1 /*
2 * Copyright (c) 1999, 2015, 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 "assembler_x86.hpp"
27 #include "c1/c1_MacroAssembler.hpp"
28 #include "c1/c1_Runtime1.hpp"
29 #include "classfile/systemDictionary.hpp"
30 #include "gc/shared/collectedHeap.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/markOop.hpp"
34 #include "runtime/basicLock.hpp"
35 #include "runtime/biasedLocking.hpp"
36 #include "runtime/os.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubRoutines.hpp"
39
40 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
41 const int aligned_mask = BytesPerWord -1;
42 const int hdr_offset = oopDesc::mark_offset_in_bytes();
43 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
44 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
45 Label done;
46 int null_check_offset = -1;
47
48 verify_oop(obj);
49
50 // save object being locked into the BasicObjectLock
51 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
52
53 if (UseBiasedLocking) {
54 assert(scratch != noreg, "should have scratch register at this point");
55 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
56 } else {
57 null_check_offset = offset();
58 }
59
60 // Load object header
61 movptr(hdr, Address(obj, hdr_offset));
62 // and mark it as unlocked
63 orptr(hdr, markOopDesc::unlocked_value);
64 // save unlocked object header into the displaced header location on the stack
65 movptr(Address(disp_hdr, 0), hdr);
66 // test if object header is still the same (i.e. unlocked), and if so, store the
67 // displaced header address in the object header - if it is not the same, get the
68 // object header instead
69 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
70 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
71 // if the object header was the same, we're done
72 if (PrintBiasedLockingStatistics) {
73 cond_inc32(Assembler::equal,
74 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
75 }
76 jcc(Assembler::equal, done);
77 // if the object header was not the same, it is now in the hdr register
78 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
79 //
80 // 1) (hdr & aligned_mask) == 0
81 // 2) rsp <= hdr
82 // 3) hdr <= rsp + page_size
83 //
84 // these 3 tests can be done by evaluating the following expression:
85 //
86 // (hdr - rsp) & (aligned_mask - page_size)
87 //
88 // assuming both the stack pointer and page_size have their least
89 // significant 2 bits cleared and page_size is a power of 2
90 subptr(hdr, rsp);
91 andptr(hdr, aligned_mask - os::vm_page_size());
92 // for recursive locking, the result is zero => save it in the displaced header
93 // location (NULL in the displaced hdr location indicates recursive locking)
94 movptr(Address(disp_hdr, 0), hdr);
95 // otherwise we don't care about the result and handle locking via runtime call
96 jcc(Assembler::notZero, slow_case);
97 // done
98 bind(done);
99 return null_check_offset;
100 }
101
102
103 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
104 const int aligned_mask = BytesPerWord -1;
105 const int hdr_offset = oopDesc::mark_offset_in_bytes();
106 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
107 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
108 Label done;
109
110 if (UseBiasedLocking) {
111 // load object
112 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
113 biased_locking_exit(obj, hdr, done);
114 }
115
116 // load displaced header
117 movptr(hdr, Address(disp_hdr, 0));
118 // if the loaded hdr is NULL we had recursive locking
119 testptr(hdr, hdr);
120 // if we had recursive locking, we are done
121 jcc(Assembler::zero, done);
122 if (!UseBiasedLocking) {
123 // load object
124 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
125 }
126 verify_oop(obj);
127 // test if object header is pointing to the displaced header, and if so, restore
128 // the displaced header in the object - if the object header is not pointing to
129 // the displaced header, get the object header instead
130 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
131 cmpxchgptr(hdr, Address(obj, hdr_offset));
132 // if the object header was not pointing to the displaced header,
133 // we do unlocking via runtime call
134 jcc(Assembler::notEqual, slow_case);
135 // done
136 bind(done);
137 }
138
139
140 // Defines obj, preserves var_size_in_bytes
141 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
142 if (UseTLAB) {
143 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
144 } else {
145 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
146 incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1);
147 }
148 }
149
150
151 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
152 assert_different_registers(obj, klass, len);
153 if (UseBiasedLocking && !len->is_valid()) {
154 assert_different_registers(obj, klass, len, t1, t2);
155 movptr(t1, Address(klass, Klass::prototype_header_offset()));
156 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
157 } else {
158 // This assumes that all prototype bits fit in an int32_t
159 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype());
160 }
161 #ifdef _LP64
162 if (UseCompressedClassPointers) { // Take care not to kill klass
163 movptr(t1, klass);
164 encode_klass_not_null(t1);
165 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
166 } else
167 #endif
168 {
169 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
170 }
171
172 if (len->is_valid()) {
173 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
174 }
175 #ifdef _LP64
176 else if (UseCompressedClassPointers) {
177 xorptr(t1, t1);
178 store_klass_gap(obj, t1);
179 }
180 #endif
181 }
182
183
184 // preserves obj, destroys len_in_bytes
185 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
186 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
187 Label done;
188
189 // len_in_bytes is positive and ptr sized
190 subptr(len_in_bytes, hdr_size_in_bytes);
191 jcc(Assembler::zero, done);
192 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
193 bind(done);
194 }
195
196
197 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
198 assert(obj == rax, "obj must be in rax, for cmpxchg");
199 assert_different_registers(obj, t1, t2); // XXX really?
200 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
201
202 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
203
204 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
205
206 HEAP_MONITORING(this, noreg, noreg, object_size * HeapWordSize, obj,
207 t1, t2, call(RuntimeAddress(Runtime1::entry_for(
208 Runtime1::heap_object_sample_id))););
209 }
210
211 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
212 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
213 "con_size_in_bytes is not multiple of alignment");
214 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
215
216 initialize_header(obj, klass, noreg, t1, t2);
217
218 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
219 // clear rest of allocated space
220 const Register t1_zero = t1;
221 const Register index = t2;
222 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
223 if (var_size_in_bytes != noreg) {
224 mov(index, var_size_in_bytes);
225 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
226 } else if (con_size_in_bytes <= threshold) {
227 // use explicit null stores
228 // code size = 2 + 3*n bytes (n = number of fields to clear)
229 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
230 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
231 movptr(Address(obj, i), t1_zero);
232 } else if (con_size_in_bytes > hdr_size_in_bytes) {
233 // use loop to null out the fields
234 // code size = 16 bytes for even n (n = number of fields to clear)
235 // initialize last object field first if odd number of fields
236 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
237 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
238 // initialize last object field if constant size is odd
239 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
240 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
241 // initialize remaining object fields: rdx is a multiple of 2
242 { Label loop;
243 bind(loop);
244 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
245 t1_zero);
246 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
247 t1_zero);)
248 decrement(index);
249 jcc(Assembler::notZero, loop);
250 }
251 }
252 }
253
254 if (CURRENT_ENV->dtrace_alloc_probes()) {
255 assert(obj == rax, "must be");
256 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
257 }
258
259 verify_oop(obj);
260 }
261
262 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
263 assert(obj == rax, "obj must be in rax, for cmpxchg");
264 assert_different_registers(obj, len, t1, t2, klass);
265
266 // determine alignment mask
267 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
268
269 // check for negative or excessive length
270 cmpptr(len, (int32_t)max_array_allocation_length);
271 jcc(Assembler::above, slow_case);
272
273 const Register arr_size = t2; // okay to be the same
274 // align object end
275 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
276 lea(arr_size, Address(arr_size, len, f));
277 andptr(arr_size, ~MinObjAlignmentInBytesMask);
278
279 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
280
281 initialize_header(obj, klass, len, t1, t2);
282
283 // clear rest of allocated space
284 const Register len_zero = len;
285 // Initialize body destroys arr_size so remember it.
286 push(arr_size);
287 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
288 pop(arr_size);
289
290 if (CURRENT_ENV->dtrace_alloc_probes()) {
291 assert(obj == rax, "must be");
292 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
293 }
294
295 HEAP_MONITORING(this, noreg, arr_size, 0, obj, t1, noreg,
296 call(RuntimeAddress(Runtime1::entry_for(
297 Runtime1::heap_object_sample_id))););
298 verify_oop(obj);
299 }
300
301
302
303 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
304 verify_oop(receiver);
305 // explicit NULL check not needed since load from [klass_offset] causes a trap
306 // check against inline cache
307 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
308 int start_offset = offset();
309
310 if (UseCompressedClassPointers) {
311 load_klass(rscratch1, receiver);
312 cmpptr(rscratch1, iCache);
313 } else {
314 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
315 }
316 // if icache check fails, then jump to runtime routine
317 // Note: RECEIVER must still contain the receiver!
318 jump_cc(Assembler::notEqual,
319 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
320 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
321 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
322 }
323
324
325 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
326 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
327 // Make sure there is enough stack space for this method's activation.
328 // Note that we do this before doing an enter(). This matches the
329 // ordering of C2's stack overflow check / rsp decrement and allows
330 // the SharedRuntime stack overflow handling to be consistent
331 // between the two compilers.
332 generate_stack_overflow_check(bang_size_in_bytes);
333
334 push(rbp);
335 if (PreserveFramePointer) {
336 mov(rbp, rsp);
337 }
338 #ifdef TIERED
339 // c2 leaves fpu stack dirty. Clean it on entry
340 if (UseSSE < 2 ) {
341 empty_FPU_stack();
342 }
343 #endif // TIERED
344 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
345 }
346
347
348 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
349 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
350 pop(rbp);
351 }
352
353
354 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
355 if (C1Breakpoint) int3();
356 inline_cache_check(receiver, ic_klass);
357 }
358
359
360 void C1_MacroAssembler::verified_entry() {
361 if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
362 // Verified Entry first instruction should be 5 bytes long for correct
363 // patching by patch_verified_entry().
364 //
365 // C1Breakpoint and VerifyFPU have one byte first instruction.
366 // Also first instruction will be one byte "push(rbp)" if stack banging
367 // code is not generated (see build_frame() above).
368 // For all these cases generate long instruction first.
369 fat_nop();
370 }
371 if (C1Breakpoint)int3();
372 // build frame
373 verify_FPU(0, "method_entry");
374 }
375
376
377 #ifndef PRODUCT
378
379 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
380 if (!VerifyOops) return;
381 verify_oop_addr(Address(rsp, stack_offset));
382 }
383
384 void C1_MacroAssembler::verify_not_null_oop(Register r) {
385 if (!VerifyOops) return;
386 Label not_null;
387 testptr(r, r);
388 jcc(Assembler::notZero, not_null);
389 stop("non-null oop required");
390 bind(not_null);
391 verify_oop(r);
392 }
393
394 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
395 #ifdef ASSERT
396 if (inv_rax) movptr(rax, 0xDEAD);
397 if (inv_rbx) movptr(rbx, 0xDEAD);
398 if (inv_rcx) movptr(rcx, 0xDEAD);
399 if (inv_rdx) movptr(rdx, 0xDEAD);
400 if (inv_rsi) movptr(rsi, 0xDEAD);
401 if (inv_rdi) movptr(rdi, 0xDEAD);
402 #endif
403 }
404
405 #endif // ifndef PRODUCT