64 670 size_t s = (thr_type == os::compiler_thread ? 4 * M : 1 * M); 671#else 672 size_t s = (thr_type == os::compiler_thread ? 2 * M : 512 * K); 673#endif // AMD64 674 return s; 675} 676 677size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 678 // Creating guard page is very expensive. Java thread has HotSpot 679 // guard page, only enable glibc guard page for non-Java threads. 680 return (thr_type == java_thread ? 0 : page_size()); 681} 682 683// Java thread: 684// 685// Low memory addresses 686// +------------------------+ 687// | |\ JavaThread created by VM does not have glibc 688// | glibc guard page | - guard, attached Java thread usually has 689// | |/ 1 page glibc guard. 690// P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 691// | |\ 692// | HotSpot Guard Pages | - red and yellow pages 693// | |/ 694// +------------------------+ JavaThread::stack_yellow_zone_base() 695// | |\ 696// | Normal Stack | - 697// | |/ 698// P2 +------------------------+ Thread::stack_base() 699// 700// Non-Java thread: 701// 702// Low memory addresses 703// +------------------------+ 704// | |\ 705// | glibc guard page | - usually 1 page 706// | |/ 707// P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 708// | |\ 709// | Normal Stack | - 710// | |/ 711// P2 +------------------------+ Thread::stack_base() 712// 713// ** P1 (aka bottom) and size ( P2 = P1 - size) are the address and stack size returned from 714// pthread_attr_getstack() 715 716static void current_stack_region(address * bottom, size_t * size) { 717 if (os::Linux::is_initial_thread()) { 718 // initial thread needs special handling because pthread_getattr_np() 719 // may return bogus value. 720 *bottom = os::Linux::initial_thread_stack_bottom(); 721 *size = os::Linux::initial_thread_stack_size(); 722 } else { 723 pthread_attr_t attr; 724 725 int rslt = pthread_getattr_np(pthread_self(), &attr); 726 727 // JVM needs to know exact stack location, abort if it fails 728 if (rslt != 0) { 729 if (rslt == ENOMEM) { 730 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 731 } else { 732 fatal(err_msg("pthread_getattr_np failed with errno = %d", rslt)); 733 } 734 } 735 736 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 737 fatal("Can not locate current stack attributes!"); 738 } 739 740 pthread_attr_destroy(&attr); 741 742 } 743 assert(os::current_stack_pointer() >= *bottom && 744 os::current_stack_pointer() < *bottom + *size, "just checking"); 745} 746 747address os::current_stack_base() { 748 address bottom; 749 size_t size; 750 current_stack_region(&bottom, &size); 751 return (bottom + size); 752} 753 754size_t os::current_stack_size() { 755 // stack size includes normal stack and HotSpot guard pages 756 address bottom; 757 size_t size; 758 current_stack_region(&bottom, &size); 759 return size; 760} 761 762///////////////////////////////////////////////////////////////////////////// 763// helper functions for fatal error handler 764 765void os::print_context(outputStream *st, void *context) { 766 if (context == NULL) return; 767 768 ucontext_t *uc = (ucontext_t*)context; 769 st->print_cr("Registers:"); 770#ifdef AMD64 771 st->print( "RAX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RAX]); 772 st->print(", RBX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RBX]); 773 st->print(", RCX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RCX]); 774 st->print(", RDX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDX]); 775 st->cr(); 776 st->print( "RSP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSP]); 777 st->print(", RBP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RBP]); 778 st->print(", RSI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSI]); 779 st->print(", RDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDI]); 780 st->cr(); 781 st->print( "R8 =" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R8]); 782 st->print(", R9 =" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R9]); 783 st->print(", R10=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R10]); 784 st->print(", R11=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R11]); 785 st->cr(); 786 st->print( "R12=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R12]); 787 st->print(", R13=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R13]); 788 st->print(", R14=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R14]); 789 st->print(", R15=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R15]); 790 st->cr(); 791 st->print( "RIP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RIP]); 792 st->print(", EFLAGS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EFL]); 793 st->print(", CSGSFS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_CSGSFS]); 794 st->print(", ERR=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ERR]); 795 st->cr(); 796 st->print(" TRAPNO=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_TRAPNO]); 797#else 798 st->print( "EAX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EAX]); 799 st->print(", EBX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EBX]); 800 st->print(", ECX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ECX]); 801 st->print(", EDX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EDX]); 802 st->cr(); 803 st->print( "ESP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_UESP]); 804 st->print(", EBP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EBP]); 805 st->print(", ESI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ESI]); 806 st->print(", EDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EDI]); 807 st->cr(); 808 st->print( "EIP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EIP]); 809 st->print(", EFLAGS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EFL]); 810 st->print(", CR2=" INTPTR_FORMAT, uc->uc_mcontext.cr2); 811#endif // AMD64 812 st->cr(); 813 st->cr(); 814 815 intptr_t *sp = (intptr_t *)os::Linux::ucontext_get_sp(uc); 816 st->print_cr("Top of Stack: (sp=" PTR_FORMAT ")", sp); 817 print_hex_dump(st, (address)sp, (address)(sp + 8*sizeof(intptr_t)), sizeof(intptr_t)); 818 st->cr(); 819 820 // Note: it may be unsafe to inspect memory near pc. For example, pc may 821 // point to garbage if entry point in an nmethod is corrupted. Leave 822 // this at the end, and hope for the best. 823 address pc = os::Linux::ucontext_get_pc(uc); 824 st->print_cr("Instructions: (pc=" PTR_FORMAT ")", pc); 825 print_hex_dump(st, pc - 32, pc + 32, sizeof(char)); 826} 827 828void os::print_register_info(outputStream *st, void *context) { 829 if (context == NULL) return; 830 831 ucontext_t *uc = (ucontext_t*)context; 832 833 st->print_cr("Register to memory mapping:"); 834 st->cr(); 835 836 // this is horrendously verbose but the layout of the registers in the 837 // context does not match how we defined our abstract Register set, so 838 // we can't just iterate through the gregs area 839 840 // this is only for the "general purpose" registers 841 842#ifdef AMD64 843 st->print("RAX="); print_location(st, uc->uc_mcontext.gregs[REG_RAX]); 844 st->print("RBX="); print_location(st, uc->uc_mcontext.gregs[REG_RBX]); 845 st->print("RCX="); print_location(st, uc->uc_mcontext.gregs[REG_RCX]); 846 st->print("RDX="); print_location(st, uc->uc_mcontext.gregs[REG_RDX]); 847 st->print("RSP="); print_location(st, uc->uc_mcontext.gregs[REG_RSP]); 848 st->print("RBP="); print_location(st, uc->uc_mcontext.gregs[REG_RBP]); 849 st->print("RSI="); print_location(st, uc->uc_mcontext.gregs[REG_RSI]); 850 st->print("RDI="); print_location(st, uc->uc_mcontext.gregs[REG_RDI]); 851 st->print("R8 ="); print_location(st, uc->uc_mcontext.gregs[REG_R8]); 852 st->print("R9 ="); print_location(st, uc->uc_mcontext.gregs[REG_R9]); 853 st->print("R10="); print_location(st, uc->uc_mcontext.gregs[REG_R10]); 854 st->print("R11="); print_location(st, uc->uc_mcontext.gregs[REG_R11]); 855 st->print("R12="); print_location(st, uc->uc_mcontext.gregs[REG_R12]); 856 st->print("R13="); print_location(st, uc->uc_mcontext.gregs[REG_R13]); 857 st->print("R14="); print_location(st, uc->uc_mcontext.gregs[REG_R14]); 858 st->print("R15="); print_location(st, uc->uc_mcontext.gregs[REG_R15]); 859#else 860 st->print("EAX="); print_location(st, uc->uc_mcontext.gregs[REG_EAX]); 861 st->print("EBX="); print_location(st, uc->uc_mcontext.gregs[REG_EBX]); 862 st->print("ECX="); print_location(st, uc->uc_mcontext.gregs[REG_ECX]); 863 st->print("EDX="); print_location(st, uc->uc_mcontext.gregs[REG_EDX]); 864 st->print("ESP="); print_location(st, uc->uc_mcontext.gregs[REG_ESP]); 865 st->print("EBP="); print_location(st, uc->uc_mcontext.gregs[REG_EBP]); 866 st->print("ESI="); print_location(st, uc->uc_mcontext.gregs[REG_ESI]); 867 st->print("EDI="); print_location(st, uc->uc_mcontext.gregs[REG_EDI]); 868#endif // AMD64 869 870 st->cr(); 871} 872 873void os::setup_fpu() { 874#ifndef AMD64 875 address fpu_cntrl = StubRoutines::addr_fpu_cntrl_wrd_std(); 876 __asm__ volatile ( "fldcw (%0)" : 877 : "r" (fpu_cntrl) : "memory"); 878#endif // !AMD64 879} 880 881#ifndef PRODUCT 882void os::verify_stack_alignment() { 883#ifdef AMD64 884 assert(((intptr_t)os::current_stack_pointer() & (StackAlignmentInBytes-1)) == 0, "incorrect stack alignment"); 885#endif 886} 887#endif 888 889 890/* 891 * IA32 only: execute code at a high address in case buggy NX emulation is present. I.e. avoid CS limit 892 * updates (JDK-8023956). 893 */ 894void os::workaround_expand_exec_shield_cs_limit() { 895#if defined(IA32) 896 size_t page_size = os::vm_page_size(); 897 /* 898 * Take the highest VA the OS will give us and exec 899 * 900 * Although using -(pagesz) as mmap hint works on newer kernel as you would 901 * think, older variants affected by this work-around don't (search forward only). 902 * 903 * On the affected distributions, we understand the memory layout to be: 904 * 905 * TASK_LIMIT= 3G, main stack base close to TASK_LIMT. 906 * 907 * A few pages south main stack will do it. 908 * 909 * If we are embedded in an app other than launcher (initial != main stack), 910 * we don't have much control or understanding of the address space, just let it slide. 911 */ 912 char* hint = (char*) (Linux::initial_thread_stack_bottom() - 913 ((StackYellowPages + StackRedPages + 1) * page_size)); 914 char* codebuf = os::attempt_reserve_memory_at(page_size, hint); 915 if ( (codebuf == NULL) || (!os::commit_memory(codebuf, page_size, true)) ) { 916 return; // No matter, we tried, best effort. 917 } 918 919 MemTracker::record_virtual_memory_type((address)codebuf, mtInternal); 920 921 if (PrintMiscellaneous && (Verbose || WizardMode)) { 922 tty->print_cr("[CS limit NX emulation work-around, exec code at: %p]", codebuf); 923 } 924 925 // Some code to exec: the 'ret' instruction 926 codebuf[0] = 0xC3; 927 928 // Call the code in the codebuf 929 __asm__ volatile("call *%0" : : "r"(codebuf)); 930 931 // keep the page mapped so CS limit isn't reduced. 932#endif 933} 934 935int os::extra_bang_size_in_bytes() { 936 // JDK-8050147 requires the full cache line bang for x86. 937 return VM_Version::L1_line_size(); 938} 939