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/barrierSet.hpp" |
30 | #include "gc/shared/barrierSetAssembler.hpp" |
31 | #include "gc/shared/collectedHeap.hpp" |
32 | #include "interpreter/interpreter.hpp" |
33 | #include "oops/arrayOop.hpp" |
34 | #include "oops/markOop.hpp" |
35 | #include "runtime/basicLock.hpp" |
36 | #include "runtime/biasedLocking.hpp" |
37 | #include "runtime/os.hpp" |
38 | #include "runtime/sharedRuntime.hpp" |
39 | #include "runtime/stubRoutines.hpp" |
40 | |
41 | int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) { |
42 | const int aligned_mask = BytesPerWord -1; |
43 | const int hdr_offset = oopDesc::mark_offset_in_bytes(); |
44 | assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction" ); |
45 | assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different" ); |
46 | Label done; |
47 | int null_check_offset = -1; |
48 | |
49 | verify_oop(obj); |
50 | |
51 | // save object being locked into the BasicObjectLock |
52 | movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj); |
53 | |
54 | if (UseBiasedLocking) { |
55 | assert(scratch != noreg, "should have scratch register at this point" ); |
56 | null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case); |
57 | } else { |
58 | null_check_offset = offset(); |
59 | } |
60 | |
61 | // Load object header |
62 | movptr(hdr, Address(obj, hdr_offset)); |
63 | // and mark it as unlocked |
64 | orptr(hdr, markOopDesc::unlocked_value); |
65 | // save unlocked object header into the displaced header location on the stack |
66 | movptr(Address(disp_hdr, 0), hdr); |
67 | // test if object header is still the same (i.e. unlocked), and if so, store the |
68 | // displaced header address in the object header - if it is not the same, get the |
69 | // object header instead |
70 | MacroAssembler::lock(); // must be immediately before cmpxchg! |
71 | cmpxchgptr(disp_hdr, Address(obj, hdr_offset)); |
72 | // if the object header was the same, we're done |
73 | if (PrintBiasedLockingStatistics) { |
74 | cond_inc32(Assembler::equal, |
75 | ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr())); |
76 | } |
77 | jcc(Assembler::equal, done); |
78 | // if the object header was not the same, it is now in the hdr register |
79 | // => test if it is a stack pointer into the same stack (recursive locking), i.e.: |
80 | // |
81 | // 1) (hdr & aligned_mask) == 0 |
82 | // 2) rsp <= hdr |
83 | // 3) hdr <= rsp + page_size |
84 | // |
85 | // these 3 tests can be done by evaluating the following expression: |
86 | // |
87 | // (hdr - rsp) & (aligned_mask - page_size) |
88 | // |
89 | // assuming both the stack pointer and page_size have their least |
90 | // significant 2 bits cleared and page_size is a power of 2 |
91 | subptr(hdr, rsp); |
92 | andptr(hdr, aligned_mask - os::vm_page_size()); |
93 | // for recursive locking, the result is zero => save it in the displaced header |
94 | // location (NULL in the displaced hdr location indicates recursive locking) |
95 | movptr(Address(disp_hdr, 0), hdr); |
96 | // otherwise we don't care about the result and handle locking via runtime call |
97 | jcc(Assembler::notZero, slow_case); |
98 | // done |
99 | bind(done); |
100 | return null_check_offset; |
101 | } |
102 | |
103 | |
104 | void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { |
105 | const int aligned_mask = BytesPerWord -1; |
106 | const int hdr_offset = oopDesc::mark_offset_in_bytes(); |
107 | assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction" ); |
108 | assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different" ); |
109 | Label done; |
110 | |
111 | if (UseBiasedLocking) { |
112 | // load object |
113 | movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
114 | biased_locking_exit(obj, hdr, done); |
115 | } |
116 | |
117 | // load displaced header |
118 | movptr(hdr, Address(disp_hdr, 0)); |
119 | // if the loaded hdr is NULL we had recursive locking |
120 | testptr(hdr, hdr); |
121 | // if we had recursive locking, we are done |
122 | jcc(Assembler::zero, done); |
123 | if (!UseBiasedLocking) { |
124 | // load object |
125 | movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
126 | } |
127 | verify_oop(obj); |
128 | // test if object header is pointing to the displaced header, and if so, restore |
129 | // the displaced header in the object - if the object header is not pointing to |
130 | // the displaced header, get the object header instead |
131 | MacroAssembler::lock(); // must be immediately before cmpxchg! |
132 | cmpxchgptr(hdr, Address(obj, hdr_offset)); |
133 | // if the object header was not pointing to the displaced header, |
134 | // we do unlocking via runtime call |
135 | jcc(Assembler::notEqual, slow_case); |
136 | // done |
137 | bind(done); |
138 | } |
139 | |
140 | |
141 | // Defines obj, preserves var_size_in_bytes |
142 | void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { |
143 | if (UseTLAB) { |
144 | tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); |
145 | } else { |
146 | eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); |
147 | } |
148 | } |
149 | |
150 | |
151 | void C1_MacroAssembler::(Register obj, Register klass, Register len, Register t1, Register t2) { |
152 | assert_different_registers(obj, klass, len); |
153 | if (UseBiasedLocking && !len->is_valid()) { |
154 | assert_different_registers(obj, klass, len, t1, t2); |
155 | movptr(t1, Address(klass, Klass::prototype_header_offset())); |
156 | movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1); |
157 | } else { |
158 | // This assumes that all prototype bits fit in an int32_t |
159 | movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype()); |
160 | } |
161 | #ifdef _LP64 |
162 | if (UseCompressedClassPointers) { // Take care not to kill klass |
163 | movptr(t1, klass); |
164 | encode_klass_not_null(t1); |
165 | movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1); |
166 | } else |
167 | #endif |
168 | { |
169 | movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass); |
170 | } |
171 | |
172 | if (len->is_valid()) { |
173 | movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len); |
174 | } |
175 | #ifdef _LP64 |
176 | else if (UseCompressedClassPointers) { |
177 | xorptr(t1, t1); |
178 | store_klass_gap(obj, t1); |
179 | } |
180 | #endif |
181 | } |
182 | |
183 | |
184 | // preserves obj, destroys len_in_bytes |
185 | void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) { |
186 | assert(hdr_size_in_bytes >= 0, "header size must be positive or 0" ); |
187 | Label done; |
188 | |
189 | // len_in_bytes is positive and ptr sized |
190 | subptr(len_in_bytes, hdr_size_in_bytes); |
191 | jcc(Assembler::zero, done); |
192 | zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1); |
193 | bind(done); |
194 | } |
195 | |
196 | |
197 | void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int , int object_size, Register klass, Label& slow_case) { |
198 | assert(obj == rax, "obj must be in rax, for cmpxchg" ); |
199 | assert_different_registers(obj, t1, t2); // XXX really? |
200 | assert(header_size >= 0 && object_size >= header_size, "illegal sizes" ); |
201 | |
202 | try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); |
203 | |
204 | initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB); |
205 | } |
206 | |
207 | 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) { |
208 | assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, |
209 | "con_size_in_bytes is not multiple of alignment" ); |
210 | const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; |
211 | |
212 | initialize_header(obj, klass, noreg, t1, t2); |
213 | |
214 | if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) { |
215 | // clear rest of allocated space |
216 | const Register t1_zero = t1; |
217 | const Register index = t2; |
218 | const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below) |
219 | if (var_size_in_bytes != noreg) { |
220 | mov(index, var_size_in_bytes); |
221 | initialize_body(obj, index, hdr_size_in_bytes, t1_zero); |
222 | } else if (con_size_in_bytes <= threshold) { |
223 | // use explicit null stores |
224 | // code size = 2 + 3*n bytes (n = number of fields to clear) |
225 | xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) |
226 | for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) |
227 | movptr(Address(obj, i), t1_zero); |
228 | } else if (con_size_in_bytes > hdr_size_in_bytes) { |
229 | // use loop to null out the fields |
230 | // code size = 16 bytes for even n (n = number of fields to clear) |
231 | // initialize last object field first if odd number of fields |
232 | xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) |
233 | movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3); |
234 | // initialize last object field if constant size is odd |
235 | if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0) |
236 | movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero); |
237 | // initialize remaining object fields: rdx is a multiple of 2 |
238 | { Label loop; |
239 | bind(loop); |
240 | movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)), |
241 | t1_zero); |
242 | NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)), |
243 | t1_zero);) |
244 | decrement(index); |
245 | jcc(Assembler::notZero, loop); |
246 | } |
247 | } |
248 | } |
249 | |
250 | if (CURRENT_ENV->dtrace_alloc_probes()) { |
251 | assert(obj == rax, "must be" ); |
252 | call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); |
253 | } |
254 | |
255 | verify_oop(obj); |
256 | } |
257 | |
258 | void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int , Address::ScaleFactor f, Register klass, Label& slow_case) { |
259 | assert(obj == rax, "obj must be in rax, for cmpxchg" ); |
260 | assert_different_registers(obj, len, t1, t2, klass); |
261 | |
262 | // determine alignment mask |
263 | assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work" ); |
264 | |
265 | // check for negative or excessive length |
266 | cmpptr(len, (int32_t)max_array_allocation_length); |
267 | jcc(Assembler::above, slow_case); |
268 | |
269 | const Register arr_size = t2; // okay to be the same |
270 | // align object end |
271 | movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask); |
272 | lea(arr_size, Address(arr_size, len, f)); |
273 | andptr(arr_size, ~MinObjAlignmentInBytesMask); |
274 | |
275 | try_allocate(obj, arr_size, 0, t1, t2, slow_case); |
276 | |
277 | initialize_header(obj, klass, len, t1, t2); |
278 | |
279 | // clear rest of allocated space |
280 | const Register len_zero = len; |
281 | initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero); |
282 | |
283 | if (CURRENT_ENV->dtrace_alloc_probes()) { |
284 | assert(obj == rax, "must be" ); |
285 | call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); |
286 | } |
287 | |
288 | verify_oop(obj); |
289 | } |
290 | |
291 | |
292 | |
293 | void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { |
294 | verify_oop(receiver); |
295 | // explicit NULL check not needed since load from [klass_offset] causes a trap |
296 | // check against inline cache |
297 | assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check" ); |
298 | int start_offset = offset(); |
299 | |
300 | if (UseCompressedClassPointers) { |
301 | load_klass(rscratch1, receiver); |
302 | cmpptr(rscratch1, iCache); |
303 | } else { |
304 | cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes())); |
305 | } |
306 | // if icache check fails, then jump to runtime routine |
307 | // Note: RECEIVER must still contain the receiver! |
308 | jump_cc(Assembler::notEqual, |
309 | RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
310 | const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9); |
311 | assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry" ); |
312 | } |
313 | |
314 | |
315 | void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) { |
316 | assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect" ); |
317 | // Make sure there is enough stack space for this method's activation. |
318 | // Note that we do this before doing an enter(). This matches the |
319 | // ordering of C2's stack overflow check / rsp decrement and allows |
320 | // the SharedRuntime stack overflow handling to be consistent |
321 | // between the two compilers. |
322 | generate_stack_overflow_check(bang_size_in_bytes); |
323 | |
324 | push(rbp); |
325 | if (PreserveFramePointer) { |
326 | mov(rbp, rsp); |
327 | } |
328 | #ifdef TIERED |
329 | // c2 leaves fpu stack dirty. Clean it on entry |
330 | if (UseSSE < 2 ) { |
331 | empty_FPU_stack(); |
332 | } |
333 | #endif // TIERED |
334 | decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0 |
335 | |
336 | BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler(); |
337 | bs->nmethod_entry_barrier(this); |
338 | } |
339 | |
340 | |
341 | void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) { |
342 | increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0 |
343 | pop(rbp); |
344 | } |
345 | |
346 | |
347 | void C1_MacroAssembler::verified_entry() { |
348 | if (C1Breakpoint || VerifyFPU || !UseStackBanging) { |
349 | // Verified Entry first instruction should be 5 bytes long for correct |
350 | // patching by patch_verified_entry(). |
351 | // |
352 | // C1Breakpoint and VerifyFPU have one byte first instruction. |
353 | // Also first instruction will be one byte "push(rbp)" if stack banging |
354 | // code is not generated (see build_frame() above). |
355 | // For all these cases generate long instruction first. |
356 | fat_nop(); |
357 | } |
358 | if (C1Breakpoint)int3(); |
359 | // build frame |
360 | verify_FPU(0, "method_entry" ); |
361 | } |
362 | |
363 | void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) { |
364 | // rbp, + 0: link |
365 | // + 1: return address |
366 | // + 2: argument with offset 0 |
367 | // + 3: argument with offset 1 |
368 | // + 4: ... |
369 | |
370 | movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord)); |
371 | } |
372 | |
373 | #ifndef PRODUCT |
374 | |
375 | void C1_MacroAssembler::verify_stack_oop(int stack_offset) { |
376 | if (!VerifyOops) return; |
377 | verify_oop_addr(Address(rsp, stack_offset)); |
378 | } |
379 | |
380 | void C1_MacroAssembler::verify_not_null_oop(Register r) { |
381 | if (!VerifyOops) return; |
382 | Label not_null; |
383 | testptr(r, r); |
384 | jcc(Assembler::notZero, not_null); |
385 | stop("non-null oop required" ); |
386 | bind(not_null); |
387 | verify_oop(r); |
388 | } |
389 | |
390 | void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) { |
391 | #ifdef ASSERT |
392 | if (inv_rax) movptr(rax, 0xDEAD); |
393 | if (inv_rbx) movptr(rbx, 0xDEAD); |
394 | if (inv_rcx) movptr(rcx, 0xDEAD); |
395 | if (inv_rdx) movptr(rdx, 0xDEAD); |
396 | if (inv_rsi) movptr(rsi, 0xDEAD); |
397 | if (inv_rdi) movptr(rdi, 0xDEAD); |
398 | #endif |
399 | } |
400 | |
401 | #endif // ifndef PRODUCT |
402 | |