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_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 shenandoah_store_check(obj);
138
139 biased_locking_exit(obj, hdr, done);
140 }
141
142 // load displaced header
143 ldr(hdr, Address(disp_hdr, 0));
144 // if the loaded hdr is NULL we had recursive locking
145 // if we had recursive locking, we are done
146 cbz(hdr, done);
147 if (!UseBiasedLocking) {
148 // load object
149 ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
150 }
151 verify_oop(obj);
152 shenandoah_store_check(obj);
153
154 // test if object header is pointing to the displaced header, and if so, restore
155 // the displaced header in the object - if the object header is not pointing to
156 // the displaced header, get the object header instead
157 // if the object header was not pointing to the displaced header,
158 // we do unlocking via runtime call
159 if (hdr_offset) {
160 lea(rscratch1, Address(obj, hdr_offset));
161 cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
162 } else {
163 cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
164 }
165 // done
166 bind(done);
167 }
168
169
170 // Defines obj, preserves var_size_in_bytes
171 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
172 if (UseTLAB) {
173 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
174 } else {
175 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
176 incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1);
177 }
178 }
179
180 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
181 assert_different_registers(obj, klass, len);
182 if (UseBiasedLocking && !len->is_valid()) {
183 assert_different_registers(obj, klass, len, t1, t2);
184 ldr(t1, Address(klass, Klass::prototype_header_offset()));
185 } else {
186 // This assumes that all prototype bits fit in an int32_t
187 mov(t1, (int32_t)(intptr_t)markOopDesc::prototype());
188 }
189 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
190
191 if (UseCompressedClassPointers) { // Take care not to kill klass
192 encode_klass_not_null(t1, klass);
193 strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
194 } else {
195 str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
196 }
197
198 if (len->is_valid()) {
199 strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
200 } else if (UseCompressedClassPointers) {
201 store_klass_gap(obj, zr);
202 }
203 }
204
205 // Zero words; len is in bytes
206 // Destroys all registers except addr
207 // len must be a nonzero multiple of wordSize
208 void C1_MacroAssembler::zero_memory(Register addr, Register len, Register t1) {
209 assert_different_registers(addr, len, t1, rscratch1, rscratch2);
210
211 #ifdef ASSERT
212 { Label L;
213 tst(len, BytesPerWord - 1);
214 br(Assembler::EQ, L);
215 stop("len is not a multiple of BytesPerWord");
216 bind(L);
217 }
218 #endif
219
220 #ifndef PRODUCT
221 block_comment("zero memory");
222 #endif
223
224 Label loop;
225 Label entry;
226
227 // Algorithm:
228 //
229 // scratch1 = cnt & 7;
230 // cnt -= scratch1;
231 // p += scratch1;
232 // switch (scratch1) {
233 // do {
234 // cnt -= 8;
235 // p[-8] = 0;
236 // case 7:
237 // p[-7] = 0;
238 // case 6:
239 // p[-6] = 0;
240 // // ...
241 // case 1:
242 // p[-1] = 0;
243 // case 0:
244 // p += 8;
245 // } while (cnt);
246 // }
247
248 const int unroll = 8; // Number of str(zr) instructions we'll unroll
249
250 lsr(len, len, LogBytesPerWord);
251 andr(rscratch1, len, unroll - 1); // tmp1 = cnt % unroll
252 sub(len, len, rscratch1); // cnt -= unroll
253 // t1 always points to the end of the region we're about to zero
254 add(t1, addr, rscratch1, Assembler::LSL, LogBytesPerWord);
255 adr(rscratch2, entry);
256 sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 2);
257 br(rscratch2);
258 bind(loop);
259 sub(len, len, unroll);
260 for (int i = -unroll; i < 0; i++)
261 str(zr, Address(t1, i * wordSize));
262 bind(entry);
263 add(t1, t1, unroll * wordSize);
264 cbnz(len, loop);
265 }
266
267 // preserves obj, destroys len_in_bytes
268 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
269 Label done;
270 assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different");
271 assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord");
272 Register index = len_in_bytes;
273 // index is positive and ptr sized
274 subs(index, index, hdr_size_in_bytes);
275 br(Assembler::EQ, done);
276 // note: for the remaining code to work, index must be a multiple of BytesPerWord
277 #ifdef ASSERT
278 { Label L;
279 tst(index, BytesPerWord - 1);
280 br(Assembler::EQ, L);
281 stop("index is not a multiple of BytesPerWord");
282 bind(L);
283 }
284 #endif
285
286 // Preserve obj
287 if (hdr_size_in_bytes)
288 add(obj, obj, hdr_size_in_bytes);
289 zero_memory(obj, index, t1);
290 if (hdr_size_in_bytes)
291 sub(obj, obj, hdr_size_in_bytes);
292
293 // done
294 bind(done);
295 }
296
297
298 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
299 assert_different_registers(obj, t1, t2); // XXX really?
300 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
301
302 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
303
304 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2);
305 }
306
307 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) {
308 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
309 "con_size_in_bytes is not multiple of alignment");
310 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
311
312 initialize_header(obj, klass, noreg, t1, t2);
313
314 // clear rest of allocated space
315 const Register index = t2;
316 const int threshold = 16 * BytesPerWord; // approximate break even point for code size (see comments below)
317 if (var_size_in_bytes != noreg) {
318 mov(index, var_size_in_bytes);
319 initialize_body(obj, index, hdr_size_in_bytes, t1);
320 } else if (con_size_in_bytes <= threshold) {
321 // use explicit null stores
322 int i = hdr_size_in_bytes;
323 if (i < con_size_in_bytes && (con_size_in_bytes % (2 * BytesPerWord))) {
324 str(zr, Address(obj, i));
325 i += BytesPerWord;
326 }
327 for (; i < con_size_in_bytes; i += 2 * BytesPerWord)
328 stp(zr, zr, Address(obj, i));
329 } else if (con_size_in_bytes > hdr_size_in_bytes) {
330 block_comment("zero memory");
331 // use loop to null out the fields
332
333 int words = (con_size_in_bytes - hdr_size_in_bytes) / BytesPerWord;
334 mov(index, words / 8);
335
336 const int unroll = 8; // Number of str(zr) instructions we'll unroll
337 int remainder = words % unroll;
338 lea(rscratch1, Address(obj, hdr_size_in_bytes + remainder * BytesPerWord));
339
340 Label entry_point, loop;
341 b(entry_point);
342
343 bind(loop);
344 sub(index, index, 1);
345 for (int i = -unroll; i < 0; i++) {
346 if (-i == remainder)
347 bind(entry_point);
348 str(zr, Address(rscratch1, i * wordSize));
349 }
350 if (remainder == 0)
351 bind(entry_point);
352 add(rscratch1, rscratch1, unroll * wordSize);
353 cbnz(index, loop);
354
355 }
356
357 membar(StoreStore);
358
359 if (CURRENT_ENV->dtrace_alloc_probes()) {
360 assert(obj == r0, "must be");
361 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
362 }
363
364 verify_oop(obj);
365 }
366 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, int f, Register klass, Label& slow_case) {
367 assert_different_registers(obj, len, t1, t2, klass);
368
369 // determine alignment mask
370 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
371
372 // check for negative or excessive length
373 mov(rscratch1, (int32_t)max_array_allocation_length);
374 cmp(len, rscratch1);
375 br(Assembler::HS, slow_case);
376
377 const Register arr_size = t2; // okay to be the same
378 // align object end
379 mov(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
380 add(arr_size, arr_size, len, ext::uxtw, f);
381 andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
382
383 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
384
385 initialize_header(obj, klass, len, t1, t2);
386
387 // clear rest of allocated space
388 const Register len_zero = len;
389 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
390
391 membar(StoreStore);
392
393 if (CURRENT_ENV->dtrace_alloc_probes()) {
394 assert(obj == r0, "must be");
395 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
396 }
397
398 verify_oop(obj);
399 }
400
401
402 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
403 verify_oop(receiver);
404 // explicit NULL check not needed since load from [klass_offset] causes a trap
405 // check against inline cache
406 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
407
408 cmp_klass(receiver, iCache, rscratch1);
409 }
410
411
412 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
413 // If we have to make this method not-entrant we'll overwrite its
414 // first instruction with a jump. For this action to be legal we
415 // must ensure that this first instruction is a B, BL, NOP, BKPT,
416 // SVC, HVC, or SMC. Make it a NOP.
417 nop();
418 assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
419 // Make sure there is enough stack space for this method's activation.
420 // Note that we do this before doing an enter().
421 generate_stack_overflow_check(bang_size_in_bytes);
422 MacroAssembler::build_frame(framesize + 2 * wordSize);
423 if (NotifySimulator) {
424 notify(Assembler::method_entry);
425 }
426 }
427
428 void C1_MacroAssembler::remove_frame(int framesize) {
429 MacroAssembler::remove_frame(framesize + 2 * wordSize);
430 if (NotifySimulator) {
431 notify(Assembler::method_reentry);
432 }
433 }
434
435
436 void C1_MacroAssembler::verified_entry() {
437 }
438
439 #ifndef PRODUCT
440
441 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
442 if (!VerifyOops) return;
443 verify_oop_addr(Address(sp, stack_offset), "oop");
444 }
445
446 void C1_MacroAssembler::verify_not_null_oop(Register r) {
447 if (!VerifyOops) return;
448 Label not_null;
449 cbnz(r, not_null);
450 stop("non-null oop required");
451 bind(not_null);
452 verify_oop(r);
453 }
454
455 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
456 #ifdef ASSERT
457 static int nn;
458 if (inv_r0) mov(r0, 0xDEAD);
459 if (inv_r19) mov(r19, 0xDEAD);
460 if (inv_r2) mov(r2, nn++);
461 if (inv_r3) mov(r3, 0xDEAD);
462 if (inv_r4) mov(r4, 0xDEAD);
463 if (inv_r5) mov(r5, 0xDEAD);
464 #endif
465 }
466 #endif // ifndef PRODUCT