42065
|
1 |
/*
|
|
2 |
* Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
|
|
3 |
* Copyright (c) 2016 SAP SE. 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::inline_cache_check(Register receiver, Register iCache) {
|
|
41 |
Label ic_miss, ic_hit;
|
|
42 |
verify_oop(receiver);
|
|
43 |
int klass_offset = oopDesc::klass_offset_in_bytes();
|
|
44 |
|
|
45 |
if (!ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(klass_offset)) {
|
|
46 |
if (VM_Version::has_CompareBranch()) {
|
|
47 |
z_cgij(receiver, 0, Assembler::bcondEqual, ic_miss);
|
|
48 |
} else {
|
|
49 |
z_ltgr(receiver, receiver);
|
|
50 |
z_bre(ic_miss);
|
|
51 |
}
|
|
52 |
}
|
|
53 |
|
|
54 |
compare_klass_ptr(iCache, klass_offset, receiver, false);
|
|
55 |
z_bre(ic_hit);
|
|
56 |
|
|
57 |
// If icache check fails, then jump to runtime routine.
|
|
58 |
// Note: RECEIVER must still contain the receiver!
|
|
59 |
load_const_optimized(Z_R1_scratch, AddressLiteral(SharedRuntime::get_ic_miss_stub()));
|
|
60 |
z_br(Z_R1_scratch);
|
|
61 |
align(CodeEntryAlignment);
|
|
62 |
bind(ic_hit);
|
|
63 |
}
|
|
64 |
|
|
65 |
void C1_MacroAssembler::explicit_null_check(Register base) {
|
|
66 |
ShouldNotCallThis(); // unused
|
|
67 |
}
|
|
68 |
|
|
69 |
void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
|
|
70 |
assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
|
|
71 |
generate_stack_overflow_check(bang_size_in_bytes);
|
|
72 |
save_return_pc();
|
|
73 |
push_frame(frame_size_in_bytes); // TODO: Must we add z_abi_160?
|
|
74 |
}
|
|
75 |
|
|
76 |
void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
|
|
77 |
ShouldNotCallThis(); // unused
|
|
78 |
}
|
|
79 |
|
|
80 |
void C1_MacroAssembler::verified_entry() {
|
|
81 |
if (C1Breakpoint) z_illtrap(0xC1);
|
|
82 |
}
|
|
83 |
|
|
84 |
void C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
|
|
85 |
const int hdr_offset = oopDesc::mark_offset_in_bytes();
|
|
86 |
assert_different_registers(hdr, obj, disp_hdr);
|
|
87 |
NearLabel done;
|
|
88 |
|
|
89 |
verify_oop(obj);
|
|
90 |
|
|
91 |
// Load object header.
|
|
92 |
z_lg(hdr, Address(obj, hdr_offset));
|
|
93 |
|
|
94 |
// Save object being locked into the BasicObjectLock...
|
|
95 |
z_stg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
|
|
96 |
|
|
97 |
if (UseBiasedLocking) {
|
|
98 |
biased_locking_enter(obj, hdr, Z_R1_scratch, Z_R0_scratch, done, &slow_case);
|
|
99 |
}
|
|
100 |
|
|
101 |
// and mark it as unlocked.
|
|
102 |
z_oill(hdr, markOopDesc::unlocked_value);
|
|
103 |
// Save unlocked object header into the displaced header location on the stack.
|
|
104 |
z_stg(hdr, Address(disp_hdr, (intptr_t)0));
|
|
105 |
// Test if object header is still the same (i.e. unlocked), and if so, store the
|
|
106 |
// displaced header address in the object header. If it is not the same, get the
|
|
107 |
// object header instead.
|
|
108 |
z_csg(hdr, disp_hdr, hdr_offset, obj);
|
|
109 |
// If the object header was the same, we're done.
|
|
110 |
if (PrintBiasedLockingStatistics) {
|
|
111 |
Unimplemented();
|
|
112 |
#if 0
|
|
113 |
cond_inc32(Assembler::equal,
|
|
114 |
ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
|
|
115 |
#endif
|
|
116 |
}
|
|
117 |
branch_optimized(Assembler::bcondEqual, done);
|
|
118 |
// If the object header was not the same, it is now in the hdr register.
|
|
119 |
// => Test if it is a stack pointer into the same stack (recursive locking), i.e.:
|
|
120 |
//
|
|
121 |
// 1) (hdr & markOopDesc::lock_mask_in_place) == 0
|
|
122 |
// 2) rsp <= hdr
|
|
123 |
// 3) hdr <= rsp + page_size
|
|
124 |
//
|
|
125 |
// These 3 tests can be done by evaluating the following expression:
|
|
126 |
//
|
|
127 |
// (hdr - Z_SP) & (~(page_size-1) | markOopDesc::lock_mask_in_place)
|
|
128 |
//
|
|
129 |
// assuming both the stack pointer and page_size have their least
|
|
130 |
// significant 2 bits cleared and page_size is a power of 2
|
|
131 |
z_sgr(hdr, Z_SP);
|
|
132 |
|
|
133 |
load_const_optimized(Z_R0_scratch, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place));
|
|
134 |
z_ngr(hdr, Z_R0_scratch); // AND sets CC (result eq/ne 0).
|
|
135 |
// For recursive locking, the result is zero. => Save it in the displaced header
|
|
136 |
// location (NULL in the displaced hdr location indicates recursive locking).
|
|
137 |
z_stg(hdr, Address(disp_hdr, (intptr_t)0));
|
|
138 |
// Otherwise we don't care about the result and handle locking via runtime call.
|
|
139 |
branch_optimized(Assembler::bcondNotZero, slow_case);
|
|
140 |
// done
|
|
141 |
bind(done);
|
|
142 |
}
|
|
143 |
|
|
144 |
void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
|
|
145 |
const int aligned_mask = BytesPerWord -1;
|
|
146 |
const int hdr_offset = oopDesc::mark_offset_in_bytes();
|
|
147 |
assert_different_registers(hdr, obj, disp_hdr);
|
|
148 |
NearLabel done;
|
|
149 |
|
|
150 |
if (UseBiasedLocking) {
|
|
151 |
// Load object.
|
|
152 |
z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
|
|
153 |
biased_locking_exit(obj, hdr, done);
|
|
154 |
}
|
|
155 |
|
|
156 |
// Load displaced header.
|
|
157 |
z_ltg(hdr, Address(disp_hdr, (intptr_t)0));
|
|
158 |
// If the loaded hdr is NULL we had recursive locking, and we are done.
|
|
159 |
z_bre(done);
|
|
160 |
if (!UseBiasedLocking) {
|
|
161 |
// Load object.
|
|
162 |
z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
|
|
163 |
}
|
|
164 |
verify_oop(obj);
|
|
165 |
// Test if object header is pointing to the displaced header, and if so, restore
|
|
166 |
// the displaced header in the object. If the object header is not pointing to
|
|
167 |
// the displaced header, get the object header instead.
|
|
168 |
z_csg(disp_hdr, hdr, hdr_offset, obj);
|
|
169 |
// If the object header was not pointing to the displaced header,
|
|
170 |
// we do unlocking via runtime call.
|
|
171 |
branch_optimized(Assembler::bcondNotEqual, slow_case);
|
|
172 |
// done
|
|
173 |
bind(done);
|
|
174 |
}
|
|
175 |
|
|
176 |
void C1_MacroAssembler::try_allocate(
|
|
177 |
Register obj, // result: Pointer to object after successful allocation.
|
|
178 |
Register var_size_in_bytes, // Object size in bytes if unknown at compile time; invalid otherwise.
|
|
179 |
int con_size_in_bytes, // Object size in bytes if known at compile time.
|
|
180 |
Register t1, // Temp register: Must be global register for incr_allocated_bytes.
|
|
181 |
Label& slow_case // Continuation point if fast allocation fails.
|
|
182 |
) {
|
|
183 |
if (UseTLAB) {
|
|
184 |
tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
|
|
185 |
} else {
|
|
186 |
// Allocation in shared Eden not implemented, because sapjvm allocation trace does not allow it.
|
|
187 |
z_brul(slow_case);
|
|
188 |
}
|
|
189 |
}
|
|
190 |
|
|
191 |
void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register Rzero, Register t1) {
|
|
192 |
assert_different_registers(obj, klass, len, t1, Rzero);
|
|
193 |
if (UseBiasedLocking && !len->is_valid()) {
|
|
194 |
assert_different_registers(obj, klass, len, t1);
|
|
195 |
z_lg(t1, Address(klass, Klass::prototype_header_offset()));
|
|
196 |
} else {
|
|
197 |
// This assumes that all prototype bits fit in an int32_t.
|
|
198 |
load_const_optimized(t1, (intx)markOopDesc::prototype());
|
|
199 |
}
|
|
200 |
z_stg(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
|
|
201 |
|
|
202 |
if (len->is_valid()) {
|
|
203 |
// Length will be in the klass gap, if one exists.
|
|
204 |
z_st(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
|
|
205 |
} else if (UseCompressedClassPointers) {
|
|
206 |
store_klass_gap(Rzero, obj); // Zero klass gap for compressed oops.
|
|
207 |
}
|
|
208 |
store_klass(klass, obj, t1);
|
|
209 |
}
|
|
210 |
|
|
211 |
void C1_MacroAssembler::initialize_body(Register objectFields, Register len_in_bytes, Register Rzero) {
|
|
212 |
Label done;
|
|
213 |
assert_different_registers(objectFields, len_in_bytes, Rzero);
|
|
214 |
|
|
215 |
// Initialize object fields.
|
|
216 |
// See documentation for MVCLE instruction!!!
|
|
217 |
assert(objectFields->encoding()%2==0, "objectFields must be an even register");
|
|
218 |
assert(len_in_bytes->encoding() == (objectFields->encoding()+1), "objectFields and len_in_bytes must be a register pair");
|
|
219 |
assert(Rzero->encoding()%2==1, "Rzero must be an odd register");
|
|
220 |
|
|
221 |
// Use Rzero as src length, then mvcle will copy nothing
|
|
222 |
// and fill the object with the padding value 0.
|
|
223 |
move_long_ext(objectFields, as_Register(Rzero->encoding()-1), 0);
|
|
224 |
bind(done);
|
|
225 |
}
|
|
226 |
|
|
227 |
void C1_MacroAssembler::allocate_object(
|
|
228 |
Register obj, // Result: pointer to object after successful allocation.
|
|
229 |
Register t1, // temp register
|
|
230 |
Register t2, // temp register: Must be a global register for try_allocate.
|
|
231 |
int hdr_size, // object header size in words
|
|
232 |
int obj_size, // object size in words
|
|
233 |
Register klass, // object klass
|
|
234 |
Label& slow_case // Continuation point if fast allocation fails.
|
|
235 |
) {
|
|
236 |
assert_different_registers(obj, t1, t2, klass);
|
|
237 |
|
|
238 |
// Allocate space and initialize header.
|
|
239 |
try_allocate(obj, noreg, obj_size * wordSize, t1, slow_case);
|
|
240 |
|
|
241 |
initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
|
|
242 |
}
|
|
243 |
|
|
244 |
void C1_MacroAssembler::initialize_object(
|
|
245 |
Register obj, // result: Pointer to object after successful allocation.
|
|
246 |
Register klass, // object klass
|
|
247 |
Register var_size_in_bytes, // Object size in bytes if unknown at compile time; invalid otherwise.
|
|
248 |
int con_size_in_bytes, // Object size in bytes if known at compile time.
|
|
249 |
Register t1, // temp register
|
|
250 |
Register t2 // temp register
|
|
251 |
) {
|
|
252 |
assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
|
|
253 |
"con_size_in_bytes is not multiple of alignment");
|
|
254 |
assert(var_size_in_bytes == noreg, "not implemented");
|
|
255 |
const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
|
|
256 |
|
|
257 |
const Register Rzero = t2;
|
|
258 |
|
|
259 |
z_xgr(Rzero, Rzero);
|
|
260 |
initialize_header(obj, klass, noreg, Rzero, t1);
|
|
261 |
|
|
262 |
// Clear rest of allocated space.
|
|
263 |
const int threshold = 4 * BytesPerWord;
|
|
264 |
if (con_size_in_bytes <= threshold) {
|
|
265 |
// Use explicit null stores.
|
|
266 |
// code size = 6*n bytes (n = number of fields to clear)
|
|
267 |
for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
|
|
268 |
z_stg(Rzero, Address(obj, i));
|
|
269 |
} else {
|
|
270 |
// Code size generated by initialize_body() is 16.
|
|
271 |
Register object_fields = Z_R0_scratch;
|
|
272 |
Register len_in_bytes = Z_R1_scratch;
|
|
273 |
z_la(object_fields, hdr_size_in_bytes, obj);
|
|
274 |
load_const_optimized(len_in_bytes, con_size_in_bytes - hdr_size_in_bytes);
|
|
275 |
initialize_body(object_fields, len_in_bytes, Rzero);
|
|
276 |
}
|
|
277 |
|
|
278 |
// Dtrace support is unimplemented.
|
|
279 |
// if (CURRENT_ENV->dtrace_alloc_probes()) {
|
|
280 |
// assert(obj == rax, "must be");
|
|
281 |
// call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id)));
|
|
282 |
// }
|
|
283 |
|
|
284 |
verify_oop(obj);
|
|
285 |
}
|
|
286 |
|
|
287 |
void C1_MacroAssembler::allocate_array(
|
|
288 |
Register obj, // result: Pointer to array after successful allocation.
|
|
289 |
Register len, // array length
|
|
290 |
Register t1, // temp register
|
|
291 |
Register t2, // temp register
|
|
292 |
int hdr_size, // object header size in words
|
|
293 |
int elt_size, // element size in bytes
|
|
294 |
Register klass, // object klass
|
|
295 |
Label& slow_case // Continuation point if fast allocation fails.
|
|
296 |
) {
|
|
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 |
compareU64_and_branch(len, (int32_t)max_array_allocation_length, bcondHigh, slow_case);
|
|
304 |
|
|
305 |
// Compute array size.
|
|
306 |
// Note: If 0 <= len <= max_length, len*elt_size + header + alignment is
|
|
307 |
// smaller or equal to the largest integer. Also, since top is always
|
|
308 |
// aligned, we can do the alignment here instead of at the end address
|
|
309 |
// computation.
|
|
310 |
const Register arr_size = t2;
|
|
311 |
switch (elt_size) {
|
|
312 |
case 1: lgr_if_needed(arr_size, len); break;
|
|
313 |
case 2: z_sllg(arr_size, len, 1); break;
|
|
314 |
case 4: z_sllg(arr_size, len, 2); break;
|
|
315 |
case 8: z_sllg(arr_size, len, 3); break;
|
|
316 |
default: ShouldNotReachHere();
|
|
317 |
}
|
|
318 |
add2reg(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask); // Add space for header & alignment.
|
|
319 |
z_nill(arr_size, (~MinObjAlignmentInBytesMask) & 0xffff); // Align array size.
|
|
320 |
|
|
321 |
try_allocate(obj, arr_size, 0, t1, slow_case);
|
|
322 |
|
|
323 |
initialize_header(obj, klass, len, noreg, t1);
|
|
324 |
|
|
325 |
// Clear rest of allocated space.
|
|
326 |
Label done;
|
|
327 |
Register object_fields = t1;
|
|
328 |
Register Rzero = Z_R1_scratch;
|
|
329 |
z_aghi(arr_size, -(hdr_size * BytesPerWord));
|
|
330 |
z_bre(done); // Jump if size of fields is zero.
|
|
331 |
z_la(object_fields, hdr_size * BytesPerWord, obj);
|
|
332 |
z_xgr(Rzero, Rzero);
|
|
333 |
initialize_body(object_fields, arr_size, Rzero);
|
|
334 |
bind(done);
|
|
335 |
|
|
336 |
// Dtrace support is unimplemented.
|
|
337 |
// if (CURRENT_ENV->dtrace_alloc_probes()) {
|
|
338 |
// assert(obj == rax, "must be");
|
|
339 |
// call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id)));
|
|
340 |
// }
|
|
341 |
|
|
342 |
verify_oop(obj);
|
|
343 |
}
|
|
344 |
|
|
345 |
|
|
346 |
#ifndef PRODUCT
|
|
347 |
|
|
348 |
void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
|
|
349 |
Unimplemented();
|
|
350 |
// if (!VerifyOops) return;
|
|
351 |
// verify_oop_addr(Address(SP, stack_offset + STACK_BIAS));
|
|
352 |
}
|
|
353 |
|
|
354 |
void C1_MacroAssembler::verify_not_null_oop(Register r) {
|
|
355 |
if (!VerifyOops) return;
|
|
356 |
NearLabel not_null;
|
|
357 |
compareU64_and_branch(r, (intptr_t)0, bcondNotEqual, not_null);
|
|
358 |
stop("non-null oop required");
|
|
359 |
bind(not_null);
|
|
360 |
verify_oop(r);
|
|
361 |
}
|
|
362 |
|
|
363 |
void C1_MacroAssembler::invalidate_registers(Register preserve1,
|
|
364 |
Register preserve2,
|
|
365 |
Register preserve3) {
|
|
366 |
Register dead_value = noreg;
|
|
367 |
for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
|
|
368 |
Register r = as_Register(i);
|
|
369 |
if (r != preserve1 && r != preserve2 && r != preserve3 && r != Z_SP && r != Z_thread) {
|
|
370 |
if (dead_value == noreg) {
|
|
371 |
load_const_optimized(r, 0xc1dead);
|
|
372 |
dead_value = r;
|
|
373 |
} else {
|
|
374 |
z_lgr(r, dead_value);
|
|
375 |
}
|
|
376 |
}
|
|
377 |
}
|
|
378 |
}
|
|
379 |
|
|
380 |
#endif // !PRODUCT
|