1 /*
2 * Copyright (c) 1997, 2017, 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 #include "utilities/macros.hpp"
34 #if INCLUDE_SHENANDOAHGC
35 #include "gc/shenandoah/shenandoahBarrierSetAssembler.hpp"
36 #endif
37
38
39 // Implementation of AbstractAssembler
40 //
41 // The AbstractAssembler is generating code into a CodeBuffer. To make code generation faster,
42 // the assembler keeps a copy of the code buffers boundaries & modifies them when
43 // emitting bytes rather than using the code buffers accessor functions all the time.
44 // The code buffer is updated via set_code_end(...) after emitting a whole instruction.
45
46 AbstractAssembler::AbstractAssembler(CodeBuffer* code) {
47 if (code == NULL) return;
48 CodeSection* cs = code->insts();
49 cs->clear_mark(); // new assembler kills old mark
50 if (cs->start() == NULL) {
51 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "CodeCache: no room for %s", code->name());
52 }
53 _code_section = cs;
54 _oop_recorder= code->oop_recorder();
55 DEBUG_ONLY( _short_branch_delta = 0; )
56 }
57
58 void AbstractAssembler::set_code_section(CodeSection* cs) {
59 assert(cs->outer() == code_section()->outer(), "sanity");
60 assert(cs->is_allocated(), "need to pre-allocate this section");
61 cs->clear_mark(); // new assembly into this section kills old mark
62 _code_section = cs;
63 }
64
65 // Inform CodeBuffer that incoming code and relocation will be for stubs
66 address AbstractAssembler::start_a_stub(int required_space) {
67 CodeBuffer* cb = code();
68 CodeSection* cs = cb->stubs();
69 assert(_code_section == cb->insts(), "not in insts?");
70 if (cs->maybe_expand_to_ensure_remaining(required_space)
71 && cb->blob() == NULL) {
72 return NULL;
73 }
74 set_code_section(cs);
75 return pc();
76 }
77
78 // Inform CodeBuffer that incoming code and relocation will be code
79 // Should not be called if start_a_stub() returned NULL
80 void AbstractAssembler::end_a_stub() {
81 assert(_code_section == code()->stubs(), "not in stubs?");
82 set_code_section(code()->insts());
83 }
84
85 // Inform CodeBuffer that incoming code and relocation will be for stubs
86 address AbstractAssembler::start_a_const(int required_space, int required_align) {
87 CodeBuffer* cb = code();
88 CodeSection* cs = cb->consts();
89 assert(_code_section == cb->insts() || _code_section == cb->stubs(), "not in insts/stubs?");
90 address end = cs->end();
91 int pad = -(intptr_t)end & (required_align-1);
92 if (cs->maybe_expand_to_ensure_remaining(pad + required_space)) {
93 if (cb->blob() == NULL) return NULL;
94 end = cs->end(); // refresh pointer
95 }
96 if (pad > 0) {
97 while (--pad >= 0) { *end++ = 0; }
98 cs->set_end(end);
99 }
100 set_code_section(cs);
101 return end;
102 }
103
104 // Inform CodeBuffer that incoming code and relocation will be code
105 // in section cs (insts or stubs).
106 void AbstractAssembler::end_a_const(CodeSection* cs) {
107 assert(_code_section == code()->consts(), "not in consts?");
108 set_code_section(cs);
109 }
110
111 void AbstractAssembler::flush() {
112 ICache::invalidate_range(addr_at(0), offset());
113 }
114
115 void AbstractAssembler::bind(Label& L) {
116 if (L.is_bound()) {
117 // Assembler can bind a label more than once to the same place.
118 guarantee(L.loc() == locator(), "attempt to redefine label");
119 return;
120 }
121 L.bind_loc(locator());
122 L.patch_instructions((MacroAssembler*)this);
123 }
124
125 void AbstractAssembler::generate_stack_overflow_check(int frame_size_in_bytes) {
126 if (UseStackBanging) {
127 // Each code entry causes one stack bang n pages down the stack where n
128 // is configurable by StackShadowPages. The setting depends on the maximum
129 // depth of VM call stack or native before going back into java code,
130 // since only java code can raise a stack overflow exception using the
131 // stack banging mechanism. The VM and native code does not detect stack
132 // overflow.
133 // The code in JavaCalls::call() checks that there is at least n pages
134 // available, so all entry code needs to do is bang once for the end of
135 // this shadow zone.
136 // The entry code may need to bang additional pages if the framesize
137 // is greater than a page.
138
139 const int page_size = os::vm_page_size();
140 int bang_end = (int)JavaThread::stack_shadow_zone_size();
141
142 // This is how far the previous frame's stack banging extended.
143 const int bang_end_safe = bang_end;
144
145 if (frame_size_in_bytes > page_size) {
146 bang_end += frame_size_in_bytes;
147 }
148
149 int bang_offset = bang_end_safe;
150 while (bang_offset <= bang_end) {
151 // Need at least one stack bang at end of shadow zone.
152 bang_stack_with_offset(bang_offset);
153 bang_offset += page_size;
154 }
155 } // end (UseStackBanging)
156 }
157
158 void Label::add_patch_at(CodeBuffer* cb, int branch_loc, const char* file, int line) {
159 assert(_loc == -1, "Label is unbound");
160 // Don't add patch locations during scratch emit.
161 if (cb->insts()->scratch_emit()) { return; }
162 if (_patch_index < PatchCacheSize) {
163 _patches[_patch_index] = branch_loc;
164 #ifdef ASSERT
165 _lines[_patch_index] = line;
166 _files[_patch_index] = file;
167 #endif
168 } else {
169 if (_patch_overflow == NULL) {
170 _patch_overflow = cb->create_patch_overflow();
171 }
172 _patch_overflow->push(branch_loc);
173 }
174 ++_patch_index;
175 }
176
177 void Label::patch_instructions(MacroAssembler* masm) {
178 assert(is_bound(), "Label is bound");
179 CodeBuffer* cb = masm->code();
180 int target_sect = CodeBuffer::locator_sect(loc());
181 address target = cb->locator_address(loc());
182 while (_patch_index > 0) {
183 --_patch_index;
184 int branch_loc;
185 int line = 0;
186 const char* file = NULL;
187 if (_patch_index >= PatchCacheSize) {
188 branch_loc = _patch_overflow->pop();
189 } else {
190 branch_loc = _patches[_patch_index];
191 #ifdef ASSERT
192 line = _lines[_patch_index];
193 file = _files[_patch_index];
194 #endif
195 }
196 int branch_sect = CodeBuffer::locator_sect(branch_loc);
197 address branch = cb->locator_address(branch_loc);
198 if (branch_sect == CodeBuffer::SECT_CONSTS) {
199 // The thing to patch is a constant word.
200 *(address*)branch = target;
201 continue;
202 }
203
204 #ifdef ASSERT
205 // Cross-section branches only work if the
206 // intermediate section boundaries are frozen.
207 if (target_sect != branch_sect) {
208 for (int n = MIN2(target_sect, branch_sect),
209 nlimit = (target_sect + branch_sect) - n;
210 n < nlimit; n++) {
211 CodeSection* cs = cb->code_section(n);
212 assert(cs->is_frozen(), "cross-section branch needs stable offsets");
213 }
214 }
215 #endif //ASSERT
216
217 // Push the target offset into the branch instruction.
218 masm->pd_patch_instruction(branch, target, file, line);
219 }
220 }
221
222 struct DelayedConstant {
223 typedef void (*value_fn_t)();
224 BasicType type;
225 intptr_t value;
226 value_fn_t value_fn;
227 // This limit of 20 is generous for initial uses.
228 // The limit needs to be large enough to store the field offsets
229 // into classes which do not have statically fixed layouts.
230 // (Initial use is for method handle object offsets.)
231 // Look for uses of "delayed_value" in the source code
232 // and make sure this number is generous enough to handle all of them.
233 enum { DC_LIMIT = 20 };
234 static DelayedConstant delayed_constants[DC_LIMIT];
235 static DelayedConstant* add(BasicType type, value_fn_t value_fn);
236 bool match(BasicType t, value_fn_t cfn) {
237 return type == t && value_fn == cfn;
238 }
239 static void update_all();
240 };
241
242 DelayedConstant DelayedConstant::delayed_constants[DC_LIMIT];
243 // Default C structure initialization rules have the following effect here:
244 // = { { (BasicType)0, (intptr_t)NULL }, ... };
245
246 DelayedConstant* DelayedConstant::add(BasicType type,
247 DelayedConstant::value_fn_t cfn) {
248 for (int i = 0; i < DC_LIMIT; i++) {
249 DelayedConstant* dcon = &delayed_constants[i];
250 if (dcon->match(type, cfn))
251 return dcon;
252 if (dcon->value_fn == NULL) {
253 dcon->value_fn = cfn;
254 dcon->type = type;
255 return dcon;
256 }
257 }
258 // If this assert is hit (in pre-integration testing!) then re-evaluate
259 // the comment on the definition of DC_LIMIT.
260 guarantee(false, "too many delayed constants");
261 return NULL;
262 }
263
264 void DelayedConstant::update_all() {
265 for (int i = 0; i < DC_LIMIT; i++) {
266 DelayedConstant* dcon = &delayed_constants[i];
267 if (dcon->value_fn != NULL && dcon->value == 0) {
268 typedef int (*int_fn_t)();
269 typedef address (*address_fn_t)();
270 switch (dcon->type) {
271 case T_INT: dcon->value = (intptr_t) ((int_fn_t) dcon->value_fn)(); break;
272 case T_ADDRESS: dcon->value = (intptr_t) ((address_fn_t)dcon->value_fn)(); break;
273 default: break;
274 }
275 }
276 }
277 }
278
279 RegisterOrConstant AbstractAssembler::delayed_value(int(*value_fn)(), Register tmp, int offset) {
280 intptr_t val = (intptr_t) (*value_fn)();
281 if (val != 0) return val + offset;
282 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
283 }
284 RegisterOrConstant AbstractAssembler::delayed_value(address(*value_fn)(), Register tmp, int offset) {
285 intptr_t val = (intptr_t) (*value_fn)();
286 if (val != 0) return val + offset;
287 return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
288 }
289 intptr_t* AbstractAssembler::delayed_value_addr(int(*value_fn)()) {
290 DelayedConstant* dcon = DelayedConstant::add(T_INT, (DelayedConstant::value_fn_t) value_fn);
291 return &dcon->value;
292 }
293 intptr_t* AbstractAssembler::delayed_value_addr(address(*value_fn)()) {
294 DelayedConstant* dcon = DelayedConstant::add(T_ADDRESS, (DelayedConstant::value_fn_t) value_fn);
295 return &dcon->value;
296 }
297 void AbstractAssembler::update_delayed_values() {
298 DelayedConstant::update_all();
299 }
300
301 void AbstractAssembler::block_comment(const char* comment) {
302 if (sect() == CodeBuffer::SECT_INSTS) {
303 code_section()->outer()->block_comment(offset(), comment);
304 }
305 }
306
307 const char* AbstractAssembler::code_string(const char* str) {
308 if (sect() == CodeBuffer::SECT_INSTS || sect() == CodeBuffer::SECT_STUBS) {
309 return code_section()->outer()->code_string(str);
310 }
311 return NULL;
312 }
313
314 bool MacroAssembler::needs_explicit_null_check(intptr_t offset) {
315 // Exception handler checks the nmethod's implicit null checks table
316 // only when this method returns false.
317 #ifdef AARCH64
318 // AArch64 addresses passed from the signal handler may have
319 // their top 8 bits zeroed. That affects the case where
320 // Shenandoah tries to load a Brooks pointer via a null oop.
321 const uintptr_t address_bits = (uintptr_t)0xfffffffffffful;
322 #else
323 const uintptr_t address_bits = ~(uintptr_t)0;
324 #endif
325 #ifdef _LP64
326 if (UseCompressedOops && Universe::narrow_oop_base() != NULL) {
327 assert (Universe::heap() != NULL, "java heap should be initialized");
328 // The first page after heap_base is unmapped and
329 // the 'offset' is equal to [heap_base + offset] for
330 // narrow oop implicit null checks.
331 uintptr_t base = (uintptr_t)Universe::narrow_oop_base();
332 int adj = 0;
333 #if INCLUDE_SHENANDOAHGC
334 if (UseShenandoahGC) {
335 adj = ShenandoahBrooksPointer::byte_offset();
336 assert(adj < 0, "no need for positive adjustments");
337 }
338 #endif
339 if ((uintptr_t)((offset - adj) & address_bits) >= base) {
340 // Normalize offset for the next check.
341 offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1));
342 }
343 }
344 #endif
345
346 #if INCLUDE_SHENANDOAHGC
347 if (UseShenandoahGC && ((offset & address_bits) == (ShenandoahBrooksPointer::byte_offset() & address_bits))) {
348 return false;
349 }
350 #endif
351
352 return offset < 0 || os::vm_page_size() <= offset;
353 }