1 /* 2 * Copyright (c) 1997, 2016, 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/codeBuffer.hpp" 27 #include "asm/macroAssembler.hpp" 28 #include "asm/macroAssembler.inline.hpp" 29 #include "runtime/atomic.hpp" 30 #include "runtime/icache.hpp" 31 #include "runtime/os.hpp" 32 #include "runtime/thread.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, "CodeCache: no room for %s", code->name()); 48 } 49 _code_section = cs; 50 _oop_recorder= code->oop_recorder(); 51 DEBUG_ONLY( _short_branch_delta = 0; ) 52 } 53 54 void AbstractAssembler::set_code_section(CodeSection* cs) { 55 assert(cs->outer() == code_section()->outer(), "sanity"); 56 assert(cs->is_allocated(), "need to pre-allocate this section"); 57 cs->clear_mark(); // new assembly into this section kills old mark 58 _code_section = cs; 59 } 60 61 // Inform CodeBuffer that incoming code and relocation will be for stubs 62 address AbstractAssembler::start_a_stub(int required_space) { 63 CodeBuffer* cb = code(); 64 CodeSection* cs = cb->stubs(); 65 assert(_code_section == cb->insts(), "not in insts?"); 66 if (cs->maybe_expand_to_ensure_remaining(required_space) 67 && cb->blob() == NULL) { 68 return NULL; 69 } 70 set_code_section(cs); 71 return pc(); 72 } 73 74 // Inform CodeBuffer that incoming code and relocation will be code 75 // Should not be called if start_a_stub() returned NULL 76 void AbstractAssembler::end_a_stub() { 77 assert(_code_section == code()->stubs(), "not in stubs?"); 78 set_code_section(code()->insts()); 79 } 80 81 // Inform CodeBuffer that incoming code and relocation will be for stubs 82 address AbstractAssembler::start_a_const(int required_space, int required_align) { 83 CodeBuffer* cb = code(); 84 CodeSection* cs = cb->consts(); 85 assert(_code_section == cb->insts() || _code_section == cb->stubs(), "not in insts/stubs?"); 86 address end = cs->end(); 87 int pad = -(intptr_t)end & (required_align-1); 88 if (cs->maybe_expand_to_ensure_remaining(pad + required_space)) { 89 if (cb->blob() == NULL) return NULL; 90 end = cs->end(); // refresh pointer 91 } 92 if (pad > 0) { 93 while (--pad >= 0) { *end++ = 0; } 94 cs->set_end(end); 95 } 96 set_code_section(cs); 97 return end; 98 } 99 100 // Inform CodeBuffer that incoming code and relocation will be code 101 // in section cs (insts or stubs). 102 void AbstractAssembler::end_a_const(CodeSection* cs) { 103 assert(_code_section == code()->consts(), "not in consts?"); 104 set_code_section(cs); 105 } 106 107 void AbstractAssembler::flush() { 108 ICache::invalidate_range(addr_at(0), offset()); 109 } 110 111 void AbstractAssembler::bind(Label& L) { 112 if (L.is_bound()) { 113 // Assembler can bind a label more than once to the same place. 114 guarantee(L.loc() == locator(), "attempt to redefine label"); 115 return; 116 } 117 L.bind_loc(locator()); 118 L.patch_instructions((MacroAssembler*)this); 119 } 120 121 void AbstractAssembler::generate_stack_overflow_check(int frame_size_in_bytes) { 122 if (UseStackBanging) { 123 // Each code entry causes one stack bang n pages down the stack where n 124 // is configurable by StackShadowPages. The setting depends on the maximum 125 // depth of VM call stack or native before going back into java code, 126 // since only java code can raise a stack overflow exception using the 127 // stack banging mechanism. The VM and native code does not detect stack 128 // overflow. 129 // The code in JavaCalls::call() checks that there is at least n pages 130 // available, so all entry code needs to do is bang once for the end of 131 // this shadow zone. 132 // The entry code may need to bang additional pages if the framesize 133 // is greater than a page. 134 135 const int page_size = os::vm_page_size(); 136 int bang_end = (int)JavaThread::stack_shadow_zone_size(); 137 138 // This is how far the previous frame's stack banging extended. 139 const int bang_end_safe = bang_end; 140 141 if (frame_size_in_bytes > page_size) { 142 bang_end += frame_size_in_bytes; 143 } 144 145 int bang_offset = bang_end_safe; 146 while (bang_offset <= bang_end) { 147 // Need at least one stack bang at end of shadow zone. 148 bang_stack_with_offset(bang_offset); 149 bang_offset += page_size; 150 } 151 } // end (UseStackBanging) 152 } 153 154 void Label::add_patch_at(CodeBuffer* cb, int branch_loc) { 155 assert(_loc == -1, "Label is unbound"); 156 // Don't add patch locations during scratch emit. 157 if (cb->insts()->scratch_emit()) { return; } 158 if (_patch_index < PatchCacheSize) { 159 _patches[_patch_index] = branch_loc; 160 } else { 161 if (_patch_overflow == NULL) { 162 _patch_overflow = cb->create_patch_overflow(); 163 } 164 _patch_overflow->push(branch_loc); 165 } 166 ++_patch_index; 167 } 168 169 void Label::patch_instructions(MacroAssembler* masm) { 170 assert(is_bound(), "Label is bound"); 171 CodeBuffer* cb = masm->code(); 172 int target_sect = CodeBuffer::locator_sect(loc()); 173 address target = cb->locator_address(loc()); 174 while (_patch_index > 0) { 175 --_patch_index; 176 int branch_loc; 177 if (_patch_index >= PatchCacheSize) { 178 branch_loc = _patch_overflow->pop(); 179 } else { 180 branch_loc = _patches[_patch_index]; 181 } 182 int branch_sect = CodeBuffer::locator_sect(branch_loc); 183 address branch = cb->locator_address(branch_loc); 184 if (branch_sect == CodeBuffer::SECT_CONSTS) { 185 // The thing to patch is a constant word. 186 *(address*)branch = target; 187 continue; 188 } 189 190 #ifdef ASSERT 191 // Cross-section branches only work if the 192 // intermediate section boundaries are frozen. 193 if (target_sect != branch_sect) { 194 for (int n = MIN2(target_sect, branch_sect), 195 nlimit = (target_sect + branch_sect) - n; 196 n < nlimit; n++) { 197 CodeSection* cs = cb->code_section(n); 198 assert(cs->is_frozen(), "cross-section branch needs stable offsets"); 199 } 200 } 201 #endif //ASSERT 202 203 // Push the target offset into the branch instruction. 204 masm->pd_patch_instruction(branch, target); 205 } 206 } 207 208 struct DelayedConstant { 209 typedef void (*value_fn_t)(); 210 BasicType type; 211 intptr_t value; 212 value_fn_t value_fn; 213 // This limit of 20 is generous for initial uses. 214 // The limit needs to be large enough to store the field offsets 215 // into classes which do not have statically fixed layouts. 216 // (Initial use is for method handle object offsets.) 217 // Look for uses of "delayed_value" in the source code 218 // and make sure this number is generous enough to handle all of them. 219 enum { DC_LIMIT = 20 }; 220 static DelayedConstant delayed_constants[DC_LIMIT]; 221 static DelayedConstant* add(BasicType type, value_fn_t value_fn); 222 bool match(BasicType t, value_fn_t cfn) { 223 return type == t && value_fn == cfn; 224 } 225 static void update_all(); 226 }; 227 228 DelayedConstant DelayedConstant::delayed_constants[DC_LIMIT]; 229 // Default C structure initialization rules have the following effect here: 230 // = { { (BasicType)0, (intptr_t)NULL }, ... }; 231 232 DelayedConstant* DelayedConstant::add(BasicType type, 233 DelayedConstant::value_fn_t cfn) { 234 for (int i = 0; i < DC_LIMIT; i++) { 235 DelayedConstant* dcon = &delayed_constants[i]; 236 if (dcon->match(type, cfn)) 237 return dcon; 238 if (dcon->value_fn == NULL) { 239 // (cmpxchg not because this is multi-threaded but because I'm paranoid) 240 if (Atomic::cmpxchg_ptr(CAST_FROM_FN_PTR(void*, cfn), &dcon->value_fn, NULL) == NULL) { 241 dcon->type = type; 242 return dcon; 243 } 244 } 245 } 246 // If this assert is hit (in pre-integration testing!) then re-evaluate 247 // the comment on the definition of DC_LIMIT. 248 guarantee(false, "too many delayed constants"); 249 return NULL; 250 } 251 252 void DelayedConstant::update_all() { 253 for (int i = 0; i < DC_LIMIT; i++) { 254 DelayedConstant* dcon = &delayed_constants[i]; 255 if (dcon->value_fn != NULL && dcon->value == 0) { 256 typedef int (*int_fn_t)(); 257 typedef address (*address_fn_t)(); 258 switch (dcon->type) { 259 case T_INT: dcon->value = (intptr_t) ((int_fn_t) dcon->value_fn)(); break; 260 case T_ADDRESS: dcon->value = (intptr_t) ((address_fn_t)dcon->value_fn)(); break; 261 } 262 } 263 } 264 } 265 266 RegisterOrConstant AbstractAssembler::delayed_value(int(*value_fn)(), Register tmp, int offset) { 267 intptr_t val = (intptr_t) (*value_fn)(); 268 if (val != 0) return val + offset; 269 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset); 270 } 271 RegisterOrConstant AbstractAssembler::delayed_value(address(*value_fn)(), Register tmp, int offset) { 272 intptr_t val = (intptr_t) (*value_fn)(); 273 if (val != 0) return val + offset; 274 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset); 275 } 276 intptr_t* AbstractAssembler::delayed_value_addr(int(*value_fn)()) { 277 DelayedConstant* dcon = DelayedConstant::add(T_INT, (DelayedConstant::value_fn_t) value_fn); 278 return &dcon->value; 279 } 280 intptr_t* AbstractAssembler::delayed_value_addr(address(*value_fn)()) { 281 DelayedConstant* dcon = DelayedConstant::add(T_ADDRESS, (DelayedConstant::value_fn_t) value_fn); 282 return &dcon->value; 283 } 284 void AbstractAssembler::update_delayed_values() { 285 DelayedConstant::update_all(); 286 } 287 288 void AbstractAssembler::block_comment(const char* comment) { 289 if (sect() == CodeBuffer::SECT_INSTS) { 290 code_section()->outer()->block_comment(offset(), comment); 291 } 292 } 293 294 const char* AbstractAssembler::code_string(const char* str) { 295 if (sect() == CodeBuffer::SECT_INSTS || sect() == CodeBuffer::SECT_STUBS) { 296 return code_section()->outer()->code_string(str); 297 } 298 return NULL; 299 } 300 301 bool MacroAssembler::needs_explicit_null_check(intptr_t offset) { 302 // Exception handler checks the nmethod's implicit null checks table 303 // only when this method returns false. 304 #ifdef _LP64 305 if (UseCompressedOops && Universe::narrow_oop_base() != NULL) { 306 assert (Universe::heap() != NULL, "java heap should be initialized"); 307 // The first page after heap_base is unmapped and 308 // the 'offset' is equal to [heap_base + offset] for 309 // narrow oop implicit null checks. 310 uintptr_t base = (uintptr_t)Universe::narrow_oop_base(); 311 if ((uintptr_t)offset >= base) { 312 // Normalize offset for the next check. 313 offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1)); 314 } 315 } 316 #endif 317 return offset < 0 || os::vm_page_size() <= offset; 318 }