hotspot/src/cpu/x86/vm/frame_x86.cpp
author sgoldman
Tue, 08 Apr 2008 12:23:15 -0400
changeset 354 3b42d6fdcb82
parent 1 489c9b5090e2
child 670 ddf3e9583f2f
permissions -rw-r--r--
6603919: Stackwalking crash on x86 -server with Sun Studio's collect -j on Summary: Rewrite frame::safe_for_sender and friends to be safe for collector/analyzer Reviewed-by: dcubed, kvn

/*
 * Copyright 1997-2007 Sun Microsystems, Inc.  All Rights Reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 *
 */

# include "incls/_precompiled.incl"
# include "incls/_frame_x86.cpp.incl"

#ifdef ASSERT
void RegisterMap::check_location_valid() {
}
#endif


// Profiling/safepoint support

bool frame::safe_for_sender(JavaThread *thread) {
  address   sp = (address)_sp;
  address   fp = (address)_fp;
  address   unextended_sp = (address)_unextended_sp;
  // sp must be within the stack
  bool sp_safe = (sp <= thread->stack_base()) &&
                 (sp >= thread->stack_base() - thread->stack_size());

  if (!sp_safe) {
    return false;
  }

  // unextended sp must be within the stack and above or equal sp
  bool unextended_sp_safe = (unextended_sp <= thread->stack_base()) &&
                            (unextended_sp >= sp);

  if (!unextended_sp_safe) {
    return false;
  }

  // an fp must be within the stack and above (but not equal) sp
  bool fp_safe = (fp <= thread->stack_base()) && (fp > sp);

  // We know sp/unextended_sp are safe only fp is questionable here

  // If the current frame is known to the code cache then we can attempt to
  // to construct the sender and do some validation of it. This goes a long way
  // toward eliminating issues when we get in frame construction code

  if (_cb != NULL ) {

    // First check if frame is complete and tester is reliable
    // Unfortunately we can only check frame complete for runtime stubs and nmethod
    // other generic buffer blobs are more problematic so we just assume they are
    // ok. adapter blobs never have a frame complete and are never ok.

    if (!_cb->is_frame_complete_at(_pc)) {
      if (_cb->is_nmethod() || _cb->is_adapter_blob() || _cb->is_runtime_stub()) {
        return false;
      }
    }
    // Entry frame checks
    if (is_entry_frame()) {
      // an entry frame must have a valid fp.

      if (!fp_safe) return false;

      // Validate the JavaCallWrapper an entry frame must have

      address jcw = (address)entry_frame_call_wrapper();

      bool jcw_safe = (jcw <= thread->stack_base()) && ( jcw > fp);

      return jcw_safe;

    }

    intptr_t* sender_sp = NULL;
    address   sender_pc = NULL;

    if (is_interpreted_frame()) {
      // fp must be safe
      if (!fp_safe) {
        return false;
      }

      sender_pc = (address) this->fp()[return_addr_offset];
      sender_sp = (intptr_t*) addr_at(sender_sp_offset);

    } else {
      // must be some sort of compiled/runtime frame
      // fp does not have to be safe (although it could be check for c1?)

      sender_sp = _unextended_sp + _cb->frame_size();
      // On Intel the return_address is always the word on the stack
      sender_pc = (address) *(sender_sp-1);
    }

    // We must always be able to find a recognizable pc
    CodeBlob* sender_blob = CodeCache::find_blob_unsafe(sender_pc);
    if (sender_pc == NULL ||  sender_blob == NULL) {
      return false;
    }


    // If the potential sender is the interpreter then we can do some more checking
    if (Interpreter::contains(sender_pc)) {

      // ebp is always saved in a recognizable place in any code we generate. However
      // only if the sender is interpreted/call_stub (c1 too?) are we certain that the saved ebp
      // is really a frame pointer.

      intptr_t *saved_fp = (intptr_t*)*(sender_sp - frame::sender_sp_offset);
      bool saved_fp_safe = ((address)saved_fp <= thread->stack_base()) && (saved_fp > sender_sp);

      if (!saved_fp_safe) {
        return false;
      }

      // construct the potential sender

      frame sender(sender_sp, saved_fp, sender_pc);

      return sender.is_interpreted_frame_valid(thread);

    }

    // Could just be some random pointer within the codeBlob

    if (!sender_blob->instructions_contains(sender_pc)) return false;

    // We should never be able to see an adapter if the current frame is something from code cache

    if ( sender_blob->is_adapter_blob()) {
      return false;
    }

    // Could be the call_stub

    if (StubRoutines::returns_to_call_stub(sender_pc)) {
      intptr_t *saved_fp = (intptr_t*)*(sender_sp - frame::sender_sp_offset);
      bool saved_fp_safe = ((address)saved_fp <= thread->stack_base()) && (saved_fp > sender_sp);

      if (!saved_fp_safe) {
        return false;
      }

      // construct the potential sender

      frame sender(sender_sp, saved_fp, sender_pc);

      // Validate the JavaCallWrapper an entry frame must have
      address jcw = (address)sender.entry_frame_call_wrapper();

      bool jcw_safe = (jcw <= thread->stack_base()) && ( jcw > (address)sender.fp());

      return jcw_safe;
    }

    // If the frame size is 0 something is bad because every nmethod has a non-zero frame size
    // because the return address counts against the callee's frame.

    if (sender_blob->frame_size() == 0) {
      assert(!sender_blob->is_nmethod(), "should count return address at least");
      return false;
    }

    // We should never be able to see anything here except an nmethod. If something in the
    // code cache (current frame) is called by an entity within the code cache that entity
    // should not be anything but the call stub (already covered), the interpreter (already covered)
    // or an nmethod.

    assert(sender_blob->is_nmethod(), "Impossible call chain");

    // Could put some more validation for the potential non-interpreted sender
    // frame we'd create by calling sender if I could think of any. Wait for next crash in forte...

    // One idea is seeing if the sender_pc we have is one that we'd expect to call to current cb

    // We've validated the potential sender that would be created
    return true;
  }

  // Must be native-compiled frame. Since sender will try and use fp to find
  // linkages it must be safe

  if (!fp_safe) {
    return false;
  }

  // Will the pc we fetch be non-zero (which we'll find at the oldest frame)

  if ( (address) this->fp()[return_addr_offset] == NULL) return false;


  // could try and do some more potential verification of native frame if we could think of some...

  return true;

}


void frame::patch_pc(Thread* thread, address pc) {
  if (TracePcPatching) {
    tty->print_cr("patch_pc at address  0x%x [0x%x -> 0x%x] ", &((address *)sp())[-1], ((address *)sp())[-1], pc);
  }
  ((address *)sp())[-1] = pc;
  _cb = CodeCache::find_blob(pc);
  if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) {
    address orig = (((nmethod*)_cb)->get_original_pc(this));
    assert(orig == _pc, "expected original to be stored before patching");
    _deopt_state = is_deoptimized;
    // leave _pc as is
  } else {
    _deopt_state = not_deoptimized;
    _pc = pc;
  }
}

bool frame::is_interpreted_frame() const  {
  return Interpreter::contains(pc());
}

int frame::frame_size() const {
  RegisterMap map(JavaThread::current(), false);
  frame sender = this->sender(&map);
  return sender.sp() - sp();
}

intptr_t* frame::entry_frame_argument_at(int offset) const {
  // convert offset to index to deal with tsi
  int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);
  // Entry frame's arguments are always in relation to unextended_sp()
  return &unextended_sp()[index];
}

// sender_sp
#ifdef CC_INTERP
intptr_t* frame::interpreter_frame_sender_sp() const {
  assert(is_interpreted_frame(), "interpreted frame expected");
  // QQQ why does this specialize method exist if frame::sender_sp() does same thing?
  // seems odd and if we always know interpreted vs. non then sender_sp() is really
  // doing too much work.
  return get_interpreterState()->sender_sp();
}

// monitor elements

BasicObjectLock* frame::interpreter_frame_monitor_begin() const {
  return get_interpreterState()->monitor_base();
}

BasicObjectLock* frame::interpreter_frame_monitor_end() const {
  return (BasicObjectLock*) get_interpreterState()->stack_base();
}

#else // CC_INTERP

intptr_t* frame::interpreter_frame_sender_sp() const {
  assert(is_interpreted_frame(), "interpreted frame expected");
  return (intptr_t*) at(interpreter_frame_sender_sp_offset);
}

void frame::set_interpreter_frame_sender_sp(intptr_t* sender_sp) {
  assert(is_interpreted_frame(), "interpreted frame expected");
  ptr_at_put(interpreter_frame_sender_sp_offset, (intptr_t) sender_sp);
}


// monitor elements

BasicObjectLock* frame::interpreter_frame_monitor_begin() const {
  return (BasicObjectLock*) addr_at(interpreter_frame_monitor_block_bottom_offset);
}

BasicObjectLock* frame::interpreter_frame_monitor_end() const {
  BasicObjectLock* result = (BasicObjectLock*) *addr_at(interpreter_frame_monitor_block_top_offset);
  // make sure the pointer points inside the frame
  assert((intptr_t) fp() >  (intptr_t) result, "result must <  than frame pointer");
  assert((intptr_t) sp() <= (intptr_t) result, "result must >= than stack pointer");
  return result;
}

void frame::interpreter_frame_set_monitor_end(BasicObjectLock* value) {
  *((BasicObjectLock**)addr_at(interpreter_frame_monitor_block_top_offset)) = value;
}

// Used by template based interpreter deoptimization
void frame::interpreter_frame_set_last_sp(intptr_t* sp) {
    *((intptr_t**)addr_at(interpreter_frame_last_sp_offset)) = sp;
}
#endif // CC_INTERP

frame frame::sender_for_entry_frame(RegisterMap* map) const {
  assert(map != NULL, "map must be set");
  // Java frame called from C; skip all C frames and return top C
  // frame of that chunk as the sender
  JavaFrameAnchor* jfa = entry_frame_call_wrapper()->anchor();
  assert(!entry_frame_is_first(), "next Java fp must be non zero");
  assert(jfa->last_Java_sp() > sp(), "must be above this frame on stack");
  map->clear();
  assert(map->include_argument_oops(), "should be set by clear");
  if (jfa->last_Java_pc() != NULL ) {
    frame fr(jfa->last_Java_sp(), jfa->last_Java_fp(), jfa->last_Java_pc());
    return fr;
  }
  frame fr(jfa->last_Java_sp(), jfa->last_Java_fp());
  return fr;
}

frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
  // sp is the raw sp from the sender after adapter or interpreter extension
  intptr_t* sp = (intptr_t*) addr_at(sender_sp_offset);

  // This is the sp before any possible extension (adapter/locals).
  intptr_t* unextended_sp = interpreter_frame_sender_sp();

  // The interpreter and compiler(s) always save EBP/RBP in a known
  // location on entry. We must record where that location is
  // so this if EBP/RBP was live on callout from c2 we can find
  // the saved copy no matter what it called.

  // Since the interpreter always saves EBP/RBP if we record where it is then
  // we don't have to always save EBP/RBP on entry and exit to c2 compiled
  // code, on entry will be enough.
#ifdef COMPILER2
  if (map->update_map()) {
    map->set_location(rbp->as_VMReg(), (address) addr_at(link_offset));
#ifdef AMD64
    // this is weird "H" ought to be at a higher address however the
    // oopMaps seems to have the "H" regs at the same address and the
    // vanilla register.
    // XXXX make this go away
    if (true) {
      map->set_location(rbp->as_VMReg()->next(), (address)addr_at(link_offset));
    }
#endif // AMD64
  }
#endif /* COMPILER2 */
  return frame(sp, unextended_sp, link(), sender_pc());
}


//------------------------------sender_for_compiled_frame-----------------------
frame frame::sender_for_compiled_frame(RegisterMap* map) const {
  assert(map != NULL, "map must be set");
  const bool c1_compiled = _cb->is_compiled_by_c1();

  // frame owned by optimizing compiler
  intptr_t* sender_sp = NULL;

  assert(_cb->frame_size() >= 0, "must have non-zero frame size");
  sender_sp = unextended_sp() + _cb->frame_size();

  // On Intel the return_address is always the word on the stack
  address sender_pc = (address) *(sender_sp-1);

  // This is the saved value of ebp which may or may not really be an fp.
  // it is only an fp if the sender is an interpreter frame (or c1?)

  intptr_t *saved_fp = (intptr_t*)*(sender_sp - frame::sender_sp_offset);

  if (map->update_map()) {
    // Tell GC to use argument oopmaps for some runtime stubs that need it.
    // For C1, the runtime stub might not have oop maps, so set this flag
    // outside of update_register_map.
    map->set_include_argument_oops(_cb->caller_must_gc_arguments(map->thread()));
    if (_cb->oop_maps() != NULL) {
      OopMapSet::update_register_map(this, map);
    }
    // Since the prolog does the save and restore of epb there is no oopmap
    // for it so we must fill in its location as if there was an oopmap entry
    // since if our caller was compiled code there could be live jvm state in it.
    map->set_location(rbp->as_VMReg(), (address) (sender_sp - frame::sender_sp_offset));
#ifdef AMD64
    // this is weird "H" ought to be at a higher address however the
    // oopMaps seems to have the "H" regs at the same address and the
    // vanilla register.
    // XXXX make this go away
    if (true) {
      map->set_location(rbp->as_VMReg()->next(), (address) (sender_sp - frame::sender_sp_offset));
    }
#endif // AMD64
  }

  assert(sender_sp != sp(), "must have changed");
  return frame(sender_sp, saved_fp, sender_pc);
}

frame frame::sender(RegisterMap* map) const {
  // Default is we done have to follow them. The sender_for_xxx will
  // update it accordingly
  map->set_include_argument_oops(false);

  if (is_entry_frame())       return sender_for_entry_frame(map);
  if (is_interpreted_frame()) return sender_for_interpreter_frame(map);
  assert(_cb == CodeCache::find_blob(pc()),"Must be the same");

  if (_cb != NULL) {
    return sender_for_compiled_frame(map);
  }
  // Must be native-compiled frame, i.e. the marshaling code for native
  // methods that exists in the core system.
  return frame(sender_sp(), link(), sender_pc());
}


bool frame::interpreter_frame_equals_unpacked_fp(intptr_t* fp) {
  assert(is_interpreted_frame(), "must be interpreter frame");
  methodOop method = interpreter_frame_method();
  // When unpacking an optimized frame the frame pointer is
  // adjusted with:
  int diff = (method->max_locals() - method->size_of_parameters()) *
             Interpreter::stackElementWords();
  return _fp == (fp - diff);
}

void frame::pd_gc_epilog() {
  // nothing done here now
}

bool frame::is_interpreted_frame_valid(JavaThread* thread) const {
// QQQ
#ifdef CC_INTERP
#else
  assert(is_interpreted_frame(), "Not an interpreted frame");
  // These are reasonable sanity checks
  if (fp() == 0 || (intptr_t(fp()) & (wordSize-1)) != 0) {
    return false;
  }
  if (sp() == 0 || (intptr_t(sp()) & (wordSize-1)) != 0) {
    return false;
  }
  if (fp() + interpreter_frame_initial_sp_offset < sp()) {
    return false;
  }
  // These are hacks to keep us out of trouble.
  // The problem with these is that they mask other problems
  if (fp() <= sp()) {        // this attempts to deal with unsigned comparison above
    return false;
  }

  // do some validation of frame elements

  // first the method

  methodOop m = *interpreter_frame_method_addr();

  // validate the method we'd find in this potential sender
  if (!Universe::heap()->is_valid_method(m)) return false;

  // stack frames shouldn't be much larger than max_stack elements

  if (fp() - sp() > 1024 + m->max_stack()*Interpreter::stackElementSize()) {
    return false;
  }

  // validate bci/bcx

  intptr_t  bcx    = interpreter_frame_bcx();
  if (m->validate_bci_from_bcx(bcx) < 0) {
    return false;
  }

  // validate constantPoolCacheOop

  constantPoolCacheOop cp = *interpreter_frame_cache_addr();

  if (cp == NULL ||
      !Space::is_aligned(cp) ||
      !Universe::heap()->is_permanent((void*)cp)) return false;

  // validate locals

  address locals =  (address) *interpreter_frame_locals_addr();

  if (locals > thread->stack_base() || locals < (address) fp()) return false;

  // We'd have to be pretty unlucky to be mislead at this point

#endif // CC_INTERP
  return true;
}

BasicType frame::interpreter_frame_result(oop* oop_result, jvalue* value_result) {
#ifdef CC_INTERP
  // Needed for JVMTI. The result should always be in the interpreterState object
  assert(false, "NYI");
  interpreterState istate = get_interpreterState();
#endif // CC_INTERP
  assert(is_interpreted_frame(), "interpreted frame expected");
  methodOop method = interpreter_frame_method();
  BasicType type = method->result_type();

  intptr_t* tos_addr;
  if (method->is_native()) {
    // Prior to calling into the runtime to report the method_exit the possible
    // return value is pushed to the native stack. If the result is a jfloat/jdouble
    // then ST0 is saved before EAX/EDX. See the note in generate_native_result
    tos_addr = (intptr_t*)sp();
    if (type == T_FLOAT || type == T_DOUBLE) {
    // QQQ seems like this code is equivalent on the two platforms
#ifdef AMD64
      // This is times two because we do a push(ltos) after pushing XMM0
      // and that takes two interpreter stack slots.
      tos_addr += 2 * Interpreter::stackElementWords();
#else
      tos_addr += 2;
#endif // AMD64
    }
  } else {
    tos_addr = (intptr_t*)interpreter_frame_tos_address();
  }

  switch (type) {
    case T_OBJECT  :
    case T_ARRAY   : {
      oop obj;
      if (method->is_native()) {
#ifdef CC_INTERP
        obj = istate->_oop_temp;
#else
        obj = (oop) at(interpreter_frame_oop_temp_offset);
#endif // CC_INTERP
      } else {
        oop* obj_p = (oop*)tos_addr;
        obj = (obj_p == NULL) ? (oop)NULL : *obj_p;
      }
      assert(obj == NULL || Universe::heap()->is_in(obj), "sanity check");
      *oop_result = obj;
      break;
    }
    case T_BOOLEAN : value_result->z = *(jboolean*)tos_addr; break;
    case T_BYTE    : value_result->b = *(jbyte*)tos_addr; break;
    case T_CHAR    : value_result->c = *(jchar*)tos_addr; break;
    case T_SHORT   : value_result->s = *(jshort*)tos_addr; break;
    case T_INT     : value_result->i = *(jint*)tos_addr; break;
    case T_LONG    : value_result->j = *(jlong*)tos_addr; break;
    case T_FLOAT   : {
#ifdef AMD64
        value_result->f = *(jfloat*)tos_addr;
#else
      if (method->is_native()) {
        jdouble d = *(jdouble*)tos_addr;  // Result was in ST0 so need to convert to jfloat
        value_result->f = (jfloat)d;
      } else {
        value_result->f = *(jfloat*)tos_addr;
      }
#endif // AMD64
      break;
    }
    case T_DOUBLE  : value_result->d = *(jdouble*)tos_addr; break;
    case T_VOID    : /* Nothing to do */ break;
    default        : ShouldNotReachHere();
  }

  return type;
}


intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
  int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);
  return &interpreter_frame_tos_address()[index];
}