1 /*
2 * Copyright (c) 1999, 2018, 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 "c1/c1_MacroAssembler.hpp"
27 #include "c1/c1_Runtime1.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "gc/shared/collectedHeap.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markOop.hpp"
33 #include "runtime/basicLock.hpp"
34 #include "runtime/biasedLocking.hpp"
35 #include "runtime/os.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38
39 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
40 const int aligned_mask = BytesPerWord -1;
41 const int hdr_offset = oopDesc::mark_offset_in_bytes();
42 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
43 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
44 Label done;
45 int null_check_offset = -1;
46
47 verify_oop(obj);
48
49 // save object being locked into the BasicObjectLock
50 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
51
52 if (UseBiasedLocking) {
53 assert(scratch != noreg, "should have scratch register at this point");
54 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
55 } else {
56 null_check_offset = offset();
57 }
58
59 // Load object header
60 movptr(hdr, Address(obj, hdr_offset));
61 // and mark it as unlocked
62 orptr(hdr, markOopDesc::unlocked_value);
63 // save unlocked object header into the displaced header location on the stack
64 movptr(Address(disp_hdr, 0), hdr);
65 // test if object header is still the same (i.e. unlocked), and if so, store the
66 // displaced header address in the object header - if it is not the same, get the
67 // object header instead
68 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
69 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
70 // if the object header was the same, we're done
71 if (PrintBiasedLockingStatistics) {
72 cond_inc32(Assembler::equal,
73 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
74 }
75 jcc(Assembler::equal, done);
76 // if the object header was not the same, it is now in the hdr register
77 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
78 //
79 // 1) (hdr & aligned_mask) == 0
80 // 2) rsp <= hdr
81 // 3) hdr <= rsp + page_size
82 //
83 // these 3 tests can be done by evaluating the following expression:
84 //
85 // (hdr - rsp) & (aligned_mask - page_size)
86 //
87 // assuming both the stack pointer and page_size have their least
88 // significant 2 bits cleared and page_size is a power of 2
89 subptr(hdr, rsp);
90 andptr(hdr, aligned_mask - os::vm_page_size());
91 // for recursive locking, the result is zero => save it in the displaced header
92 // location (NULL in the displaced hdr location indicates recursive locking)
93 movptr(Address(disp_hdr, 0), hdr);
94 // otherwise we don't care about the result and handle locking via runtime call
95 jcc(Assembler::notZero, slow_case);
96 // done
97 bind(done);
98 return null_check_offset;
99 }
100
101
102 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
103 const int aligned_mask = BytesPerWord -1;
104 const int hdr_offset = oopDesc::mark_offset_in_bytes();
105 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
106 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
107 Label done;
108
109 if (UseBiasedLocking) {
110 // load object
111 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
112 biased_locking_exit(obj, hdr, done);
113 }
114
115 // load displaced header
116 movptr(hdr, Address(disp_hdr, 0));
117 // if the loaded hdr is NULL we had recursive locking
118 testptr(hdr, hdr);
119 // if we had recursive locking, we are done
120 jcc(Assembler::zero, done);
121 if (!UseBiasedLocking) {
122 // load object
123 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
124 }
125 verify_oop(obj);
126 // test if object header is pointing to the displaced header, and if so, restore
127 // the displaced header in the object - if the object header is not pointing to
128 // the displaced header, get the object header instead
129 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
130 cmpxchgptr(hdr, Address(obj, hdr_offset));
131 // if the object header was not pointing to the displaced header,
132 // we do unlocking via runtime call
133 jcc(Assembler::notEqual, slow_case);
134 // done
135 bind(done);
136 }
137
138
139 // Defines obj, preserves var_size_in_bytes
140 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
141 if (UseTLAB) {
142 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
143 } else {
144 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
145 incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1);
146 }
147 }
148
149
150 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
151 assert_different_registers(obj, klass, len);
152 if (UseBiasedLocking && !len->is_valid()) {
153 assert_different_registers(obj, klass, len, t1, t2);
154 movptr(t1, Address(klass, Klass::prototype_header_offset()));
155 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
156 } else {
157 // This assumes that all prototype bits fit in an int32_t
158 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype());
159 }
160 #ifdef _LP64
161 if (UseCompressedClassPointers) { // Take care not to kill klass
162 movptr(t1, klass);
163 encode_klass_not_null(t1);
164 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
165 } else
166 #endif
167 {
168 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
169 }
170
171 if (len->is_valid()) {
172 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
173 }
174 #ifdef _LP64
175 else if (UseCompressedClassPointers) {
176 xorptr(t1, t1);
177 store_klass_gap(obj, t1);
178 }
179 #endif
180 }
181
182
183 // preserves obj, destroys len_in_bytes
184 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
185 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
186 Label done;
187
188 // len_in_bytes is positive and ptr sized
189 subptr(len_in_bytes, hdr_size_in_bytes);
190 jcc(Assembler::zero, done);
191 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
192 bind(done);
193 }
194
195
196 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
197 assert(obj == rax, "obj must be in rax, for cmpxchg");
198 assert_different_registers(obj, t1, t2); // XXX really?
199 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
200
201 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
202
203 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
204 }
205
206 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) {
207 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
208 "con_size_in_bytes is not multiple of alignment");
209 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
210
211 initialize_header(obj, klass, noreg, t1, t2);
212
213 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
214 // clear rest of allocated space
215 const Register t1_zero = t1;
216 const Register index = t2;
217 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
218 if (var_size_in_bytes != noreg) {
219 mov(index, var_size_in_bytes);
220 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
221 } else if (con_size_in_bytes <= threshold) {
222 // use explicit null stores
223 // code size = 2 + 3*n bytes (n = number of fields to clear)
224 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
225 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
226 movptr(Address(obj, i), t1_zero);
227 } else if (con_size_in_bytes > hdr_size_in_bytes) {
228 // use loop to null out the fields
229 // code size = 16 bytes for even n (n = number of fields to clear)
230 // initialize last object field first if odd number of fields
231 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
232 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
233 // initialize last object field if constant size is odd
234 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
235 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
236 // initialize remaining object fields: rdx is a multiple of 2
237 { Label loop;
238 bind(loop);
239 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
240 t1_zero);
241 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
242 t1_zero);)
243 decrement(index);
244 jcc(Assembler::notZero, loop);
245 }
246 }
247 }
248
249 if (CURRENT_ENV->dtrace_alloc_probes()) {
250 assert(obj == rax, "must be");
251 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
252 }
253
254 verify_oop(obj);
255 }
256
257 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
258 assert(obj == rax, "obj must be in rax, for cmpxchg");
259 assert_different_registers(obj, len, t1, t2, klass);
260
261 // determine alignment mask
262 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
263
264 // check for negative or excessive length
265 cmpptr(len, (int32_t)max_array_allocation_length);
266 jcc(Assembler::above, slow_case);
267
268 const Register arr_size = t2; // okay to be the same
269 // align object end
270 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
271 lea(arr_size, Address(arr_size, len, f));
272 andptr(arr_size, ~MinObjAlignmentInBytesMask);
273
274 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
275
276 initialize_header(obj, klass, len, t1, t2);
277
278 // clear rest of allocated space
279 const Register len_zero = len;
280 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
281
282 if (CURRENT_ENV->dtrace_alloc_probes()) {
283 assert(obj == rax, "must be");
284 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
285 }
286
287 verify_oop(obj);
288 }
289
290
291
292 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
293 verify_oop(receiver);
294 // explicit NULL check not needed since load from [klass_offset] causes a trap
295 // check against inline cache
296 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
297 int start_offset = offset();
298
299 if (UseCompressedClassPointers) {
300 load_klass(rscratch1, receiver);
301 cmpptr(rscratch1, iCache);
302 } else {
303 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
304 }
305 // if icache check fails, then jump to runtime routine
306 // Note: RECEIVER must still contain the receiver!
307 jump_cc(Assembler::notEqual,
308 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
309 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
310 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
311 }
312
313
314 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
315 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
316 // Make sure there is enough stack space for this method's activation.
317 // Note that we do this before doing an enter(). This matches the
318 // ordering of C2's stack overflow check / rsp decrement and allows
319 // the SharedRuntime stack overflow handling to be consistent
320 // between the two compilers.
321 generate_stack_overflow_check(bang_size_in_bytes);
322
323 push(rbp);
324 if (PreserveFramePointer) {
325 mov(rbp, rsp);
326 }
327 #ifdef TIERED
328 // c2 leaves fpu stack dirty. Clean it on entry
329 if (UseSSE < 2 ) {
330 empty_FPU_stack();
331 }
332 #endif // TIERED
333 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
334 }
335
336
337 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
338 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
339 pop(rbp);
340 }
341
342
343 void C1_MacroAssembler::verified_entry() {
344 if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
345 // Verified Entry first instruction should be 5 bytes long for correct
346 // patching by patch_verified_entry().
347 //
348 // C1Breakpoint and VerifyFPU have one byte first instruction.
349 // Also first instruction will be one byte "push(rbp)" if stack banging
350 // code is not generated (see build_frame() above).
351 // For all these cases generate long instruction first.
352 fat_nop();
353 }
354 if (C1Breakpoint)int3();
355 // build frame
356 verify_FPU(0, "method_entry");
357 }
358
359 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
360 // rbp, + 0: link
361 // + 1: return address
362 // + 2: argument with offset 0
363 // + 3: argument with offset 1
364 // + 4: ...
365
366 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
367 }
368
369 #ifndef PRODUCT
370
371 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
372 if (!VerifyOops) return;
373 verify_oop_addr(Address(rsp, stack_offset));
374 }
375
376 void C1_MacroAssembler::verify_not_null_oop(Register r) {
377 if (!VerifyOops) return;
378 Label not_null;
379 testptr(r, r);
380 jcc(Assembler::notZero, not_null);
381 stop("non-null oop required");
382 bind(not_null);
383 verify_oop(r);
384 }
385
386 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
387 #ifdef ASSERT
388 if (inv_rax) movptr(rax, 0xDEAD);
389 if (inv_rbx) movptr(rbx, 0xDEAD);
390 if (inv_rcx) movptr(rcx, 0xDEAD);
391 if (inv_rdx) movptr(rdx, 0xDEAD);
392 if (inv_rsi) movptr(rsi, 0xDEAD);
393 if (inv_rdi) movptr(rdi, 0xDEAD);
394 #endif
395 }
396
397 #endif // ifndef PRODUCT