1 /*

     2  * Copyright 1997-2007 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,

    20  * CA 95054 USA or visit www.sun.com if you need additional information or

    21  * have any questions.

    22  *

    23  */

    24 

    25 # include "incls/_precompiled.incl"

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

    27 

    28 #ifdef ASSERT

    29 void RegisterMap::check_location_valid() {

    30 }

    31 #endif

    32 

    33 

    34 // Profiling/safepoint support

    35 

    36 bool frame::safe_for_sender(JavaThread *thread) {

    37   address   sp = (address)_sp;

    38   address   fp = (address)_fp;

    39   address   unextended_sp = (address)_unextended_sp;

    40   bool sp_safe = (sp != NULL &&

    41                  (sp <= thread->stack_base()) &&

    42                  (sp >= thread->stack_base() - thread->stack_size()));

    43   bool unextended_sp_safe = (unextended_sp != NULL &&

    44                  (unextended_sp <= thread->stack_base()) &&

    45                  (unextended_sp >= thread->stack_base() - thread->stack_size()));

    46   bool fp_safe = (fp != NULL &&

    47                  (fp <= thread->stack_base()) &&

    48                  (fp >= thread->stack_base() - thread->stack_size()));

    49   if (sp_safe && unextended_sp_safe && fp_safe) {

    50     // Unfortunately we can only check frame complete for runtime stubs and nmethod

    51     // other generic buffer blobs are more problematic so we just assume they are

    52     // ok. adapter blobs never have a frame complete and are never ok.

    53     if (_cb != NULL && !_cb->is_frame_complete_at(_pc)) {

    54       if (_cb->is_nmethod() || _cb->is_adapter_blob() || _cb->is_runtime_stub()) {

    55         return false;

    56       }

    57     }

    58     return true;

    59   }

    60   // Note: fp == NULL is not really a prerequisite for this to be safe to

    61   // walk for c2. However we've modified the code such that if we get

    62   // a failure with fp != NULL that we then try with FP == NULL.

    63   // This is basically to mimic what a last_frame would look like if

    64   // c2 had generated it.

    65   if (sp_safe && unextended_sp_safe && fp == NULL) {

    66     // frame must be complete if fp == NULL as fp == NULL is only sensible

    67     // if we are looking at a nmethod and frame complete assures us of that.

    68     if (_cb != NULL && _cb->is_frame_complete_at(_pc) && _cb->is_compiled_by_c2()) {

    69         return true;

    70     }

    71   }

    72   return false;

    73 }

    74 

    75 

    76 void frame::patch_pc(Thread* thread, address pc) {

    77   if (TracePcPatching) {

    78     tty->print_cr("patch_pc at address  0x%x [0x%x -> 0x%x] ", &((address *)sp())[-1], ((address *)sp())[-1], pc);

    79   }

    80   ((address *)sp())[-1] = pc;

    81   _cb = CodeCache::find_blob(pc);

    82   if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) {

    83     address orig = (((nmethod*)_cb)->get_original_pc(this));

    84     assert(orig == _pc, "expected original to be stored before patching");

    85     _deopt_state = is_deoptimized;

    86     // leave _pc as is

    87   } else {

    88     _deopt_state = not_deoptimized;

    89     _pc = pc;

    90   }

    91 }

    92 

    93 bool frame::is_interpreted_frame() const  {

    94   return Interpreter::contains(pc());

    95 }

    96 

    97 int frame::frame_size() const {

    98   RegisterMap map(JavaThread::current(), false);

    99   frame sender = this->sender(&map);

   100   return sender.sp() - sp();

   101 }

   102 

   103 intptr_t* frame::entry_frame_argument_at(int offset) const {

   104   // convert offset to index to deal with tsi

   105   int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);

   106   // Entry frame's arguments are always in relation to unextended_sp()

   107   return &unextended_sp()[index];

   108 }

   109 

   110 // sender_sp

   111 #ifdef CC_INTERP

   112 intptr_t* frame::interpreter_frame_sender_sp() const {

   113   assert(is_interpreted_frame(), "interpreted frame expected");

   114   // QQQ why does this specialize method exist if frame::sender_sp() does same thing?

   115   // seems odd and if we always know interpreted vs. non then sender_sp() is really

   116   // doing too much work.

   117   return get_interpreterState()->sender_sp();

   118 }

   119 

   120 // monitor elements

   121 

   122 BasicObjectLock* frame::interpreter_frame_monitor_begin() const {

   123   return get_interpreterState()->monitor_base();

   124 }

   125 

   126 BasicObjectLock* frame::interpreter_frame_monitor_end() const {

   127   return (BasicObjectLock*) get_interpreterState()->stack_base();

   128 }

   129 

   130 #else // CC_INTERP

   131 

   132 intptr_t* frame::interpreter_frame_sender_sp() const {

   133   assert(is_interpreted_frame(), "interpreted frame expected");

   134   return (intptr_t*) at(interpreter_frame_sender_sp_offset);

   135 }

   136 

   137 void frame::set_interpreter_frame_sender_sp(intptr_t* sender_sp) {

   138   assert(is_interpreted_frame(), "interpreted frame expected");

   139   ptr_at_put(interpreter_frame_sender_sp_offset, (intptr_t) sender_sp);

   140 }

   141 

   142 

   143 // monitor elements

   144 

   145 BasicObjectLock* frame::interpreter_frame_monitor_begin() const {

   146   return (BasicObjectLock*) addr_at(interpreter_frame_monitor_block_bottom_offset);

   147 }

   148 

   149 BasicObjectLock* frame::interpreter_frame_monitor_end() const {

   150   BasicObjectLock* result = (BasicObjectLock*) *addr_at(interpreter_frame_monitor_block_top_offset);

   151   // make sure the pointer points inside the frame

   152   assert((intptr_t) fp() >  (intptr_t) result, "result must <  than frame pointer");

   153   assert((intptr_t) sp() <= (intptr_t) result, "result must >= than stack pointer");

   154   return result;

   155 }

   156 

   157 void frame::interpreter_frame_set_monitor_end(BasicObjectLock* value) {

   158   *((BasicObjectLock**)addr_at(interpreter_frame_monitor_block_top_offset)) = value;

   159 }

   160 

   161 // Used by template based interpreter deoptimization

   162 void frame::interpreter_frame_set_last_sp(intptr_t* sp) {

   163     *((intptr_t**)addr_at(interpreter_frame_last_sp_offset)) = sp;

   164 }

   165 #endif // CC_INTERP

   166 

   167 frame frame::sender_for_entry_frame(RegisterMap* map) const {

   168   assert(map != NULL, "map must be set");

   169   // Java frame called from C; skip all C frames and return top C

   170   // frame of that chunk as the sender

   171   JavaFrameAnchor* jfa = entry_frame_call_wrapper()->anchor();

   172   assert(!entry_frame_is_first(), "next Java fp must be non zero");

   173   assert(jfa->last_Java_sp() > sp(), "must be above this frame on stack");

   174   map->clear();

   175   assert(map->include_argument_oops(), "should be set by clear");

   176   if (jfa->last_Java_pc() != NULL ) {

   177     frame fr(jfa->last_Java_sp(), jfa->last_Java_fp(), jfa->last_Java_pc());

   178     return fr;

   179   }

   180   frame fr(jfa->last_Java_sp(), jfa->last_Java_fp());

   181   return fr;

   182 }

   183 

   184 frame frame::sender_for_interpreter_frame(RegisterMap* map) const {

   185   // sp is the raw sp from the sender after adapter or interpreter extension

   186   intptr_t* sp = (intptr_t*) addr_at(sender_sp_offset);

   187 

   188   // This is the sp before any possible extension (adapter/locals).

   189   intptr_t* unextended_sp = interpreter_frame_sender_sp();

   190 

   191   // The interpreter and compiler(s) always save EBP/RBP in a known

   192   // location on entry. We must record where that location is

   193   // so this if EBP/RBP was live on callout from c2 we can find

   194   // the saved copy no matter what it called.

   195 

   196   // Since the interpreter always saves EBP/RBP if we record where it is then

   197   // we don't have to always save EBP/RBP on entry and exit to c2 compiled

   198   // code, on entry will be enough.

   199 #ifdef COMPILER2

   200   if (map->update_map()) {

   201     map->set_location(rbp->as_VMReg(), (address) addr_at(link_offset));

   202 #ifdef AMD64

   203     // this is weird "H" ought to be at a higher address however the

   204     // oopMaps seems to have the "H" regs at the same address and the

   205     // vanilla register.

   206     // XXXX make this go away

   207     if (true) {

   208       map->set_location(rbp->as_VMReg()->next(), (address)addr_at(link_offset));

   209     }

   210 #endif // AMD64

   211   }

   212 #endif /* COMPILER2 */

   213   return frame(sp, unextended_sp, link(), sender_pc());

   214 }

   215 

   216 

   217 //------------------------------sender_for_compiled_frame-----------------------

   218 frame frame::sender_for_compiled_frame(RegisterMap* map) const {

   219   assert(map != NULL, "map must be set");

   220   const bool c1_compiled = _cb->is_compiled_by_c1();

   221 

   222   // frame owned by optimizing compiler

   223   intptr_t* sender_sp = NULL;

   224 

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

   226   sender_sp = unextended_sp() + _cb->frame_size();

   227 

   228   // On Intel the return_address is always the word on the stack

   229   address sender_pc = (address) *(sender_sp-1);

   230 

   231   // This is the saved value of ebp which may or may not really be an fp.

   232   // it is only an fp if the sender is an interpreter frame (or c1?)

   233 

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

   235 

   236   if (map->update_map()) {

   237     // Tell GC to use argument oopmaps for some runtime stubs that need it.

   238     // For C1, the runtime stub might not have oop maps, so set this flag

   239     // outside of update_register_map.

   240     map->set_include_argument_oops(_cb->caller_must_gc_arguments(map->thread()));

   241     if (_cb->oop_maps() != NULL) {

   242       OopMapSet::update_register_map(this, map);

   243     }

   244     // Since the prolog does the save and restore of epb there is no oopmap

   245     // for it so we must fill in its location as if there was an oopmap entry

   246     // since if our caller was compiled code there could be live jvm state in it.

   247     map->set_location(rbp->as_VMReg(), (address) (sender_sp - frame::sender_sp_offset));

   248 #ifdef AMD64

   249     // this is weird "H" ought to be at a higher address however the

   250     // oopMaps seems to have the "H" regs at the same address and the

   251     // vanilla register.

   252     // XXXX make this go away

   253     if (true) {

   254       map->set_location(rbp->as_VMReg()->next(), (address) (sender_sp - frame::sender_sp_offset));

   255     }

   256 #endif // AMD64

   257   }

   258 

   259   assert(sender_sp != sp(), "must have changed");

   260   return frame(sender_sp, saved_fp, sender_pc);

   261 }

   262 

   263 frame frame::sender(RegisterMap* map) const {

   264   // Default is we done have to follow them. The sender_for_xxx will

   265   // update it accordingly

   266   map->set_include_argument_oops(false);

   267 

   268   if (is_entry_frame())       return sender_for_entry_frame(map);

   269   if (is_interpreted_frame()) return sender_for_interpreter_frame(map);

   270   assert(_cb == CodeCache::find_blob(pc()),"Must be the same");

   271 

   272   if (_cb != NULL) {

   273     return sender_for_compiled_frame(map);

   274   }

   275   // Must be native-compiled frame, i.e. the marshaling code for native

   276   // methods that exists in the core system.

   277   return frame(sender_sp(), link(), sender_pc());

   278 }

   279 

   280 

   281 bool frame::interpreter_frame_equals_unpacked_fp(intptr_t* fp) {

   282   assert(is_interpreted_frame(), "must be interpreter frame");

   283   methodOop method = interpreter_frame_method();

   284   // When unpacking an optimized frame the frame pointer is

   285   // adjusted with:

   286   int diff = (method->max_locals() - method->size_of_parameters()) *

   287              Interpreter::stackElementWords();

   288   return _fp == (fp - diff);

   289 }

   290 

   291 void frame::pd_gc_epilog() {

   292   // nothing done here now

   293 }

   294 

   295 bool frame::is_interpreted_frame_valid() const {

   296 // QQQ

   297 #ifdef CC_INTERP

   298 #else

   299   assert(is_interpreted_frame(), "Not an interpreted frame");

   300   // These are reasonable sanity checks

   301   if (fp() == 0 || (intptr_t(fp()) & (wordSize-1)) != 0) {

   302     return false;

   303   }

   304   if (sp() == 0 || (intptr_t(sp()) & (wordSize-1)) != 0) {

   305     return false;

   306   }

   307   if (fp() + interpreter_frame_initial_sp_offset < sp()) {

   308     return false;

   309   }

   310   // These are hacks to keep us out of trouble.

   311   // The problem with these is that they mask other problems

   312   if (fp() <= sp()) {        // this attempts to deal with unsigned comparison above

   313     return false;

   314   }

   315   if (fp() - sp() > 4096) {  // stack frames shouldn't be large.

   316     return false;

   317   }

   318 #endif // CC_INTERP

   319   return true;

   320 }

   321 

   322 BasicType frame::interpreter_frame_result(oop* oop_result, jvalue* value_result) {

   323 #ifdef CC_INTERP

   324   // Needed for JVMTI. The result should always be in the interpreterState object

   325   assert(false, "NYI");

   326   interpreterState istate = get_interpreterState();

   327 #endif // CC_INTERP

   328   assert(is_interpreted_frame(), "interpreted frame expected");

   329   methodOop method = interpreter_frame_method();

   330   BasicType type = method->result_type();

   331 

   332   intptr_t* tos_addr;

   333   if (method->is_native()) {

   334     // Prior to calling into the runtime to report the method_exit the possible

   335     // return value is pushed to the native stack. If the result is a jfloat/jdouble

   336     // then ST0 is saved before EAX/EDX. See the note in generate_native_result

   337     tos_addr = (intptr_t*)sp();

   338     if (type == T_FLOAT || type == T_DOUBLE) {

   339     // QQQ seems like this code is equivalent on the two platforms

   340 #ifdef AMD64

   341       // This is times two because we do a push(ltos) after pushing XMM0

   342       // and that takes two interpreter stack slots.

   343       tos_addr += 2 * Interpreter::stackElementWords();

   344 #else

   345       tos_addr += 2;

   346 #endif // AMD64

   347     }

   348   } else {

   349     tos_addr = (intptr_t*)interpreter_frame_tos_address();

   350   }

   351 

   352   switch (type) {

   353     case T_OBJECT  :

   354     case T_ARRAY   : {

   355       oop obj;

   356       if (method->is_native()) {

   357 #ifdef CC_INTERP

   358         obj = istate->_oop_temp;

   359 #else

   360         obj = (oop) at(interpreter_frame_oop_temp_offset);

   361 #endif // CC_INTERP

   362       } else {

   363         oop* obj_p = (oop*)tos_addr;

   364         obj = (obj_p == NULL) ? (oop)NULL : *obj_p;

   365       }

   366       assert(obj == NULL || Universe::heap()->is_in(obj), "sanity check");

   367       *oop_result = obj;

   368       break;

   369     }

   370     case T_BOOLEAN : value_result->z = *(jboolean*)tos_addr; break;

   371     case T_BYTE    : value_result->b = *(jbyte*)tos_addr; break;

   372     case T_CHAR    : value_result->c = *(jchar*)tos_addr; break;

   373     case T_SHORT   : value_result->s = *(jshort*)tos_addr; break;

   374     case T_INT     : value_result->i = *(jint*)tos_addr; break;

   375     case T_LONG    : value_result->j = *(jlong*)tos_addr; break;

   376     case T_FLOAT   : {

   377 #ifdef AMD64

   378         value_result->f = *(jfloat*)tos_addr;

   379 #else

   380       if (method->is_native()) {

   381         jdouble d = *(jdouble*)tos_addr;  // Result was in ST0 so need to convert to jfloat

   382         value_result->f = (jfloat)d;

   383       } else {

   384         value_result->f = *(jfloat*)tos_addr;

   385       }

   386 #endif // AMD64

   387       break;

   388     }

   389     case T_DOUBLE  : value_result->d = *(jdouble*)tos_addr; break;

   390     case T_VOID    : /* Nothing to do */ break;

   391     default        : ShouldNotReachHere();

   392   }

   393 

   394   return type;

   395 }

   396 

   397 

   398 intptr_t* frame::interpreter_frame_tos_at(jint offset) const {

   399   int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);

   400   return &interpreter_frame_tos_address()[index];

   401 }