|
1 /* |
|
2 * Copyright (c) 1999, 2013, 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_interface/collectedHeap.hpp" |
|
30 #include "interpreter/interpreter.hpp" |
|
31 #include "oops/arrayOop.hpp" |
|
32 #include "oops/markOop.hpp" |
|
33 #include "runtime/basicLock.hpp" |
|
34 #include "runtime/biasedLocking.hpp" |
|
35 #include "runtime/os.hpp" |
|
36 #include "runtime/stubRoutines.hpp" |
|
37 |
|
38 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) { |
|
39 const int aligned_mask = BytesPerWord -1; |
|
40 const int hdr_offset = oopDesc::mark_offset_in_bytes(); |
|
41 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction"); |
|
42 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); |
|
43 Label done; |
|
44 int null_check_offset = -1; |
|
45 |
|
46 verify_oop(obj); |
|
47 |
|
48 // save object being locked into the BasicObjectLock |
|
49 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj); |
|
50 |
|
51 if (UseBiasedLocking) { |
|
52 assert(scratch != noreg, "should have scratch register at this point"); |
|
53 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case); |
|
54 } else { |
|
55 null_check_offset = offset(); |
|
56 } |
|
57 |
|
58 // Load object header |
|
59 movptr(hdr, Address(obj, hdr_offset)); |
|
60 // and mark it as unlocked |
|
61 orptr(hdr, markOopDesc::unlocked_value); |
|
62 // save unlocked object header into the displaced header location on the stack |
|
63 movptr(Address(disp_hdr, 0), hdr); |
|
64 // test if object header is still the same (i.e. unlocked), and if so, store the |
|
65 // displaced header address in the object header - if it is not the same, get the |
|
66 // object header instead |
|
67 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! |
|
68 cmpxchgptr(disp_hdr, Address(obj, hdr_offset)); |
|
69 // if the object header was the same, we're done |
|
70 if (PrintBiasedLockingStatistics) { |
|
71 cond_inc32(Assembler::equal, |
|
72 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr())); |
|
73 } |
|
74 jcc(Assembler::equal, done); |
|
75 // if the object header was not the same, it is now in the hdr register |
|
76 // => test if it is a stack pointer into the same stack (recursive locking), i.e.: |
|
77 // |
|
78 // 1) (hdr & aligned_mask) == 0 |
|
79 // 2) rsp <= hdr |
|
80 // 3) hdr <= rsp + page_size |
|
81 // |
|
82 // these 3 tests can be done by evaluating the following expression: |
|
83 // |
|
84 // (hdr - rsp) & (aligned_mask - page_size) |
|
85 // |
|
86 // assuming both the stack pointer and page_size have their least |
|
87 // significant 2 bits cleared and page_size is a power of 2 |
|
88 subptr(hdr, rsp); |
|
89 andptr(hdr, aligned_mask - os::vm_page_size()); |
|
90 // for recursive locking, the result is zero => save it in the displaced header |
|
91 // location (NULL in the displaced hdr location indicates recursive locking) |
|
92 movptr(Address(disp_hdr, 0), hdr); |
|
93 // otherwise we don't care about the result and handle locking via runtime call |
|
94 jcc(Assembler::notZero, slow_case); |
|
95 // done |
|
96 bind(done); |
|
97 return null_check_offset; |
|
98 } |
|
99 |
|
100 |
|
101 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { |
|
102 const int aligned_mask = BytesPerWord -1; |
|
103 const int hdr_offset = oopDesc::mark_offset_in_bytes(); |
|
104 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction"); |
|
105 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); |
|
106 Label done; |
|
107 |
|
108 if (UseBiasedLocking) { |
|
109 // load object |
|
110 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
|
111 biased_locking_exit(obj, hdr, done); |
|
112 } |
|
113 |
|
114 // load displaced header |
|
115 movptr(hdr, Address(disp_hdr, 0)); |
|
116 // if the loaded hdr is NULL we had recursive locking |
|
117 testptr(hdr, hdr); |
|
118 // if we had recursive locking, we are done |
|
119 jcc(Assembler::zero, done); |
|
120 if (!UseBiasedLocking) { |
|
121 // load object |
|
122 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
|
123 } |
|
124 verify_oop(obj); |
|
125 // test if object header is pointing to the displaced header, and if so, restore |
|
126 // the displaced header in the object - if the object header is not pointing to |
|
127 // the displaced header, get the object header instead |
|
128 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! |
|
129 cmpxchgptr(hdr, Address(obj, hdr_offset)); |
|
130 // if the object header was not pointing to the displaced header, |
|
131 // we do unlocking via runtime call |
|
132 jcc(Assembler::notEqual, slow_case); |
|
133 // done |
|
134 bind(done); |
|
135 } |
|
136 |
|
137 |
|
138 // Defines obj, preserves var_size_in_bytes |
|
139 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { |
|
140 if (UseTLAB) { |
|
141 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); |
|
142 } else { |
|
143 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); |
|
144 incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1); |
|
145 } |
|
146 } |
|
147 |
|
148 |
|
149 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { |
|
150 assert_different_registers(obj, klass, len); |
|
151 if (UseBiasedLocking && !len->is_valid()) { |
|
152 assert_different_registers(obj, klass, len, t1, t2); |
|
153 movptr(t1, Address(klass, Klass::prototype_header_offset())); |
|
154 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1); |
|
155 } else { |
|
156 // This assumes that all prototype bits fit in an int32_t |
|
157 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype()); |
|
158 } |
|
159 #ifdef _LP64 |
|
160 if (UseCompressedClassPointers) { // Take care not to kill klass |
|
161 movptr(t1, klass); |
|
162 encode_klass_not_null(t1); |
|
163 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1); |
|
164 } else |
|
165 #endif |
|
166 { |
|
167 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass); |
|
168 } |
|
169 |
|
170 if (len->is_valid()) { |
|
171 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len); |
|
172 } |
|
173 #ifdef _LP64 |
|
174 else if (UseCompressedClassPointers) { |
|
175 xorptr(t1, t1); |
|
176 store_klass_gap(obj, t1); |
|
177 } |
|
178 #endif |
|
179 } |
|
180 |
|
181 |
|
182 // preserves obj, destroys len_in_bytes |
|
183 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) { |
|
184 Label done; |
|
185 assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different"); |
|
186 assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord"); |
|
187 Register index = len_in_bytes; |
|
188 // index is positive and ptr sized |
|
189 subptr(index, hdr_size_in_bytes); |
|
190 jcc(Assembler::zero, done); |
|
191 // initialize topmost word, divide index by 2, check if odd and test if zero |
|
192 // note: for the remaining code to work, index must be a multiple of BytesPerWord |
|
193 #ifdef ASSERT |
|
194 { Label L; |
|
195 testptr(index, BytesPerWord - 1); |
|
196 jcc(Assembler::zero, L); |
|
197 stop("index is not a multiple of BytesPerWord"); |
|
198 bind(L); |
|
199 } |
|
200 #endif |
|
201 xorptr(t1, t1); // use _zero reg to clear memory (shorter code) |
|
202 if (UseIncDec) { |
|
203 shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set |
|
204 } else { |
|
205 shrptr(index, 2); // use 2 instructions to avoid partial flag stall |
|
206 shrptr(index, 1); |
|
207 } |
|
208 #ifndef _LP64 |
|
209 // index could have been not a multiple of 8 (i.e., bit 2 was set) |
|
210 { Label even; |
|
211 // note: if index was a multiple of 8, than it cannot |
|
212 // be 0 now otherwise it must have been 0 before |
|
213 // => if it is even, we don't need to check for 0 again |
|
214 jcc(Assembler::carryClear, even); |
|
215 // clear topmost word (no jump needed if conditional assignment would work here) |
|
216 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1); |
|
217 // index could be 0 now, need to check again |
|
218 jcc(Assembler::zero, done); |
|
219 bind(even); |
|
220 } |
|
221 #endif // !_LP64 |
|
222 // initialize remaining object fields: rdx is a multiple of 2 now |
|
223 { Label loop; |
|
224 bind(loop); |
|
225 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 1*BytesPerWord), t1); |
|
226 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 2*BytesPerWord), t1);) |
|
227 decrement(index); |
|
228 jcc(Assembler::notZero, loop); |
|
229 } |
|
230 |
|
231 // done |
|
232 bind(done); |
|
233 } |
|
234 |
|
235 |
|
236 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) { |
|
237 assert(obj == rax, "obj must be in rax, for cmpxchg"); |
|
238 assert_different_registers(obj, t1, t2); // XXX really? |
|
239 assert(header_size >= 0 && object_size >= header_size, "illegal sizes"); |
|
240 |
|
241 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); |
|
242 |
|
243 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2); |
|
244 } |
|
245 |
|
246 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) { |
|
247 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, |
|
248 "con_size_in_bytes is not multiple of alignment"); |
|
249 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; |
|
250 |
|
251 initialize_header(obj, klass, noreg, t1, t2); |
|
252 |
|
253 // clear rest of allocated space |
|
254 const Register t1_zero = t1; |
|
255 const Register index = t2; |
|
256 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below) |
|
257 if (var_size_in_bytes != noreg) { |
|
258 mov(index, var_size_in_bytes); |
|
259 initialize_body(obj, index, hdr_size_in_bytes, t1_zero); |
|
260 } else if (con_size_in_bytes <= threshold) { |
|
261 // use explicit null stores |
|
262 // code size = 2 + 3*n bytes (n = number of fields to clear) |
|
263 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) |
|
264 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) |
|
265 movptr(Address(obj, i), t1_zero); |
|
266 } else if (con_size_in_bytes > hdr_size_in_bytes) { |
|
267 // use loop to null out the fields |
|
268 // code size = 16 bytes for even n (n = number of fields to clear) |
|
269 // initialize last object field first if odd number of fields |
|
270 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) |
|
271 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3); |
|
272 // initialize last object field if constant size is odd |
|
273 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0) |
|
274 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero); |
|
275 // initialize remaining object fields: rdx is a multiple of 2 |
|
276 { Label loop; |
|
277 bind(loop); |
|
278 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)), |
|
279 t1_zero); |
|
280 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)), |
|
281 t1_zero);) |
|
282 decrement(index); |
|
283 jcc(Assembler::notZero, loop); |
|
284 } |
|
285 } |
|
286 |
|
287 if (CURRENT_ENV->dtrace_alloc_probes()) { |
|
288 assert(obj == rax, "must be"); |
|
289 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); |
|
290 } |
|
291 |
|
292 verify_oop(obj); |
|
293 } |
|
294 |
|
295 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) { |
|
296 assert(obj == rax, "obj must be in rax, for cmpxchg"); |
|
297 assert_different_registers(obj, len, t1, t2, klass); |
|
298 |
|
299 // determine alignment mask |
|
300 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); |
|
301 |
|
302 // check for negative or excessive length |
|
303 cmpptr(len, (int32_t)max_array_allocation_length); |
|
304 jcc(Assembler::above, slow_case); |
|
305 |
|
306 const Register arr_size = t2; // okay to be the same |
|
307 // align object end |
|
308 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask); |
|
309 lea(arr_size, Address(arr_size, len, f)); |
|
310 andptr(arr_size, ~MinObjAlignmentInBytesMask); |
|
311 |
|
312 try_allocate(obj, arr_size, 0, t1, t2, slow_case); |
|
313 |
|
314 initialize_header(obj, klass, len, t1, t2); |
|
315 |
|
316 // clear rest of allocated space |
|
317 const Register len_zero = len; |
|
318 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero); |
|
319 |
|
320 if (CURRENT_ENV->dtrace_alloc_probes()) { |
|
321 assert(obj == rax, "must be"); |
|
322 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); |
|
323 } |
|
324 |
|
325 verify_oop(obj); |
|
326 } |
|
327 |
|
328 |
|
329 |
|
330 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { |
|
331 verify_oop(receiver); |
|
332 // explicit NULL check not needed since load from [klass_offset] causes a trap |
|
333 // check against inline cache |
|
334 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check"); |
|
335 int start_offset = offset(); |
|
336 |
|
337 if (UseCompressedClassPointers) { |
|
338 load_klass(rscratch1, receiver); |
|
339 cmpptr(rscratch1, iCache); |
|
340 } else { |
|
341 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes())); |
|
342 } |
|
343 // if icache check fails, then jump to runtime routine |
|
344 // Note: RECEIVER must still contain the receiver! |
|
345 jump_cc(Assembler::notEqual, |
|
346 RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
|
347 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9); |
|
348 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry"); |
|
349 } |
|
350 |
|
351 |
|
352 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) { |
|
353 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect"); |
|
354 // Make sure there is enough stack space for this method's activation. |
|
355 // Note that we do this before doing an enter(). This matches the |
|
356 // ordering of C2's stack overflow check / rsp decrement and allows |
|
357 // the SharedRuntime stack overflow handling to be consistent |
|
358 // between the two compilers. |
|
359 generate_stack_overflow_check(bang_size_in_bytes); |
|
360 |
|
361 push(rbp); |
|
362 #ifdef TIERED |
|
363 // c2 leaves fpu stack dirty. Clean it on entry |
|
364 if (UseSSE < 2 ) { |
|
365 empty_FPU_stack(); |
|
366 } |
|
367 #endif // TIERED |
|
368 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0 |
|
369 } |
|
370 |
|
371 |
|
372 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) { |
|
373 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0 |
|
374 pop(rbp); |
|
375 } |
|
376 |
|
377 |
|
378 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { |
|
379 if (C1Breakpoint) int3(); |
|
380 inline_cache_check(receiver, ic_klass); |
|
381 } |
|
382 |
|
383 |
|
384 void C1_MacroAssembler::verified_entry() { |
|
385 if (C1Breakpoint || VerifyFPU || !UseStackBanging) { |
|
386 // Verified Entry first instruction should be 5 bytes long for correct |
|
387 // patching by patch_verified_entry(). |
|
388 // |
|
389 // C1Breakpoint and VerifyFPU have one byte first instruction. |
|
390 // Also first instruction will be one byte "push(rbp)" if stack banging |
|
391 // code is not generated (see build_frame() above). |
|
392 // For all these cases generate long instruction first. |
|
393 fat_nop(); |
|
394 } |
|
395 if (C1Breakpoint)int3(); |
|
396 // build frame |
|
397 verify_FPU(0, "method_entry"); |
|
398 } |
|
399 |
|
400 |
|
401 #ifndef PRODUCT |
|
402 |
|
403 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { |
|
404 if (!VerifyOops) return; |
|
405 verify_oop_addr(Address(rsp, stack_offset)); |
|
406 } |
|
407 |
|
408 void C1_MacroAssembler::verify_not_null_oop(Register r) { |
|
409 if (!VerifyOops) return; |
|
410 Label not_null; |
|
411 testptr(r, r); |
|
412 jcc(Assembler::notZero, not_null); |
|
413 stop("non-null oop required"); |
|
414 bind(not_null); |
|
415 verify_oop(r); |
|
416 } |
|
417 |
|
418 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) { |
|
419 #ifdef ASSERT |
|
420 if (inv_rax) movptr(rax, 0xDEAD); |
|
421 if (inv_rbx) movptr(rbx, 0xDEAD); |
|
422 if (inv_rcx) movptr(rcx, 0xDEAD); |
|
423 if (inv_rdx) movptr(rdx, 0xDEAD); |
|
424 if (inv_rsi) movptr(rsi, 0xDEAD); |
|
425 if (inv_rdi) movptr(rdi, 0xDEAD); |
|
426 #endif |
|
427 } |
|
428 |
|
429 #endif // ifndef PRODUCT |