Tue, 27 Nov 2012 17:41:38 -0800
8003850: add support for constants in stub code
Summary: remember the code section and switch back to the proper one when adding constants.
Reviewed-by: twisti, kvn
Contributed-by: goetz.lindenmaier@sap.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/assembler.hpp"
27 #include "asm/assembler.inline.hpp"
28 #include "asm/codeBuffer.hpp"
29 #include "runtime/icache.hpp"
30 #include "runtime/os.hpp"
31 #ifdef TARGET_ARCH_x86
32 # include "assembler_x86.inline.hpp"
33 #endif
34 #ifdef TARGET_ARCH_sparc
35 # include "assembler_sparc.inline.hpp"
36 #endif
37 #ifdef TARGET_ARCH_zero
38 # include "assembler_zero.inline.hpp"
39 #endif
40 #ifdef TARGET_ARCH_arm
41 # include "assembler_arm.inline.hpp"
42 #endif
43 #ifdef TARGET_ARCH_ppc
44 # include "assembler_ppc.inline.hpp"
45 #endif
48 // Implementation of AbstractAssembler
49 //
50 // The AbstractAssembler is generating code into a CodeBuffer. To make code generation faster,
51 // the assembler keeps a copy of the code buffers boundaries & modifies them when
52 // emitting bytes rather than using the code buffers accessor functions all the time.
53 // The code buffer is updated via set_code_end(...) after emitting a whole instruction.
55 AbstractAssembler::AbstractAssembler(CodeBuffer* code) {
56 if (code == NULL) return;
57 CodeSection* cs = code->insts();
58 cs->clear_mark(); // new assembler kills old mark
59 _code_section = cs;
60 _code_begin = cs->start();
61 _code_limit = cs->limit();
62 _code_pos = cs->end();
63 _oop_recorder= code->oop_recorder();
64 DEBUG_ONLY( _short_branch_delta = 0; )
65 if (_code_begin == NULL) {
66 vm_exit_out_of_memory(0, err_msg("CodeCache: no room for %s",
67 code->name()));
68 }
69 }
71 void AbstractAssembler::set_code_section(CodeSection* cs) {
72 assert(cs->outer() == code_section()->outer(), "sanity");
73 assert(cs->is_allocated(), "need to pre-allocate this section");
74 cs->clear_mark(); // new assembly into this section kills old mark
75 _code_section = cs;
76 _code_begin = cs->start();
77 _code_limit = cs->limit();
78 _code_pos = cs->end();
79 }
81 // Inform CodeBuffer that incoming code and relocation will be for stubs
82 address AbstractAssembler::start_a_stub(int required_space) {
83 CodeBuffer* cb = code();
84 CodeSection* cs = cb->stubs();
85 assert(_code_section == cb->insts(), "not in insts?");
86 sync();
87 if (cs->maybe_expand_to_ensure_remaining(required_space)
88 && cb->blob() == NULL) {
89 return NULL;
90 }
91 set_code_section(cs);
92 return pc();
93 }
95 // Inform CodeBuffer that incoming code and relocation will be code
96 // Should not be called if start_a_stub() returned NULL
97 void AbstractAssembler::end_a_stub() {
98 assert(_code_section == code()->stubs(), "not in stubs?");
99 sync();
100 set_code_section(code()->insts());
101 }
103 // Inform CodeBuffer that incoming code and relocation will be for stubs
104 address AbstractAssembler::start_a_const(int required_space, int required_align) {
105 CodeBuffer* cb = code();
106 CodeSection* cs = cb->consts();
107 assert(_code_section == cb->insts() || _code_section == cb->stubs(), "not in insts/stubs?");
108 sync();
109 address end = cs->end();
110 int pad = -(intptr_t)end & (required_align-1);
111 if (cs->maybe_expand_to_ensure_remaining(pad + required_space)) {
112 if (cb->blob() == NULL) return NULL;
113 end = cs->end(); // refresh pointer
114 }
115 if (pad > 0) {
116 while (--pad >= 0) { *end++ = 0; }
117 cs->set_end(end);
118 }
119 set_code_section(cs);
120 return end;
121 }
123 // Inform CodeBuffer that incoming code and relocation will be code
124 // in section cs (insts or stubs).
125 void AbstractAssembler::end_a_const(CodeSection* cs) {
126 assert(_code_section == code()->consts(), "not in consts?");
127 sync();
128 set_code_section(cs);
129 }
131 void AbstractAssembler::flush() {
132 sync();
133 ICache::invalidate_range(addr_at(0), offset());
134 }
137 void AbstractAssembler::a_byte(int x) {
138 emit_byte(x);
139 }
142 void AbstractAssembler::a_long(jint x) {
143 emit_long(x);
144 }
146 // Labels refer to positions in the (to be) generated code. There are bound
147 // and unbound
148 //
149 // Bound labels refer to known positions in the already generated code.
150 // offset() is the position the label refers to.
151 //
152 // Unbound labels refer to unknown positions in the code to be generated; it
153 // may contain a list of unresolved displacements that refer to it
154 #ifndef PRODUCT
155 void AbstractAssembler::print(Label& L) {
156 if (L.is_bound()) {
157 tty->print_cr("bound label to %d|%d", L.loc_pos(), L.loc_sect());
158 } else if (L.is_unbound()) {
159 L.print_instructions((MacroAssembler*)this);
160 } else {
161 tty->print_cr("label in inconsistent state (loc = %d)", L.loc());
162 }
163 }
164 #endif // PRODUCT
167 void AbstractAssembler::bind(Label& L) {
168 if (L.is_bound()) {
169 // Assembler can bind a label more than once to the same place.
170 guarantee(L.loc() == locator(), "attempt to redefine label");
171 return;
172 }
173 L.bind_loc(locator());
174 L.patch_instructions((MacroAssembler*)this);
175 }
177 void AbstractAssembler::generate_stack_overflow_check( int frame_size_in_bytes) {
178 if (UseStackBanging) {
179 // Each code entry causes one stack bang n pages down the stack where n
180 // is configurable by StackBangPages. The setting depends on the maximum
181 // depth of VM call stack or native before going back into java code,
182 // since only java code can raise a stack overflow exception using the
183 // stack banging mechanism. The VM and native code does not detect stack
184 // overflow.
185 // The code in JavaCalls::call() checks that there is at least n pages
186 // available, so all entry code needs to do is bang once for the end of
187 // this shadow zone.
188 // The entry code may need to bang additional pages if the framesize
189 // is greater than a page.
191 const int page_size = os::vm_page_size();
192 int bang_end = StackShadowPages*page_size;
194 // This is how far the previous frame's stack banging extended.
195 const int bang_end_safe = bang_end;
197 if (frame_size_in_bytes > page_size) {
198 bang_end += frame_size_in_bytes;
199 }
201 int bang_offset = bang_end_safe;
202 while (bang_offset <= bang_end) {
203 // Need at least one stack bang at end of shadow zone.
204 bang_stack_with_offset(bang_offset);
205 bang_offset += page_size;
206 }
207 } // end (UseStackBanging)
208 }
210 void Label::add_patch_at(CodeBuffer* cb, int branch_loc) {
211 assert(_loc == -1, "Label is unbound");
212 if (_patch_index < PatchCacheSize) {
213 _patches[_patch_index] = branch_loc;
214 } else {
215 if (_patch_overflow == NULL) {
216 _patch_overflow = cb->create_patch_overflow();
217 }
218 _patch_overflow->push(branch_loc);
219 }
220 ++_patch_index;
221 }
223 void Label::patch_instructions(MacroAssembler* masm) {
224 assert(is_bound(), "Label is bound");
225 CodeBuffer* cb = masm->code();
226 int target_sect = CodeBuffer::locator_sect(loc());
227 address target = cb->locator_address(loc());
228 while (_patch_index > 0) {
229 --_patch_index;
230 int branch_loc;
231 if (_patch_index >= PatchCacheSize) {
232 branch_loc = _patch_overflow->pop();
233 } else {
234 branch_loc = _patches[_patch_index];
235 }
236 int branch_sect = CodeBuffer::locator_sect(branch_loc);
237 address branch = cb->locator_address(branch_loc);
238 if (branch_sect == CodeBuffer::SECT_CONSTS) {
239 // The thing to patch is a constant word.
240 *(address*)branch = target;
241 continue;
242 }
244 #ifdef ASSERT
245 // Cross-section branches only work if the
246 // intermediate section boundaries are frozen.
247 if (target_sect != branch_sect) {
248 for (int n = MIN2(target_sect, branch_sect),
249 nlimit = (target_sect + branch_sect) - n;
250 n < nlimit; n++) {
251 CodeSection* cs = cb->code_section(n);
252 assert(cs->is_frozen(), "cross-section branch needs stable offsets");
253 }
254 }
255 #endif //ASSERT
257 // Push the target offset into the branch instruction.
258 masm->pd_patch_instruction(branch, target);
259 }
260 }
262 struct DelayedConstant {
263 typedef void (*value_fn_t)();
264 BasicType type;
265 intptr_t value;
266 value_fn_t value_fn;
267 // This limit of 20 is generous for initial uses.
268 // The limit needs to be large enough to store the field offsets
269 // into classes which do not have statically fixed layouts.
270 // (Initial use is for method handle object offsets.)
271 // Look for uses of "delayed_value" in the source code
272 // and make sure this number is generous enough to handle all of them.
273 enum { DC_LIMIT = 20 };
274 static DelayedConstant delayed_constants[DC_LIMIT];
275 static DelayedConstant* add(BasicType type, value_fn_t value_fn);
276 bool match(BasicType t, value_fn_t cfn) {
277 return type == t && value_fn == cfn;
278 }
279 static void update_all();
280 };
282 DelayedConstant DelayedConstant::delayed_constants[DC_LIMIT];
283 // Default C structure initialization rules have the following effect here:
284 // = { { (BasicType)0, (intptr_t)NULL }, ... };
286 DelayedConstant* DelayedConstant::add(BasicType type,
287 DelayedConstant::value_fn_t cfn) {
288 for (int i = 0; i < DC_LIMIT; i++) {
289 DelayedConstant* dcon = &delayed_constants[i];
290 if (dcon->match(type, cfn))
291 return dcon;
292 if (dcon->value_fn == NULL) {
293 // (cmpxchg not because this is multi-threaded but because I'm paranoid)
294 if (Atomic::cmpxchg_ptr(CAST_FROM_FN_PTR(void*, cfn), &dcon->value_fn, NULL) == NULL) {
295 dcon->type = type;
296 return dcon;
297 }
298 }
299 }
300 // If this assert is hit (in pre-integration testing!) then re-evaluate
301 // the comment on the definition of DC_LIMIT.
302 guarantee(false, "too many delayed constants");
303 return NULL;
304 }
306 void DelayedConstant::update_all() {
307 for (int i = 0; i < DC_LIMIT; i++) {
308 DelayedConstant* dcon = &delayed_constants[i];
309 if (dcon->value_fn != NULL && dcon->value == 0) {
310 typedef int (*int_fn_t)();
311 typedef address (*address_fn_t)();
312 switch (dcon->type) {
313 case T_INT: dcon->value = (intptr_t) ((int_fn_t) dcon->value_fn)(); break;
314 case T_ADDRESS: dcon->value = (intptr_t) ((address_fn_t)dcon->value_fn)(); break;
315 }
316 }
317 }
318 }
320 RegisterOrConstant AbstractAssembler::delayed_value(int(*value_fn)(), Register tmp, int offset) {
321 intptr_t val = (intptr_t) (*value_fn)();
322 if (val != 0) return val + offset;
323 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
324 }
325 RegisterOrConstant AbstractAssembler::delayed_value(address(*value_fn)(), Register tmp, int offset) {
326 intptr_t val = (intptr_t) (*value_fn)();
327 if (val != 0) return val + offset;
328 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
329 }
330 intptr_t* AbstractAssembler::delayed_value_addr(int(*value_fn)()) {
331 DelayedConstant* dcon = DelayedConstant::add(T_INT, (DelayedConstant::value_fn_t) value_fn);
332 return &dcon->value;
333 }
334 intptr_t* AbstractAssembler::delayed_value_addr(address(*value_fn)()) {
335 DelayedConstant* dcon = DelayedConstant::add(T_ADDRESS, (DelayedConstant::value_fn_t) value_fn);
336 return &dcon->value;
337 }
338 void AbstractAssembler::update_delayed_values() {
339 DelayedConstant::update_all();
340 }
345 void AbstractAssembler::block_comment(const char* comment) {
346 if (sect() == CodeBuffer::SECT_INSTS) {
347 code_section()->outer()->block_comment(offset(), comment);
348 }
349 }
351 bool MacroAssembler::needs_explicit_null_check(intptr_t offset) {
352 // Exception handler checks the nmethod's implicit null checks table
353 // only when this method returns false.
354 #ifdef _LP64
355 if (UseCompressedOops && Universe::narrow_oop_base() != NULL) {
356 assert (Universe::heap() != NULL, "java heap should be initialized");
357 // The first page after heap_base is unmapped and
358 // the 'offset' is equal to [heap_base + offset] for
359 // narrow oop implicit null checks.
360 uintptr_t base = (uintptr_t)Universe::narrow_oop_base();
361 if ((uintptr_t)offset >= base) {
362 // Normalize offset for the next check.
363 offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1));
364 }
365 }
366 #endif
367 return offset < 0 || os::vm_page_size() <= offset;
368 }
370 #ifndef PRODUCT
371 void Label::print_instructions(MacroAssembler* masm) const {
372 CodeBuffer* cb = masm->code();
373 for (int i = 0; i < _patch_index; ++i) {
374 int branch_loc;
375 if (i >= PatchCacheSize) {
376 branch_loc = _patch_overflow->at(i - PatchCacheSize);
377 } else {
378 branch_loc = _patches[i];
379 }
380 int branch_pos = CodeBuffer::locator_pos(branch_loc);
381 int branch_sect = CodeBuffer::locator_sect(branch_loc);
382 address branch = cb->locator_address(branch_loc);
383 tty->print_cr("unbound label");
384 tty->print("@ %d|%d ", branch_pos, branch_sect);
385 if (branch_sect == CodeBuffer::SECT_CONSTS) {
386 tty->print_cr(PTR_FORMAT, *(address*)branch);
387 continue;
388 }
389 masm->pd_print_patched_instruction(branch);
390 tty->cr();
391 }
392 }
393 #endif // ndef PRODUCT