Tue, 18 Dec 2012 10:47:23 -0800
8004536: replace AbstractAssembler emit_word with emit_int16
Reviewed-by: jrose, kvn, twisti
Contributed-by: Morris Meyer <morris.meyer@oracle.com>
1 /*
2 * Copyright (c) 1997, 2012, 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 */
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"
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.
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, 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 }
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 }
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 }
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 }
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 }
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 }
108 void AbstractAssembler::flush() {
109 ICache::invalidate_range(addr_at(0), offset());
110 }
113 void AbstractAssembler::a_byte(int x) {
114 emit_byte(x);
115 }
118 void AbstractAssembler::a_long(jint x) {
119 emit_long(x);
120 }
123 void AbstractAssembler::bind(Label& L) {
124 if (L.is_bound()) {
125 // Assembler can bind a label more than once to the same place.
126 guarantee(L.loc() == locator(), "attempt to redefine label");
127 return;
128 }
129 L.bind_loc(locator());
130 L.patch_instructions((MacroAssembler*)this);
131 }
133 void AbstractAssembler::generate_stack_overflow_check( int frame_size_in_bytes) {
134 if (UseStackBanging) {
135 // Each code entry causes one stack bang n pages down the stack where n
136 // is configurable by StackBangPages. The setting depends on the maximum
137 // depth of VM call stack or native before going back into java code,
138 // since only java code can raise a stack overflow exception using the
139 // stack banging mechanism. The VM and native code does not detect stack
140 // overflow.
141 // The code in JavaCalls::call() checks that there is at least n pages
142 // available, so all entry code needs to do is bang once for the end of
143 // this shadow zone.
144 // The entry code may need to bang additional pages if the framesize
145 // is greater than a page.
147 const int page_size = os::vm_page_size();
148 int bang_end = StackShadowPages*page_size;
150 // This is how far the previous frame's stack banging extended.
151 const int bang_end_safe = bang_end;
153 if (frame_size_in_bytes > page_size) {
154 bang_end += frame_size_in_bytes;
155 }
157 int bang_offset = bang_end_safe;
158 while (bang_offset <= bang_end) {
159 // Need at least one stack bang at end of shadow zone.
160 bang_stack_with_offset(bang_offset);
161 bang_offset += page_size;
162 }
163 } // end (UseStackBanging)
164 }
166 void Label::add_patch_at(CodeBuffer* cb, int branch_loc) {
167 assert(_loc == -1, "Label is unbound");
168 if (_patch_index < PatchCacheSize) {
169 _patches[_patch_index] = branch_loc;
170 } else {
171 if (_patch_overflow == NULL) {
172 _patch_overflow = cb->create_patch_overflow();
173 }
174 _patch_overflow->push(branch_loc);
175 }
176 ++_patch_index;
177 }
179 void Label::patch_instructions(MacroAssembler* masm) {
180 assert(is_bound(), "Label is bound");
181 CodeBuffer* cb = masm->code();
182 int target_sect = CodeBuffer::locator_sect(loc());
183 address target = cb->locator_address(loc());
184 while (_patch_index > 0) {
185 --_patch_index;
186 int branch_loc;
187 if (_patch_index >= PatchCacheSize) {
188 branch_loc = _patch_overflow->pop();
189 } else {
190 branch_loc = _patches[_patch_index];
191 }
192 int branch_sect = CodeBuffer::locator_sect(branch_loc);
193 address branch = cb->locator_address(branch_loc);
194 if (branch_sect == CodeBuffer::SECT_CONSTS) {
195 // The thing to patch is a constant word.
196 *(address*)branch = target;
197 continue;
198 }
200 #ifdef ASSERT
201 // Cross-section branches only work if the
202 // intermediate section boundaries are frozen.
203 if (target_sect != branch_sect) {
204 for (int n = MIN2(target_sect, branch_sect),
205 nlimit = (target_sect + branch_sect) - n;
206 n < nlimit; n++) {
207 CodeSection* cs = cb->code_section(n);
208 assert(cs->is_frozen(), "cross-section branch needs stable offsets");
209 }
210 }
211 #endif //ASSERT
213 // Push the target offset into the branch instruction.
214 masm->pd_patch_instruction(branch, target);
215 }
216 }
218 struct DelayedConstant {
219 typedef void (*value_fn_t)();
220 BasicType type;
221 intptr_t value;
222 value_fn_t value_fn;
223 // This limit of 20 is generous for initial uses.
224 // The limit needs to be large enough to store the field offsets
225 // into classes which do not have statically fixed layouts.
226 // (Initial use is for method handle object offsets.)
227 // Look for uses of "delayed_value" in the source code
228 // and make sure this number is generous enough to handle all of them.
229 enum { DC_LIMIT = 20 };
230 static DelayedConstant delayed_constants[DC_LIMIT];
231 static DelayedConstant* add(BasicType type, value_fn_t value_fn);
232 bool match(BasicType t, value_fn_t cfn) {
233 return type == t && value_fn == cfn;
234 }
235 static void update_all();
236 };
238 DelayedConstant DelayedConstant::delayed_constants[DC_LIMIT];
239 // Default C structure initialization rules have the following effect here:
240 // = { { (BasicType)0, (intptr_t)NULL }, ... };
242 DelayedConstant* DelayedConstant::add(BasicType type,
243 DelayedConstant::value_fn_t cfn) {
244 for (int i = 0; i < DC_LIMIT; i++) {
245 DelayedConstant* dcon = &delayed_constants[i];
246 if (dcon->match(type, cfn))
247 return dcon;
248 if (dcon->value_fn == NULL) {
249 // (cmpxchg not because this is multi-threaded but because I'm paranoid)
250 if (Atomic::cmpxchg_ptr(CAST_FROM_FN_PTR(void*, cfn), &dcon->value_fn, NULL) == NULL) {
251 dcon->type = type;
252 return dcon;
253 }
254 }
255 }
256 // If this assert is hit (in pre-integration testing!) then re-evaluate
257 // the comment on the definition of DC_LIMIT.
258 guarantee(false, "too many delayed constants");
259 return NULL;
260 }
262 void DelayedConstant::update_all() {
263 for (int i = 0; i < DC_LIMIT; i++) {
264 DelayedConstant* dcon = &delayed_constants[i];
265 if (dcon->value_fn != NULL && dcon->value == 0) {
266 typedef int (*int_fn_t)();
267 typedef address (*address_fn_t)();
268 switch (dcon->type) {
269 case T_INT: dcon->value = (intptr_t) ((int_fn_t) dcon->value_fn)(); break;
270 case T_ADDRESS: dcon->value = (intptr_t) ((address_fn_t)dcon->value_fn)(); break;
271 }
272 }
273 }
274 }
276 RegisterOrConstant AbstractAssembler::delayed_value(int(*value_fn)(), Register tmp, int offset) {
277 intptr_t val = (intptr_t) (*value_fn)();
278 if (val != 0) return val + offset;
279 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
280 }
281 RegisterOrConstant AbstractAssembler::delayed_value(address(*value_fn)(), Register tmp, int offset) {
282 intptr_t val = (intptr_t) (*value_fn)();
283 if (val != 0) return val + offset;
284 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
285 }
286 intptr_t* AbstractAssembler::delayed_value_addr(int(*value_fn)()) {
287 DelayedConstant* dcon = DelayedConstant::add(T_INT, (DelayedConstant::value_fn_t) value_fn);
288 return &dcon->value;
289 }
290 intptr_t* AbstractAssembler::delayed_value_addr(address(*value_fn)()) {
291 DelayedConstant* dcon = DelayedConstant::add(T_ADDRESS, (DelayedConstant::value_fn_t) value_fn);
292 return &dcon->value;
293 }
294 void AbstractAssembler::update_delayed_values() {
295 DelayedConstant::update_all();
296 }
301 void AbstractAssembler::block_comment(const char* comment) {
302 if (sect() == CodeBuffer::SECT_INSTS) {
303 code_section()->outer()->block_comment(offset(), comment);
304 }
305 }
307 bool MacroAssembler::needs_explicit_null_check(intptr_t offset) {
308 // Exception handler checks the nmethod's implicit null checks table
309 // only when this method returns false.
310 #ifdef _LP64
311 if (UseCompressedOops && Universe::narrow_oop_base() != NULL) {
312 assert (Universe::heap() != NULL, "java heap should be initialized");
313 // The first page after heap_base is unmapped and
314 // the 'offset' is equal to [heap_base + offset] for
315 // narrow oop implicit null checks.
316 uintptr_t base = (uintptr_t)Universe::narrow_oop_base();
317 if ((uintptr_t)offset >= base) {
318 // Normalize offset for the next check.
319 offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1));
320 }
321 }
322 #endif
323 return offset < 0 || os::vm_page_size() <= offset;
324 }