1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.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 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
41 FloatRegister f0, FloatRegister f1,
42 Register result)
43 {
44 Label done;
45 if (is_float) {
46 fcmps(f0, f1);
47 } else {
48 fcmpd(f0, f1);
49 }
50 if (unordered_result < 0) {
51 // we want -1 for unordered or less than, 0 for equal and 1 for
52 // greater than.
53 cset(result, NE); // Not equal or unordered
54 cneg(result, result, LT); // Less than or unordered
55 } else {
56 // we want -1 for less than, 0 for equal and 1 for unordered or
57 // greater than.
58 cset(result, NE); // Not equal or unordered
59 cneg(result, result, LO); // Less than
60 }
61 }
62
63 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
64 const int aligned_mask = BytesPerWord -1;
65 const int hdr_offset = oopDesc::mark_offset_in_bytes();
66 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
67 Label done, fail;
68 int null_check_offset = -1;
69
70 verify_oop(obj);
71
72 shenandoah_store_addr_check(obj);
73
74 // save object being locked into the BasicObjectLock
75 str(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
76
77 if (UseBiasedLocking) {
78 assert(scratch != noreg, "should have scratch register at this point");
79 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
80 } else {
81 null_check_offset = offset();
82 }
83
84 // Load object header
85 ldr(hdr, Address(obj, hdr_offset));
86 // and mark it as unlocked
87 orr(hdr, hdr, markOopDesc::unlocked_value);
88 // save unlocked object header into the displaced header location on the stack
89 str(hdr, Address(disp_hdr, 0));
90 // test if object header is still the same (i.e. unlocked), and if so, store the
91 // displaced header address in the object header - if it is not the same, get the
92 // object header instead
93 lea(rscratch2, Address(obj, hdr_offset));
94 cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/NULL);
95 // if the object header was the same, we're done
96 // if the object header was not the same, it is now in the hdr register
97 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
98 //
99 // 1) (hdr & aligned_mask) == 0
100 // 2) sp <= hdr
101 // 3) hdr <= sp + page_size
102 //
103 // these 3 tests can be done by evaluating the following expression:
104 //
105 // (hdr - sp) & (aligned_mask - page_size)
106 //
107 // assuming both the stack pointer and page_size have their least
108 // significant 2 bits cleared and page_size is a power of 2
109 mov(rscratch1, sp);
110 sub(hdr, hdr, rscratch1);
111 ands(hdr, hdr, aligned_mask - os::vm_page_size());
112 // for recursive locking, the result is zero => save it in the displaced header
113 // location (NULL in the displaced hdr location indicates recursive locking)
114 str(hdr, Address(disp_hdr, 0));
115 // otherwise we don't care about the result and handle locking via runtime call
116 cbnz(hdr, slow_case);
117 // done
118 bind(done);
119 if (PrintBiasedLockingStatistics) {
120 lea(rscratch2, ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
121 addmw(Address(rscratch2, 0), 1, rscratch1);
122 }
123 return null_check_offset;
124 }
125
126
127 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
128 const int aligned_mask = BytesPerWord -1;
129 const int hdr_offset = oopDesc::mark_offset_in_bytes();
130 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
131 Label done;
132
133 if (UseBiasedLocking) {
134 // load object
135 ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
136
137 biased_locking_exit(obj, hdr, done);
138 }
139
140 // load displaced header
141 ldr(hdr, Address(disp_hdr, 0));
142 // if the loaded hdr is NULL we had recursive locking
143 // if we had recursive locking, we are done
144 cbz(hdr, done);
145 if (!UseBiasedLocking) {
146 // load object
147 ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
148 }
149 verify_oop(obj);
150
151 shenandoah_store_addr_check(obj);
152
153 // test if object header is pointing to the displaced header, and if so, restore
154 // the displaced header in the object - if the object header is not pointing to
155 // the displaced header, get the object header instead
156 // if the object header was not pointing to the displaced header,
157 // we do unlocking via runtime call
158 if (hdr_offset) {
159 lea(rscratch1, Address(obj, hdr_offset));
160 cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
161 } else {
162 cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
163 }
164 // done
165 bind(done);
166 }
167
168
169 // Defines obj, preserves var_size_in_bytes
170 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
171 if (UseTLAB) {
172 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
173 } else {
174 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
175 incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1);
176 }
177 }
178
179 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
180 assert_different_registers(obj, klass, len);
181 if (UseBiasedLocking && !len->is_valid()) {
182 assert_different_registers(obj, klass, len, t1, t2);
183 ldr(t1, Address(klass, Klass::prototype_header_offset()));
184 } else {
185 // This assumes that all prototype bits fit in an int32_t
186 mov(t1, (int32_t)(intptr_t)markOopDesc::prototype());
187 }
188 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
189
190 if (UseCompressedClassPointers) { // Take care not to kill klass
191 encode_klass_not_null(t1, klass);
192 strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
193 } else {
194 str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
195 }
196
197 if (len->is_valid()) {
198 strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
199 } else if (UseCompressedClassPointers) {
200 store_klass_gap(obj, zr);
201 }
202 }
203
204 // Zero words; len is in bytes
205 // Destroys all registers except addr
206 // len must be a nonzero multiple of wordSize
207 void C1_MacroAssembler::zero_memory(Register addr, Register len, Register t1) {
208 assert_different_registers(addr, len, t1, rscratch1, rscratch2);
209
210 #ifdef ASSERT
211 { Label L;
212 tst(len, BytesPerWord - 1);
213 br(Assembler::EQ, L);
214 stop("len is not a multiple of BytesPerWord");
215 bind(L);
216 }
217 #endif
218
219 #ifndef PRODUCT
220 block_comment("zero memory");
221 #endif
222
223 Label loop;
224 Label entry;
225
226 // Algorithm:
227 //
228 // scratch1 = cnt & 7;
229 // cnt -= scratch1;
230 // p += scratch1;
231 // switch (scratch1) {
232 // do {
233 // cnt -= 8;
234 // p[-8] = 0;
235 // case 7:
236 // p[-7] = 0;
237 // case 6:
238 // p[-6] = 0;
239 // // ...
240 // case 1:
241 // p[-1] = 0;
242 // case 0:
243 // p += 8;
244 // } while (cnt);
245 // }
246
247 const int unroll = 8; // Number of str(zr) instructions we'll unroll
248
249 lsr(len, len, LogBytesPerWord);
250 andr(rscratch1, len, unroll - 1); // tmp1 = cnt % unroll
251 sub(len, len, rscratch1); // cnt -= unroll
252 // t1 always points to the end of the region we're about to zero
253 add(t1, addr, rscratch1, Assembler::LSL, LogBytesPerWord);
254 adr(rscratch2, entry);
255 sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 2);
256 br(rscratch2);
257 bind(loop);
258 sub(len, len, unroll);
259 for (int i = -unroll; i < 0; i++)
260 str(zr, Address(t1, i * wordSize));
261 bind(entry);
262 add(t1, t1, unroll * wordSize);
263 cbnz(len, loop);
264 }
265
266 // preserves obj, destroys len_in_bytes
267 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
268 Label done;
269 assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different");
270 assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord");
271 Register index = len_in_bytes;
272 // index is positive and ptr sized
273 subs(index, index, hdr_size_in_bytes);
274 br(Assembler::EQ, done);
275 // note: for the remaining code to work, index must be a multiple of BytesPerWord
276 #ifdef ASSERT
277 { Label L;
278 tst(index, BytesPerWord - 1);
279 br(Assembler::EQ, L);
280 stop("index is not a multiple of BytesPerWord");
281 bind(L);
282 }
283 #endif
284
285 // Preserve obj
286 if (hdr_size_in_bytes)
287 add(obj, obj, hdr_size_in_bytes);
288 zero_memory(obj, index, t1);
289 if (hdr_size_in_bytes)
290 sub(obj, obj, hdr_size_in_bytes);
291
292 // done
293 bind(done);
294 }
295
296
297 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
298 assert_different_registers(obj, t1, t2); // XXX really?
299 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
300
301 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
302
303 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2);
304 }
305
306 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) {
307 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
308 "con_size_in_bytes is not multiple of alignment");
309 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
310
311 initialize_header(obj, klass, noreg, t1, t2);
312
313 // clear rest of allocated space
314 const Register index = t2;
315 const int threshold = 16 * BytesPerWord; // approximate break even point for code size (see comments below)
316 if (var_size_in_bytes != noreg) {
317 mov(index, var_size_in_bytes);
318 initialize_body(obj, index, hdr_size_in_bytes, t1);
319 } else if (con_size_in_bytes <= threshold) {
320 // use explicit null stores
321 int i = hdr_size_in_bytes;
322 if (i < con_size_in_bytes && (con_size_in_bytes % (2 * BytesPerWord))) {
323 str(zr, Address(obj, i));
324 i += BytesPerWord;
325 }
326 for (; i < con_size_in_bytes; i += 2 * BytesPerWord)
327 stp(zr, zr, Address(obj, i));
328 } else if (con_size_in_bytes > hdr_size_in_bytes) {
329 block_comment("zero memory");
330 // use loop to null out the fields
331
332 int words = (con_size_in_bytes - hdr_size_in_bytes) / BytesPerWord;
333 mov(index, words / 8);
334
335 const int unroll = 8; // Number of str(zr) instructions we'll unroll
336 int remainder = words % unroll;
337 lea(rscratch1, Address(obj, hdr_size_in_bytes + remainder * BytesPerWord));
338
339 Label entry_point, loop;
340 b(entry_point);
341
342 bind(loop);
343 sub(index, index, 1);
344 for (int i = -unroll; i < 0; i++) {
345 if (-i == remainder)
346 bind(entry_point);
347 str(zr, Address(rscratch1, i * wordSize));
348 }
349 if (remainder == 0)
350 bind(entry_point);
351 add(rscratch1, rscratch1, unroll * wordSize);
352 cbnz(index, loop);
353
354 }
355
356 membar(StoreStore);
357
358 if (CURRENT_ENV->dtrace_alloc_probes()) {
359 assert(obj == r0, "must be");
360 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
361 }
362
363 verify_oop(obj);
364 }
365 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, int f, Register klass, Label& slow_case) {
366 assert_different_registers(obj, len, t1, t2, klass);
367
368 // determine alignment mask
369 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
370
371 // check for negative or excessive length
372 mov(rscratch1, (int32_t)max_array_allocation_length);
373 cmp(len, rscratch1);
374 br(Assembler::HS, slow_case);
375
376 const Register arr_size = t2; // okay to be the same
377 // align object end
378 mov(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
379 add(arr_size, arr_size, len, ext::uxtw, f);
380 andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
381
382 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
383
384 initialize_header(obj, klass, len, t1, t2);
385
386 // clear rest of allocated space
387 const Register len_zero = len;
388 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
389
390 membar(StoreStore);
391
392 if (CURRENT_ENV->dtrace_alloc_probes()) {
393 assert(obj == r0, "must be");
394 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
395 }
396
397 verify_oop(obj);
398 }
399
400
401 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
402 verify_oop(receiver);
403 // explicit NULL check not needed since load from [klass_offset] causes a trap
404 // check against inline cache
405 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
406
407 cmp_klass(receiver, iCache, rscratch1);
408 }
409
410
411 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
412 // If we have to make this method not-entrant we'll overwrite its
413 // first instruction with a jump. For this action to be legal we
414 // must ensure that this first instruction is a B, BL, NOP, BKPT,
415 // SVC, HVC, or SMC. Make it a NOP.
416 nop();
417 assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
418 // Make sure there is enough stack space for this method's activation.
419 // Note that we do this before doing an enter().
420 generate_stack_overflow_check(bang_size_in_bytes);
421 MacroAssembler::build_frame(framesize + 2 * wordSize);
422 if (NotifySimulator) {
423 notify(Assembler::method_entry);
424 }
425 }
426
427 void C1_MacroAssembler::remove_frame(int framesize) {
428 MacroAssembler::remove_frame(framesize + 2 * wordSize);
429 if (NotifySimulator) {
430 notify(Assembler::method_reentry);
431 }
432 }
433
434
435 void C1_MacroAssembler::verified_entry() {
436 }
437
438 #ifndef PRODUCT
439
440 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
441 if (!VerifyOops) return;
442 verify_oop_addr(Address(sp, stack_offset), "oop");
443 }
444
445 void C1_MacroAssembler::verify_not_null_oop(Register r) {
446 if (!VerifyOops) return;
447 Label not_null;
448 cbnz(r, not_null);
449 stop("non-null oop required");
450 bind(not_null);
451 verify_oop(r);
452 }
453
454 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
455 #ifdef ASSERT
456 static int nn;
457 if (inv_r0) mov(r0, 0xDEAD);
458 if (inv_r19) mov(r19, 0xDEAD);
459 if (inv_r2) mov(r2, nn++);
460 if (inv_r3) mov(r3, 0xDEAD);
461 if (inv_r4) mov(r4, 0xDEAD);
462 if (inv_r5) mov(r5, 0xDEAD);
463 #endif
464 }
465 #endif // ifndef PRODUCT