1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2012, 2015 SAP AG. 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/stubRoutines.hpp"
38 #include "runtime/sharedRuntime.hpp"
39
40
41 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
42 const Register temp_reg = R12_scratch2;
43 verify_oop(receiver);
44 load_klass(temp_reg, receiver);
45 if (TrapBasedICMissChecks) {
46 trap_ic_miss_check(temp_reg, iCache);
47 } else {
48 Label L;
49 cmpd(CCR0, temp_reg, iCache);
50 beq(CCR0, L);
51 //load_const_optimized(temp_reg, SharedRuntime::get_ic_miss_stub(), R0);
52 calculate_address_from_global_toc(temp_reg, SharedRuntime::get_ic_miss_stub(), true, true, false);
53 mtctr(temp_reg);
54 bctr();
55 align(32, 12);
56 bind(L);
57 }
58 }
59
60
61 void C1_MacroAssembler::explicit_null_check(Register base) {
62 Unimplemented();
63 }
64
65
66 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
67 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
68 // Make sure there is enough stack space for this method's activation.
69 generate_stack_overflow_check(bang_size_in_bytes);
70
71 // Create the frame.
72 const Register return_pc = R0;
73
74 mflr(return_pc);
75 // Get callers sp.
76 std(return_pc, _abi(lr), R1_SP); // SP->lr = return_pc
77 push_frame(frame_size_in_bytes, R0); // SP -= frame_size_in_bytes
78 }
79
80
81 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
82 Unimplemented(); // Currently unused.
83 //if (C1Breakpoint) illtrap();
84 //inline_cache_check(receiver, ic_klass);
85 }
86
87
88 void C1_MacroAssembler::verified_entry() {
89 if (C1Breakpoint) illtrap();
90 // build frame
91 }
92
93
94 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) {
95 assert_different_registers(Rmark, Roop, Rbox, Rscratch);
96
97 Label done, cas_failed, slow_int;
98
99 // The following move must be the first instruction of emitted since debug
100 // information may be generated for it.
101 // Load object header.
102 ld(Rmark, oopDesc::mark_offset_in_bytes(), Roop);
103
104 verify_oop(Roop);
105
106 // Save object being locked into the BasicObjectLock...
107 std(Roop, BasicObjectLock::obj_offset_in_bytes(), Rbox);
108
109 if (UseBiasedLocking) {
110 biased_locking_enter(CCR0, Roop, Rmark, Rscratch, R0, done, &slow_int);
111 }
112
113 // ... and mark it unlocked.
114 ori(Rmark, Rmark, markOopDesc::unlocked_value);
115
116 // Save unlocked object header into the displaced header location on the stack.
117 std(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
118
119 // Compare object markOop with Rmark and if equal exchange Rscratch with object markOop.
120 assert(oopDesc::mark_offset_in_bytes() == 0, "cas must take a zero displacement");
121 cmpxchgd(/*flag=*/CCR0,
122 /*current_value=*/Rscratch,
123 /*compare_value=*/Rmark,
124 /*exchange_value=*/Rbox,
125 /*where=*/Roop/*+0==mark_offset_in_bytes*/,
126 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
127 MacroAssembler::cmpxchgx_hint_acquire_lock(),
128 noreg,
129 &cas_failed,
130 /*check without membar and ldarx first*/true);
131 // If compare/exchange succeeded we found an unlocked object and we now have locked it
132 // hence we are done.
133 b(done);
134
135 bind(slow_int);
136 b(slow_case); // far
137
138 bind(cas_failed);
139 // We did not find an unlocked object so see if this is a recursive case.
140 sub(Rscratch, Rscratch, R1_SP);
141 load_const_optimized(R0, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place));
142 and_(R0/*==0?*/, Rscratch, R0);
143 std(R0/*==0, perhaps*/, BasicLock::displaced_header_offset_in_bytes(), Rbox);
144 bne(CCR0, slow_int);
145
146 bind(done);
147 }
148
149
150 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
151 assert_different_registers(Rmark, Roop, Rbox);
152
153 Label slow_int, done;
154
155 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
156 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
157
158 if (UseBiasedLocking) {
159 // Load the object out of the BasicObjectLock.
160 ld(Roop, BasicObjectLock::obj_offset_in_bytes(), Rbox);
161 verify_oop(Roop);
162 biased_locking_exit(CCR0, Roop, R0, done);
163 }
164 // Test first it it is a fast recursive unlock.
165 ld(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
166 cmpdi(CCR0, Rmark, 0);
167 beq(CCR0, done);
168 if (!UseBiasedLocking) {
169 // Load object.
170 ld(Roop, BasicObjectLock::obj_offset_in_bytes(), Rbox);
171 verify_oop(Roop);
172 }
173
174 // Check if it is still a light weight lock, this is is true if we see
175 // the stack address of the basicLock in the markOop of the object.
176 cmpxchgd(/*flag=*/CCR0,
177 /*current_value=*/R0,
178 /*compare_value=*/Rbox,
179 /*exchange_value=*/Rmark,
180 /*where=*/Roop,
181 MacroAssembler::MemBarRel,
182 MacroAssembler::cmpxchgx_hint_release_lock(),
183 noreg,
184 &slow_int);
185 b(done);
186 bind(slow_int);
187 b(slow_case); // far
188
189 // Done
190 bind(done);
191 }
192
193
194 void C1_MacroAssembler::try_allocate(
195 Register obj, // result: pointer to object after successful allocation
196 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
197 int con_size_in_bytes, // object size in bytes if known at compile time
198 Register t1, // temp register, must be global register for incr_allocated_bytes
199 Register t2, // temp register
200 Label& slow_case // continuation point if fast allocation fails
201 ) {
202 if (UseTLAB) {
203 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
204 } else {
205 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
206 RegisterOrConstant size_in_bytes = var_size_in_bytes->is_valid()
207 ? RegisterOrConstant(var_size_in_bytes)
208 : RegisterOrConstant(con_size_in_bytes);
209 incr_allocated_bytes(size_in_bytes, t1, t2);
210 }
211 }
212
213
214 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
215 assert_different_registers(obj, klass, len, t1, t2);
216 if (UseBiasedLocking && !len->is_valid()) {
217 ld(t1, in_bytes(Klass::prototype_header_offset()), klass);
218 } else {
219 load_const_optimized(t1, (intx)markOopDesc::prototype());
220 }
221 std(t1, oopDesc::mark_offset_in_bytes(), obj);
222 store_klass(obj, klass);
223 if (len->is_valid()) {
224 stw(len, arrayOopDesc::length_offset_in_bytes(), obj);
225 } else if (UseCompressedClassPointers) {
226 // Otherwise length is in the class gap.
227 store_klass_gap(obj);
228 }
229 }
230
231
232 void C1_MacroAssembler::initialize_body(Register base, Register index) {
233 assert_different_registers(base, index);
234 srdi(index, index, LogBytesPerWord);
235 clear_memory_doubleword(base, index);
236 }
237
238 void C1_MacroAssembler::initialize_body(Register obj, Register tmp1, Register tmp2,
239 int obj_size_in_bytes, int hdr_size_in_bytes) {
240 const int index = (obj_size_in_bytes - hdr_size_in_bytes) / HeapWordSize;
241
242 const int cl_size = VM_Version::L1_data_cache_line_size(),
243 cl_dwords = cl_size>>3,
244 cl_dw_addr_bits = exact_log2(cl_dwords);
245
246 const Register tmp = R0,
247 base_ptr = tmp1,
248 cnt_dwords = tmp2;
249
250 if (index <= 6) {
251 // Use explicit NULL stores.
252 if (index > 0) { li(tmp, 0); }
253 for (int i = 0; i < index; ++i) { std(tmp, hdr_size_in_bytes + i * HeapWordSize, obj); }
254
255 } else if (index < (2<<cl_dw_addr_bits)-1) {
256 // simple loop
257 Label loop;
258
259 li(cnt_dwords, index);
260 addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
261 li(tmp, 0);
262 mtctr(cnt_dwords); // Load counter.
263 bind(loop);
264 std(tmp, 0, base_ptr); // Clear 8byte aligned block.
265 addi(base_ptr, base_ptr, 8);
266 bdnz(loop);
267
268 } else {
269 // like clear_memory_doubleword
270 Label startloop, fast, fastloop, restloop, done;
271
272 addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
273 load_const_optimized(cnt_dwords, index);
274 rldicl_(tmp, base_ptr, 64-3, 64-cl_dw_addr_bits); // Extract dword offset within first cache line.
275 beq(CCR0, fast); // Already 128byte aligned.
276
277 subfic(tmp, tmp, cl_dwords);
278 mtctr(tmp); // Set ctr to hit 128byte boundary (0<ctr<cl_dwords).
279 subf(cnt_dwords, tmp, cnt_dwords); // rest.
280 li(tmp, 0);
281
282 bind(startloop); // Clear at the beginning to reach 128byte boundary.
283 std(tmp, 0, base_ptr); // Clear 8byte aligned block.
284 addi(base_ptr, base_ptr, 8);
285 bdnz(startloop);
286
287 bind(fast); // Clear 128byte blocks.
288 srdi(tmp, cnt_dwords, cl_dw_addr_bits); // Loop count for 128byte loop (>0).
289 andi(cnt_dwords, cnt_dwords, cl_dwords-1); // Rest in dwords.
290 mtctr(tmp); // Load counter.
291
292 bind(fastloop);
293 dcbz(base_ptr); // Clear 128byte aligned block.
294 addi(base_ptr, base_ptr, cl_size);
295 bdnz(fastloop);
296
297 cmpdi(CCR0, cnt_dwords, 0); // size 0?
298 beq(CCR0, done); // rest == 0
299 li(tmp, 0);
300 mtctr(cnt_dwords); // Load counter.
301
302 bind(restloop); // Clear rest.
303 std(tmp, 0, base_ptr); // Clear 8byte aligned block.
304 addi(base_ptr, base_ptr, 8);
305 bdnz(restloop);
306
307 bind(done);
308 }
309 }
310
311 void C1_MacroAssembler::allocate_object(
312 Register obj, // result: pointer to object after successful allocation
313 Register t1, // temp register
314 Register t2, // temp register
315 Register t3, // temp register
316 int hdr_size, // object header size in words
317 int obj_size, // object size in words
318 Register klass, // object klass
319 Label& slow_case // continuation point if fast allocation fails
320 ) {
321 assert_different_registers(obj, t1, t2, t3, klass);
322
323 // allocate space & initialize header
324 if (!is_simm16(obj_size * wordSize)) {
325 // Would need to use extra register to load
326 // object size => go the slow case for now.
327 b(slow_case);
328 return;
329 }
330 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case);
331
332 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
333 }
334
335 void C1_MacroAssembler::initialize_object(
336 Register obj, // result: pointer to object after successful allocation
337 Register klass, // object klass
338 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
339 int con_size_in_bytes, // object size in bytes if known at compile time
340 Register t1, // temp register
341 Register t2 // temp register
342 ) {
343 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
344
345 initialize_header(obj, klass, noreg, t1, t2);
346
347 #ifdef ASSERT
348 {
349 lwz(t1, in_bytes(Klass::layout_helper_offset()), klass);
350 if (var_size_in_bytes != noreg) {
351 cmpw(CCR0, t1, var_size_in_bytes);
352 } else {
353 cmpwi(CCR0, t1, con_size_in_bytes);
354 }
355 asm_assert_eq("bad size in initialize_object", 0x753);
356 }
357 #endif
358
359 // Initialize body.
360 if (var_size_in_bytes != noreg) {
361 // Use a loop.
362 addi(t1, obj, hdr_size_in_bytes); // Compute address of first element.
363 addi(t2, var_size_in_bytes, -hdr_size_in_bytes); // Compute size of body.
364 initialize_body(t1, t2);
365 } else if (con_size_in_bytes > hdr_size_in_bytes) {
366 // Use a loop.
367 initialize_body(obj, t1, t2, con_size_in_bytes, hdr_size_in_bytes);
368 }
369
370 if (CURRENT_ENV->dtrace_alloc_probes()) {
371 Unimplemented();
372 // assert(obj == O0, "must be");
373 // call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
374 // relocInfo::runtime_call_type);
375 }
376
377 verify_oop(obj);
378 }
379
380
381 void C1_MacroAssembler::allocate_array(
382 Register obj, // result: pointer to array after successful allocation
383 Register len, // array length
384 Register t1, // temp register
385 Register t2, // temp register
386 Register t3, // temp register
387 int hdr_size, // object header size in words
388 int elt_size, // element size in bytes
389 Register klass, // object klass
390 Label& slow_case // continuation point if fast allocation fails
391 ) {
392 assert_different_registers(obj, len, t1, t2, t3, klass);
393
394 // Determine alignment mask.
395 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
396 int log2_elt_size = exact_log2(elt_size);
397
398 // Check for negative or excessive length.
399 size_t max_length = max_array_allocation_length >> log2_elt_size;
400 if (UseTLAB) {
401 size_t max_tlab = align_size_up(ThreadLocalAllocBuffer::max_size() >> log2_elt_size, 64*K);
402 if (max_tlab < max_length) { max_length = max_tlab; }
403 }
404 load_const_optimized(t1, max_length);
405 cmpld(CCR0, len, t1);
406 bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::greater), slow_case);
407
408 // compute array size
409 // note: If 0 <= len <= max_length, len*elt_size + header + alignment is
410 // smaller or equal to the largest integer; also, since top is always
411 // aligned, we can do the alignment here instead of at the end address
412 // computation.
413 const Register arr_size = t1;
414 Register arr_len_in_bytes = len;
415 if (elt_size != 1) {
416 sldi(t1, len, log2_elt_size);
417 arr_len_in_bytes = t1;
418 }
419 addi(arr_size, arr_len_in_bytes, hdr_size * wordSize + MinObjAlignmentInBytesMask); // Add space for header & alignment.
420 clrrdi(arr_size, arr_size, LogMinObjAlignmentInBytes); // Align array size.
421
422 // Allocate space & initialize header.
423 if (UseTLAB) {
424 tlab_allocate(obj, arr_size, 0, t2, slow_case);
425 } else {
426 eden_allocate(obj, arr_size, 0, t2, t3, slow_case);
427 }
428 initialize_header(obj, klass, len, t2, t3);
429
430 // Initialize body.
431 const Register base = t2;
432 const Register index = t3;
433 addi(base, obj, hdr_size * wordSize); // compute address of first element
434 addi(index, arr_size, -(hdr_size * wordSize)); // compute index = number of bytes to clear
435 initialize_body(base, index);
436
437 if (CURRENT_ENV->dtrace_alloc_probes()) {
438 Unimplemented();
439 //assert(obj == O0, "must be");
440 //call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
441 // relocInfo::runtime_call_type);
442 }
443
444 verify_oop(obj);
445 }
446
447
448 #ifndef PRODUCT
449
450 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
451 verify_oop_addr((RegisterOrConstant)(stack_offset + STACK_BIAS), R1_SP, "broken oop in stack slot");
452 }
453
454 void C1_MacroAssembler::verify_not_null_oop(Register r) {
455 Label not_null;
456 cmpdi(CCR0, r, 0);
457 bne(CCR0, not_null);
458 stop("non-null oop required");
459 bind(not_null);
460 if (!VerifyOops) return;
461 verify_oop(r);
462 }
463
464 #endif // PRODUCT
465
466 void C1_MacroAssembler::null_check(Register r, Label* Lnull) {
467 if (TrapBasedNullChecks) { // SIGTRAP based
468 trap_null_check(r);
469 } else { // explicit
470 //const address exception_entry = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id);
471 assert(Lnull != NULL, "must have Label for explicit check");
472 cmpdi(CCR0, r, 0);
473 bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::equal), *Lnull);
474 }
475 }
476
477 address C1_MacroAssembler::call_c_with_frame_resize(address dest, int frame_resize) {
478 if (frame_resize) { resize_frame(-frame_resize, R0); }
479 #if defined(ABI_ELFv2)
480 address return_pc = call_c(dest, relocInfo::runtime_call_type);
481 #else
482 address return_pc = call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, dest), relocInfo::runtime_call_type);
483 #endif
484 if (frame_resize) { resize_frame(frame_resize, R0); }
485 return return_pc;
486 }