duke@435: /*
mikael@6198:  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
trims@1907:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907:  * or visit www.oracle.com if you need additional information or have any
trims@1907:  * questions.
duke@435:  *
duke@435:  */
duke@435: 
stefank@2314: #include "precompiled.hpp"
twisti@4318: #include "asm/macroAssembler.hpp"
twisti@4318: #include "asm/macroAssembler.inline.hpp"
stefank@2314: #include "asm/codeBuffer.hpp"
twisti@4318: #include "runtime/atomic.hpp"
twisti@4318: #include "runtime/atomic.inline.hpp"
stefank@2314: #include "runtime/icache.hpp"
stefank@2314: #include "runtime/os.hpp"
duke@435: 
duke@435: 
duke@435: // Implementation of AbstractAssembler
duke@435: //
duke@435: // The AbstractAssembler is generating code into a CodeBuffer. To make code generation faster,
duke@435: // the assembler keeps a copy of the code buffers boundaries & modifies them when
duke@435: // emitting bytes rather than using the code buffers accessor functions all the time.
twisti@1040: // The code buffer is updated via set_code_end(...) after emitting a whole instruction.
duke@435: 
duke@435: AbstractAssembler::AbstractAssembler(CodeBuffer* code) {
duke@435:   if (code == NULL)  return;
duke@435:   CodeSection* cs = code->insts();
duke@435:   cs->clear_mark();   // new assembler kills old mark
twisti@4317:   if (cs->start() == NULL)  {
ccheung@4993:     vm_exit_out_of_memory(0, OOM_MMAP_ERROR, err_msg("CodeCache: no room for %s",
jcoomes@1845:                                      code->name()));
duke@435:   }
twisti@4317:   _code_section = cs;
twisti@4317:   _oop_recorder= code->oop_recorder();
twisti@4317:   DEBUG_ONLY( _short_branch_delta = 0; )
duke@435: }
duke@435: 
duke@435: void AbstractAssembler::set_code_section(CodeSection* cs) {
duke@435:   assert(cs->outer() == code_section()->outer(), "sanity");
duke@435:   assert(cs->is_allocated(), "need to pre-allocate this section");
duke@435:   cs->clear_mark();  // new assembly into this section kills old mark
duke@435:   _code_section = cs;
duke@435: }
duke@435: 
duke@435: // Inform CodeBuffer that incoming code and relocation will be for stubs
duke@435: address AbstractAssembler::start_a_stub(int required_space) {
duke@435:   CodeBuffer*  cb = code();
duke@435:   CodeSection* cs = cb->stubs();
duke@435:   assert(_code_section == cb->insts(), "not in insts?");
duke@435:   if (cs->maybe_expand_to_ensure_remaining(required_space)
duke@435:       && cb->blob() == NULL) {
duke@435:     return NULL;
duke@435:   }
duke@435:   set_code_section(cs);
duke@435:   return pc();
duke@435: }
duke@435: 
duke@435: // Inform CodeBuffer that incoming code and relocation will be code
duke@435: // Should not be called if start_a_stub() returned NULL
duke@435: void AbstractAssembler::end_a_stub() {
duke@435:   assert(_code_section == code()->stubs(), "not in stubs?");
duke@435:   set_code_section(code()->insts());
duke@435: }
duke@435: 
duke@435: // Inform CodeBuffer that incoming code and relocation will be for stubs
duke@435: address AbstractAssembler::start_a_const(int required_space, int required_align) {
duke@435:   CodeBuffer*  cb = code();
duke@435:   CodeSection* cs = cb->consts();
kvn@4316:   assert(_code_section == cb->insts() || _code_section == cb->stubs(), "not in insts/stubs?");
duke@435:   address end = cs->end();
duke@435:   int pad = -(intptr_t)end & (required_align-1);
duke@435:   if (cs->maybe_expand_to_ensure_remaining(pad + required_space)) {
duke@435:     if (cb->blob() == NULL)  return NULL;
duke@435:     end = cs->end();  // refresh pointer
duke@435:   }
duke@435:   if (pad > 0) {
duke@435:     while (--pad >= 0) { *end++ = 0; }
duke@435:     cs->set_end(end);
duke@435:   }
duke@435:   set_code_section(cs);
duke@435:   return end;
duke@435: }
duke@435: 
duke@435: // Inform CodeBuffer that incoming code and relocation will be code
kvn@4316: // in section cs (insts or stubs).
kvn@4316: void AbstractAssembler::end_a_const(CodeSection* cs) {
duke@435:   assert(_code_section == code()->consts(), "not in consts?");
kvn@4316:   set_code_section(cs);
duke@435: }
duke@435: 
duke@435: void AbstractAssembler::flush() {
duke@435:   ICache::invalidate_range(addr_at(0), offset());
duke@435: }
duke@435: 
duke@435: void AbstractAssembler::bind(Label& L) {
duke@435:   if (L.is_bound()) {
duke@435:     // Assembler can bind a label more than once to the same place.
duke@435:     guarantee(L.loc() == locator(), "attempt to redefine label");
duke@435:     return;
duke@435:   }
duke@435:   L.bind_loc(locator());
duke@435:   L.patch_instructions((MacroAssembler*)this);
duke@435: }
duke@435: 
duke@435: void AbstractAssembler::generate_stack_overflow_check( int frame_size_in_bytes) {
duke@435:   if (UseStackBanging) {
duke@435:     // Each code entry causes one stack bang n pages down the stack where n
mikael@6072:     // is configurable by StackShadowPages.  The setting depends on the maximum
duke@435:     // depth of VM call stack or native before going back into java code,
duke@435:     // since only java code can raise a stack overflow exception using the
duke@435:     // stack banging mechanism.  The VM and native code does not detect stack
duke@435:     // overflow.
duke@435:     // The code in JavaCalls::call() checks that there is at least n pages
duke@435:     // available, so all entry code needs to do is bang once for the end of
duke@435:     // this shadow zone.
duke@435:     // The entry code may need to bang additional pages if the framesize
duke@435:     // is greater than a page.
duke@435: 
duke@435:     const int page_size = os::vm_page_size();
duke@435:     int bang_end = StackShadowPages*page_size;
duke@435: 
duke@435:     // This is how far the previous frame's stack banging extended.
duke@435:     const int bang_end_safe = bang_end;
duke@435: 
duke@435:     if (frame_size_in_bytes > page_size) {
duke@435:       bang_end += frame_size_in_bytes;
duke@435:     }
duke@435: 
duke@435:     int bang_offset = bang_end_safe;
duke@435:     while (bang_offset <= bang_end) {
duke@435:       // Need at least one stack bang at end of shadow zone.
duke@435:       bang_stack_with_offset(bang_offset);
duke@435:       bang_offset += page_size;
duke@435:     }
duke@435:   } // end (UseStackBanging)
duke@435: }
duke@435: 
duke@435: void Label::add_patch_at(CodeBuffer* cb, int branch_loc) {
duke@435:   assert(_loc == -1, "Label is unbound");
duke@435:   if (_patch_index < PatchCacheSize) {
duke@435:     _patches[_patch_index] = branch_loc;
duke@435:   } else {
duke@435:     if (_patch_overflow == NULL) {
duke@435:       _patch_overflow = cb->create_patch_overflow();
duke@435:     }
duke@435:     _patch_overflow->push(branch_loc);
duke@435:   }
duke@435:   ++_patch_index;
duke@435: }
duke@435: 
duke@435: void Label::patch_instructions(MacroAssembler* masm) {
duke@435:   assert(is_bound(), "Label is bound");
duke@435:   CodeBuffer* cb = masm->code();
duke@435:   int target_sect = CodeBuffer::locator_sect(loc());
duke@435:   address target = cb->locator_address(loc());
duke@435:   while (_patch_index > 0) {
duke@435:     --_patch_index;
duke@435:     int branch_loc;
duke@435:     if (_patch_index >= PatchCacheSize) {
duke@435:       branch_loc = _patch_overflow->pop();
duke@435:     } else {
duke@435:       branch_loc = _patches[_patch_index];
duke@435:     }
duke@435:     int branch_sect = CodeBuffer::locator_sect(branch_loc);
duke@435:     address branch = cb->locator_address(branch_loc);
duke@435:     if (branch_sect == CodeBuffer::SECT_CONSTS) {
duke@435:       // The thing to patch is a constant word.
duke@435:       *(address*)branch = target;
duke@435:       continue;
duke@435:     }
duke@435: 
duke@435: #ifdef ASSERT
duke@435:     // Cross-section branches only work if the
duke@435:     // intermediate section boundaries are frozen.
duke@435:     if (target_sect != branch_sect) {
duke@435:       for (int n = MIN2(target_sect, branch_sect),
duke@435:                nlimit = (target_sect + branch_sect) - n;
duke@435:            n < nlimit; n++) {
duke@435:         CodeSection* cs = cb->code_section(n);
duke@435:         assert(cs->is_frozen(), "cross-section branch needs stable offsets");
duke@435:       }
duke@435:     }
duke@435: #endif //ASSERT
duke@435: 
duke@435:     // Push the target offset into the branch instruction.
duke@435:     masm->pd_patch_instruction(branch, target);
duke@435:   }
duke@435: }
duke@435: 
jrose@1057: struct DelayedConstant {
jrose@1057:   typedef void (*value_fn_t)();
jrose@1057:   BasicType type;
jrose@1057:   intptr_t value;
jrose@1057:   value_fn_t value_fn;
jrose@1057:   // This limit of 20 is generous for initial uses.
jrose@1057:   // The limit needs to be large enough to store the field offsets
jrose@1057:   // into classes which do not have statically fixed layouts.
jrose@1057:   // (Initial use is for method handle object offsets.)
jrose@1057:   // Look for uses of "delayed_value" in the source code
jrose@1057:   // and make sure this number is generous enough to handle all of them.
jrose@1057:   enum { DC_LIMIT = 20 };
jrose@1057:   static DelayedConstant delayed_constants[DC_LIMIT];
jrose@1057:   static DelayedConstant* add(BasicType type, value_fn_t value_fn);
jrose@1057:   bool match(BasicType t, value_fn_t cfn) {
jrose@1057:     return type == t && value_fn == cfn;
jrose@1057:   }
jrose@1057:   static void update_all();
jrose@1057: };
jrose@1057: 
jrose@1057: DelayedConstant DelayedConstant::delayed_constants[DC_LIMIT];
jrose@1057: // Default C structure initialization rules have the following effect here:
jrose@1057: // = { { (BasicType)0, (intptr_t)NULL }, ... };
jrose@1057: 
jrose@1057: DelayedConstant* DelayedConstant::add(BasicType type,
jrose@1057:                                       DelayedConstant::value_fn_t cfn) {
jrose@1057:   for (int i = 0; i < DC_LIMIT; i++) {
jrose@1057:     DelayedConstant* dcon = &delayed_constants[i];
jrose@1057:     if (dcon->match(type, cfn))
jrose@1057:       return dcon;
jrose@1057:     if (dcon->value_fn == NULL) {
jrose@1057:       // (cmpxchg not because this is multi-threaded but because I'm paranoid)
jrose@1057:       if (Atomic::cmpxchg_ptr(CAST_FROM_FN_PTR(void*, cfn), &dcon->value_fn, NULL) == NULL) {
jrose@1057:         dcon->type = type;
jrose@1057:         return dcon;
jrose@1057:       }
jrose@1057:     }
jrose@1057:   }
jrose@1057:   // If this assert is hit (in pre-integration testing!) then re-evaluate
jrose@1057:   // the comment on the definition of DC_LIMIT.
jrose@1057:   guarantee(false, "too many delayed constants");
jrose@1057:   return NULL;
jrose@1057: }
jrose@1057: 
jrose@1057: void DelayedConstant::update_all() {
jrose@1057:   for (int i = 0; i < DC_LIMIT; i++) {
jrose@1057:     DelayedConstant* dcon = &delayed_constants[i];
jrose@1057:     if (dcon->value_fn != NULL && dcon->value == 0) {
jrose@1057:       typedef int     (*int_fn_t)();
jrose@1057:       typedef address (*address_fn_t)();
jrose@1057:       switch (dcon->type) {
jrose@1057:       case T_INT:     dcon->value = (intptr_t) ((int_fn_t)    dcon->value_fn)(); break;
jrose@1057:       case T_ADDRESS: dcon->value = (intptr_t) ((address_fn_t)dcon->value_fn)(); break;
jrose@1057:       }
jrose@1057:     }
jrose@1057:   }
jrose@1057: }
jrose@1057: 
twisti@3969: RegisterOrConstant AbstractAssembler::delayed_value(int(*value_fn)(), Register tmp, int offset) {
twisti@3969:   intptr_t val = (intptr_t) (*value_fn)();
twisti@3969:   if (val != 0)  return val + offset;
twisti@3969:   return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
twisti@3969: }
twisti@3969: RegisterOrConstant AbstractAssembler::delayed_value(address(*value_fn)(), Register tmp, int offset) {
twisti@3969:   intptr_t val = (intptr_t) (*value_fn)();
twisti@3969:   if (val != 0)  return val + offset;
twisti@3969:   return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
twisti@3969: }
jrose@1057: intptr_t* AbstractAssembler::delayed_value_addr(int(*value_fn)()) {
jrose@1057:   DelayedConstant* dcon = DelayedConstant::add(T_INT, (DelayedConstant::value_fn_t) value_fn);
jrose@1057:   return &dcon->value;
jrose@1057: }
jrose@1057: intptr_t* AbstractAssembler::delayed_value_addr(address(*value_fn)()) {
jrose@1057:   DelayedConstant* dcon = DelayedConstant::add(T_ADDRESS, (DelayedConstant::value_fn_t) value_fn);
jrose@1057:   return &dcon->value;
jrose@1057: }
jrose@1057: void AbstractAssembler::update_delayed_values() {
jrose@1057:   DelayedConstant::update_all();
jrose@1057: }
jrose@1057: 
duke@435: void AbstractAssembler::block_comment(const char* comment) {
duke@435:   if (sect() == CodeBuffer::SECT_INSTS) {
duke@435:     code_section()->outer()->block_comment(offset(), comment);
duke@435:   }
duke@435: }
duke@435: 
roland@4767: const char* AbstractAssembler::code_string(const char* str) {
roland@4767:   if (sect() == CodeBuffer::SECT_INSTS || sect() == CodeBuffer::SECT_STUBS) {
roland@4767:     return code_section()->outer()->code_string(str);
roland@4767:   }
roland@4767:   return NULL;
roland@4767: }
roland@4767: 
coleenp@672: bool MacroAssembler::needs_explicit_null_check(intptr_t offset) {
coleenp@672:   // Exception handler checks the nmethod's implicit null checks table
coleenp@672:   // only when this method returns false.
kvn@1077: #ifdef _LP64
kvn@1077:   if (UseCompressedOops && Universe::narrow_oop_base() != NULL) {
kvn@1077:     assert (Universe::heap() != NULL, "java heap should be initialized");
coleenp@672:     // The first page after heap_base is unmapped and
coleenp@672:     // the 'offset' is equal to [heap_base + offset] for
coleenp@672:     // narrow oop implicit null checks.
kvn@1077:     uintptr_t base = (uintptr_t)Universe::narrow_oop_base();
kvn@1077:     if ((uintptr_t)offset >= base) {
coleenp@672:       // Normalize offset for the next check.
kvn@1077:       offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1));
coleenp@672:     }
coleenp@672:   }
kvn@1077: #endif
coleenp@672:   return offset < 0 || os::vm_page_size() <= offset;
coleenp@672: }