1 /* 2 * Copyright (c) 1997, 2013, 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 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "asm/macroAssembler.inline.hpp" 28 #include "asm/codeBuffer.hpp" 29 #include "runtime/atomic.hpp" 30 #include "runtime/atomic.inline.hpp" 31 #include "runtime/icache.hpp" 32 #include "runtime/os.hpp" 33 34 35 // Implementation of AbstractAssembler 36 // 37 // The AbstractAssembler is generating code into a CodeBuffer. To make code generation faster, 38 // the assembler keeps a copy of the code buffers boundaries & modifies them when 39 // emitting bytes rather than using the code buffers accessor functions all the time. 40 // The code buffer is updated via set_code_end(...) after emitting a whole instruction. 41 42 AbstractAssembler::AbstractAssembler(CodeBuffer* code) { 43 if (code == NULL) return; 44 CodeSection* cs = code->insts(); 45 cs->clear_mark(); // new assembler kills old mark 46 if (cs->start() == NULL) { 47 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, err_msg("CodeCache: no room for %s", 48 code->name())); 49 } 50 _code_section = cs; 51 _oop_recorder= code->oop_recorder(); 52 DEBUG_ONLY( _short_branch_delta = 0; ) 53 } 54 55 void AbstractAssembler::set_code_section(CodeSection* cs) { 56 assert(cs->outer() == code_section()->outer(), "sanity"); 57 assert(cs->is_allocated(), "need to pre-allocate this section"); 58 cs->clear_mark(); // new assembly into this section kills old mark 59 _code_section = cs; 60 } 61 62 // Inform CodeBuffer that incoming code and relocation will be for stubs 63 address AbstractAssembler::start_a_stub(int required_space) { 64 CodeBuffer* cb = code(); 65 CodeSection* cs = cb->stubs(); 66 assert(_code_section == cb->insts(), "not in insts?"); 67 if (cs->maybe_expand_to_ensure_remaining(required_space) 68 && cb->blob() == NULL) { 69 return NULL; 70 } 71 set_code_section(cs); 72 return pc(); 73 } 74 75 // Inform CodeBuffer that incoming code and relocation will be code 76 // Should not be called if start_a_stub() returned NULL 77 void AbstractAssembler::end_a_stub() { 78 assert(_code_section == code()->stubs(), "not in stubs?"); 79 set_code_section(code()->insts()); 80 } 81 82 // Inform CodeBuffer that incoming code and relocation will be for stubs 83 address AbstractAssembler::start_a_const(int required_space, int required_align) { 84 CodeBuffer* cb = code(); 85 CodeSection* cs = cb->consts(); 86 assert(_code_section == cb->insts() || _code_section == cb->stubs(), "not in insts/stubs?"); 87 address end = cs->end(); 88 int pad = -(intptr_t)end & (required_align-1); 89 if (cs->maybe_expand_to_ensure_remaining(pad + required_space)) { 90 if (cb->blob() == NULL) return NULL; 91 end = cs->end(); // refresh pointer 92 } 93 if (pad > 0) { 94 while (--pad >= 0) { *end++ = 0; } 95 cs->set_end(end); 96 } 97 set_code_section(cs); 98 return end; 99 } 100 101 // Inform CodeBuffer that incoming code and relocation will be code 102 // in section cs (insts or stubs). 103 void AbstractAssembler::end_a_const(CodeSection* cs) { 104 assert(_code_section == code()->consts(), "not in consts?"); 105 set_code_section(cs); 106 } 107 108 void AbstractAssembler::flush() { 109 ICache::invalidate_range(addr_at(0), offset()); 110 } 111 112 void AbstractAssembler::bind(Label& L) { 113 if (L.is_bound()) { 114 // Assembler can bind a label more than once to the same place. 115 guarantee(L.loc() == locator(), "attempt to redefine label"); 116 return; 117 } 118 L.bind_loc(locator()); 119 L.patch_instructions((MacroAssembler*)this); 120 } 121 122 void AbstractAssembler::generate_stack_overflow_check(int frame_size_in_bytes) { 123 if (UseStackBanging) { 124 // Each code entry causes one stack bang n pages down the stack where n 125 // is configurable by StackShadowPages. The setting depends on the maximum 126 // depth of VM call stack or native before going back into java code, 127 // since only java code can raise a stack overflow exception using the 128 // stack banging mechanism. The VM and native code does not detect stack 129 // overflow. 130 // The code in JavaCalls::call() checks that there is at least n pages 131 // available, so all entry code needs to do is bang once for the end of 132 // this shadow zone. 133 // The entry code may need to bang additional pages if the framesize 134 // is greater than a page. 135 136 const int page_size = os::vm_page_size(); 137 int bang_end = StackShadowPages * page_size; 138 139 // This is how far the previous frame's stack banging extended. 140 const int bang_end_safe = bang_end; 141 142 if (frame_size_in_bytes > page_size) { 143 bang_end += frame_size_in_bytes; 144 } 145 146 int bang_offset = bang_end_safe; 147 while (bang_offset <= bang_end) { 148 // Need at least one stack bang at end of shadow zone. 149 bang_stack_with_offset(bang_offset); 150 bang_offset += page_size; 151 } 152 } // end (UseStackBanging) 153 } 154 155 void Label::add_patch_at(CodeBuffer* cb, int branch_loc) { 156 assert(_loc == -1, "Label is unbound"); 157 if (_patch_index < PatchCacheSize) { 158 _patches[_patch_index] = branch_loc; 159 } else { 160 if (_patch_overflow == NULL) { 161 _patch_overflow = cb->create_patch_overflow(); 162 } 163 _patch_overflow->push(branch_loc); 164 } 165 ++_patch_index; 166 } 167 168 void Label::patch_instructions(MacroAssembler* masm) { 169 assert(is_bound(), "Label is bound"); 170 CodeBuffer* cb = masm->code(); 171 int target_sect = CodeBuffer::locator_sect(loc()); 172 address target = cb->locator_address(loc()); 173 while (_patch_index > 0) { 174 --_patch_index; 175 int branch_loc; 176 if (_patch_index >= PatchCacheSize) { 177 branch_loc = _patch_overflow->pop(); 178 } else { 179 branch_loc = _patches[_patch_index]; 180 } 181 int branch_sect = CodeBuffer::locator_sect(branch_loc); 182 address branch = cb->locator_address(branch_loc); 183 if (branch_sect == CodeBuffer::SECT_CONSTS) { 184 // The thing to patch is a constant word. 185 *(address*)branch = target; 186 continue; 187 } 188 189 #ifdef ASSERT 190 // Cross-section branches only work if the 191 // intermediate section boundaries are frozen. 192 if (target_sect != branch_sect) { 193 for (int n = MIN2(target_sect, branch_sect), 194 nlimit = (target_sect + branch_sect) - n; 195 n < nlimit; n++) { 196 CodeSection* cs = cb->code_section(n); 197 assert(cs->is_frozen(), "cross-section branch needs stable offsets"); 198 } 199 } 200 #endif //ASSERT 201 202 // Push the target offset into the branch instruction. 203 masm->pd_patch_instruction(branch, target); 204 } 205 } 206 207 struct DelayedConstant { 208 typedef void (*value_fn_t)(); 209 BasicType type; 210 intptr_t value; 211 value_fn_t value_fn; 212 // This limit of 20 is generous for initial uses. 213 // The limit needs to be large enough to store the field offsets 214 // into classes which do not have statically fixed layouts. 215 // (Initial use is for method handle object offsets.) 216 // Look for uses of "delayed_value" in the source code 217 // and make sure this number is generous enough to handle all of them. 218 enum { DC_LIMIT = 20 }; 219 static DelayedConstant delayed_constants[DC_LIMIT]; 220 static DelayedConstant* add(BasicType type, value_fn_t value_fn); 221 bool match(BasicType t, value_fn_t cfn) { 222 return type == t && value_fn == cfn; 223 } 224 static void update_all(); 225 }; 226 227 DelayedConstant DelayedConstant::delayed_constants[DC_LIMIT]; 228 // Default C structure initialization rules have the following effect here: 229 // = { { (BasicType)0, (intptr_t)NULL }, ... }; 230 231 DelayedConstant* DelayedConstant::add(BasicType type, 232 DelayedConstant::value_fn_t cfn) { 233 for (int i = 0; i < DC_LIMIT; i++) { 234 DelayedConstant* dcon = &delayed_constants[i]; 235 if (dcon->match(type, cfn)) 236 return dcon; 237 if (dcon->value_fn == NULL) { 238 // (cmpxchg not because this is multi-threaded but because I'm paranoid) 239 if (Atomic::cmpxchg_ptr(CAST_FROM_FN_PTR(void*, cfn), &dcon->value_fn, NULL) == NULL) { 240 dcon->type = type; 241 return dcon; 242 } 243 } 244 } 245 // If this assert is hit (in pre-integration testing!) then re-evaluate 246 // the comment on the definition of DC_LIMIT. 247 guarantee(false, "too many delayed constants"); 248 return NULL; 249 } 250 251 void DelayedConstant::update_all() { 252 for (int i = 0; i < DC_LIMIT; i++) { 253 DelayedConstant* dcon = &delayed_constants[i]; 254 if (dcon->value_fn != NULL && dcon->value == 0) { 255 typedef int (*int_fn_t)(); 256 typedef address (*address_fn_t)(); 257 switch (dcon->type) { 258 case T_INT: dcon->value = (intptr_t) ((int_fn_t) dcon->value_fn)(); break; 259 case T_ADDRESS: dcon->value = (intptr_t) ((address_fn_t)dcon->value_fn)(); break; 260 } 261 } 262 } 263 } 264 265 RegisterOrConstant AbstractAssembler::delayed_value(int(*value_fn)(), Register tmp, int offset) { 266 intptr_t val = (intptr_t) (*value_fn)(); 267 if (val != 0) return val + offset; 268 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset); 269 } 270 RegisterOrConstant AbstractAssembler::delayed_value(address(*value_fn)(), Register tmp, int offset) { 271 intptr_t val = (intptr_t) (*value_fn)(); 272 if (val != 0) return val + offset; 273 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset); 274 } 275 intptr_t* AbstractAssembler::delayed_value_addr(int(*value_fn)()) { 276 DelayedConstant* dcon = DelayedConstant::add(T_INT, (DelayedConstant::value_fn_t) value_fn); 277 return &dcon->value; 278 } 279 intptr_t* AbstractAssembler::delayed_value_addr(address(*value_fn)()) { 280 DelayedConstant* dcon = DelayedConstant::add(T_ADDRESS, (DelayedConstant::value_fn_t) value_fn); 281 return &dcon->value; 282 } 283 void AbstractAssembler::update_delayed_values() { 284 DelayedConstant::update_all(); 285 } 286 287 void AbstractAssembler::block_comment(const char* comment) { 288 if (sect() == CodeBuffer::SECT_INSTS) { 289 code_section()->outer()->block_comment(offset(), comment); 290 } 291 } 292 293 const char* AbstractAssembler::code_string(const char* str) { 294 if (sect() == CodeBuffer::SECT_INSTS || sect() == CodeBuffer::SECT_STUBS) { 295 return code_section()->outer()->code_string(str); 296 } 297 return NULL; 298 } 299 300 bool MacroAssembler::needs_explicit_null_check(intptr_t offset) { 301 // Exception handler checks the nmethod's implicit null checks table 302 // only when this method returns false. 303 #ifdef _LP64 304 if (UseCompressedOops && Universe::narrow_oop_base() != NULL) { 305 assert (Universe::heap() != NULL, "java heap should be initialized"); 306 // The first page after heap_base is unmapped and 307 // the 'offset' is equal to [heap_base + offset] for 308 // narrow oop implicit null checks. 309 uintptr_t base = (uintptr_t)Universe::narrow_oop_base(); 310 if ((uintptr_t)offset >= base) { 311 // Normalize offset for the next check. 312 offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1)); 313 } 314 } 315 #endif 316 return offset < 0 || os::vm_page_size() <= offset; 317 }