1 /*
2 * Copyright (c) 2018, Red Hat, Inc. All rights reserved.
3 *
4 * This code is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 only, as
6 * published by the Free Software Foundation.
7 *
8 * This code is distributed in the hope that it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
11 * version 2 for more details (a copy is included in the LICENSE file that
12 * accompanied this code).
13 *
14 * You should have received a copy of the GNU General Public License version
15 * 2 along with this work; if not, write to the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
17 *
18 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
19 * or visit www.oracle.com if you need additional information or have any
20 * questions.
21 *
22 */
23
24 #include "precompiled.hpp"
25 #include "c1/c1_MacroAssembler.hpp"
26 #include "c1/c1_LIRAssembler.hpp"
27 #include "macroAssembler_aarch64.hpp"
28 #include "shenandoahBarrierSetAssembler_aarch64.hpp"
29 #include "gc_implementation/shenandoah/shenandoahBarrierSet.hpp"
30 #include "gc_implementation/shenandoah/shenandoahForwarding.hpp"
31 #include "gc_implementation/shenandoah/shenandoahHeap.hpp"
32 #include "gc_implementation/shenandoah/shenandoahHeapRegion.hpp"
33 #include "gc_implementation/shenandoah/shenandoahRuntime.hpp"
34 #include "gc_implementation/shenandoah/c1/shenandoahBarrierSetC1.hpp"
35 #include "runtime/stubCodeGenerator.hpp"
36 #include "runtime/thread.hpp"
37
38 ShenandoahBarrierSetAssembler* ShenandoahBarrierSetAssembler::bsasm() {
39 return ShenandoahBarrierSet::barrier_set()->bsasm();
40 }
41
42 #define __ masm->
43
44 void ShenandoahBarrierSetAssembler::arraycopy_prologue(MacroAssembler* masm, bool dest_uninitialized,
45 Register src, Register dst, Register count) {
46 if ((ShenandoahSATBBarrier && !dest_uninitialized) || ShenandoahStoreValEnqueueBarrier || ShenandoahLoadRefBarrier) {
47
48 Label done;
49
50 // Avoid calling runtime if count == 0
51 __ cbz(count, done);
52
53 // Is GC active?
54 Address gc_state(rthread, in_bytes(JavaThread::gc_state_offset()));
55 __ ldrb(rscratch1, gc_state);
56 if (ShenandoahSATBBarrier && dest_uninitialized) {
57 __ tbz(rscratch1, ShenandoahHeap::HAS_FORWARDED_BITPOS, done);
58 } else {
59 __ mov(rscratch2, ShenandoahHeap::HAS_FORWARDED | ShenandoahHeap::MARKING);
60 __ tst(rscratch1, rscratch2);
61 __ br(Assembler::EQ, done);
62 }
63
64 __ push_call_clobbered_registers();
65 if (UseCompressedOops) {
66 __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::arraycopy_barrier_narrow_oop_entry), src, dst, count);
67 } else {
68 __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::arraycopy_barrier_oop_entry), src, dst, count);
69 }
70 __ pop_call_clobbered_registers();
71 __ bind(done);
72 }
73 }
74
75 void ShenandoahBarrierSetAssembler::resolve_forward_pointer(MacroAssembler* masm, Register dst, Register tmp) {
76 assert(ShenandoahCASBarrier, "should be enabled");
77 Label is_null;
78 __ cbz(dst, is_null);
79 resolve_forward_pointer_not_null(masm, dst, tmp);
80 __ bind(is_null);
81 }
82
83 // IMPORTANT: This must preserve all registers, even rscratch1 and rscratch2, except those explicitely
84 // passed in.
85 void ShenandoahBarrierSetAssembler::resolve_forward_pointer_not_null(MacroAssembler* masm, Register dst, Register tmp) {
86 assert(ShenandoahCASBarrier || ShenandoahLoadRefBarrier, "should be enabled");
87 // The below loads the mark word, checks if the lowest two bits are
88 // set, and if so, clear the lowest two bits and copy the result
89 // to dst. Otherwise it leaves dst alone.
90 // Implementing this is surprisingly awkward. I do it here by:
91 // - Inverting the mark word
92 // - Test lowest two bits == 0
93 // - If so, set the lowest two bits
94 // - Invert the result back, and copy to dst
95
96 bool borrow_reg = (tmp == noreg);
97 if (borrow_reg) {
98 // No free registers available. Make one useful.
99 tmp = rscratch1;
100 if (tmp == dst) {
101 tmp = rscratch2;
102 }
103 __ push(RegSet::of(tmp), sp);
104 }
105
106 assert_different_registers(tmp, dst);
107
108 Label done;
109 __ ldr(tmp, Address(dst, oopDesc::mark_offset_in_bytes()));
110 __ eon(tmp, tmp, zr);
111 __ ands(zr, tmp, markOopDesc::lock_mask_in_place);
112 __ br(Assembler::NE, done);
113 __ orr(tmp, tmp, markOopDesc::marked_value);
114 __ eon(dst, tmp, zr);
115 __ bind(done);
116
117 if (borrow_reg) {
118 __ pop(RegSet::of(tmp), sp);
119 }
120 }
121
122 void ShenandoahBarrierSetAssembler::load_reference_barrier(MacroAssembler* masm, Register dst, Address load_addr) {
123 if (!ShenandoahLoadRefBarrier) {
124 return;
125 }
126
127 assert(dst != rscratch2, "need rscratch2");
128 assert_different_registers(load_addr.base(), load_addr.index(), rscratch1, rscratch2);
129
130 bool is_narrow = UseCompressedOops;
131
132 Label heap_stable, not_cset;
133 __ enter();
134 Address gc_state(rthread, in_bytes(JavaThread::gc_state_offset()));
135 __ ldrb(rscratch2, gc_state);
136
137 // Check for heap stability
138 __ tbz(rscratch2, ShenandoahHeap::HAS_FORWARDED_BITPOS, heap_stable);
139
140 // use r1 for load address
141 Register result_dst = dst;
142 if (dst == r1) {
143 __ mov(rscratch1, dst);
144 dst = rscratch1;
145 }
146
147 // Save r0 and r1, unless it is an output register
148 RegSet to_save = RegSet::of(r0, r1) - result_dst;
149 __ push(to_save, sp);
150 __ lea(r1, load_addr);
151 __ mov(r0, dst);
152
153 // Test for in-cset
154 __ mov(rscratch2, ShenandoahHeap::in_cset_fast_test_addr());
155 __ lsr(rscratch1, r0, ShenandoahHeapRegion::region_size_bytes_shift_jint());
156 __ ldrb(rscratch2, Address(rscratch2, rscratch1));
157 __ tbz(rscratch2, 0, not_cset);
158
159 __ push_call_clobbered_registers();
160 if (is_narrow) {
161 __ mov(lr, CAST_FROM_FN_PTR(address, ShenandoahRuntime::load_reference_barrier_narrow));
162 } else {
163 __ mov(lr, CAST_FROM_FN_PTR(address, ShenandoahRuntime::load_reference_barrier));
164 }
165 __ blr(lr);
166 __ mov(rscratch1, r0);
167 __ pop_call_clobbered_registers();
168 __ mov(r0, rscratch1);
169
170 __ bind(not_cset);
171
172 __ mov(result_dst, r0);
173 __ pop(to_save, sp);
174
175 __ bind(heap_stable);
176 __ leave();
177 }
178
179 void ShenandoahBarrierSetAssembler::storeval_barrier(MacroAssembler* masm, Register dst, Register tmp) {
180 if (ShenandoahStoreValEnqueueBarrier) {
181 // Save possibly live regs.
182 RegSet live_regs = RegSet::range(r0, r4) - dst;
183 __ push(live_regs, sp);
184 __ strd(v0, __ pre(sp, 2 * -wordSize));
185
186 __ g1_write_barrier_pre(noreg, dst, rthread, tmp, true, false);
187
188 // Restore possibly live regs.
189 __ ldrd(v0, __ post(sp, 2 * wordSize));
190 __ pop(live_regs, sp);
191 }
192 }
193
194 void ShenandoahBarrierSetAssembler::load_heap_oop(MacroAssembler* masm, Register dst, Address src) {
195 Register result_dst = dst;
196
197 // Preserve src location for LRB
198 if (dst == src.base() || dst == src.index()) {
199 dst = rscratch1;
200 }
201 assert_different_registers(dst, src.base(), src.index());
202
203 if (UseCompressedOops) {
204 __ ldrw(dst, src);
205 __ decode_heap_oop(dst);
206 } else {
207 __ ldr(dst, src);
208 }
209
210 load_reference_barrier(masm, dst, src);
211
212 if (dst != result_dst) {
213 __ mov(result_dst, dst);
214 dst = result_dst;
215 }
216 }
217
218 void ShenandoahBarrierSetAssembler::cmpxchg_oop(MacroAssembler* masm, Register addr, Register expected, Register new_val,
219 bool acquire, bool release, bool weak, bool is_cae,
220 Register result) {
221
222 Register tmp1 = rscratch1;
223 Register tmp2 = rscratch2;
224 bool is_narrow = UseCompressedOops;
225 Assembler::operand_size size = is_narrow ? Assembler::word : Assembler::xword;
226
227 assert_different_registers(addr, expected, new_val, tmp1, tmp2);
228
229 Label retry, done, fail;
230
231 // CAS, using LL/SC pair.
232 __ bind(retry);
233 __ load_exclusive(tmp1, addr, size, acquire);
234 if (is_narrow) {
235 __ cmpw(tmp1, expected);
236 } else {
237 __ cmp(tmp1, expected);
238 }
239 __ br(Assembler::NE, fail);
240 __ store_exclusive(tmp2, new_val, addr, size, release);
241 if (weak) {
242 __ cmpw(tmp2, 0u); // If the store fails, return NE to our caller
243 } else {
244 __ cbnzw(tmp2, retry);
245 }
246 __ b(done);
247
248 __ bind(fail);
249 // Check if rb(expected)==rb(tmp1)
250 // Shuffle registers so that we have memory value ready for next expected.
251 __ mov(tmp2, expected);
252 __ mov(expected, tmp1);
253 if (is_narrow) {
254 __ decode_heap_oop(tmp1, tmp1);
255 __ decode_heap_oop(tmp2, tmp2);
256 }
257 resolve_forward_pointer(masm, tmp1);
258 resolve_forward_pointer(masm, tmp2);
259 __ cmp(tmp1, tmp2);
260 // Retry with expected now being the value we just loaded from addr.
261 __ br(Assembler::EQ, retry);
262 if (is_cae && is_narrow) {
263 // For cmp-and-exchange and narrow oops, we need to restore
264 // the compressed old-value. We moved it to 'expected' a few lines up.
265 __ mov(result, expected);
266 }
267 __ bind(done);
268
269 if (is_cae) {
270 __ mov(result, tmp1);
271 } else {
272 __ cset(result, Assembler::EQ);
273 }
274 }
275
276 #undef __
277
278 #ifdef COMPILER1
279
280 #define __ ce->masm()->
281
282 void ShenandoahBarrierSetAssembler::gen_load_reference_barrier_stub(LIR_Assembler* ce, ShenandoahLoadReferenceBarrierStub* stub) {
283 __ bind(*stub->entry());
284
285 Register obj = stub->obj()->as_register();
286 Register res = stub->result()->as_register();
287 Register addr = stub->addr()->as_pointer_register();
288 Register tmp1 = stub->tmp1()->as_register();
289 Register tmp2 = stub->tmp2()->as_register();
290
291 assert(res == r0, "result must arrive in r0");
292
293 if (res != obj) {
294 __ mov(res, obj);
295 }
296
297 // Check for null.
298 __ cbz(res, *stub->continuation());
299
300 // Check for object in cset.
301 __ mov(tmp2, ShenandoahHeap::in_cset_fast_test_addr());
302 __ lsr(tmp1, res, ShenandoahHeapRegion::region_size_bytes_shift_jint());
303 __ ldrb(tmp2, Address(tmp2, tmp1));
304 __ cbz(tmp2, *stub->continuation());
305
306 // Check if object is already forwarded.
307 Label slow_path;
308 __ ldr(tmp1, Address(res, oopDesc::mark_offset_in_bytes()));
309 __ eon(tmp1, tmp1, zr);
310 __ ands(zr, tmp1, markOopDesc::lock_mask_in_place);
311 __ br(Assembler::NE, slow_path);
312
313 // Decode forwarded object.
314 __ orr(tmp1, tmp1, markOopDesc::marked_value);
315 __ eon(res, tmp1, zr);
316 __ b(*stub->continuation());
317
318 __ bind(slow_path);
319 ce->store_parameter(res, 0);
320 ce->store_parameter(addr, 1);
321 __ far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::shenandoah_lrb_slow_id)));
322
323 __ b(*stub->continuation());
324 }
325
326 #undef __
327
328 #endif // COMPILER1