Thu, 15 Aug 2013 20:04:10 -0400
8003424: Enable Class Data Sharing for CompressedOops
8016729: ObjectAlignmentInBytes=16 now forces the use of heap based compressed oops
8005933: The -Xshare:auto option is ignored for -server
Summary: Move klass metaspace above the heap and support CDS with compressed klass ptrs.
Reviewed-by: coleenp, kvn, mgerdin, tschatzl, stefank
1 /*
2 * Copyright (c) 2003, 2012, 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 */
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "code/vtableStubs.hpp"
28 #include "interp_masm_x86_64.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "oops/instanceKlass.hpp"
31 #include "oops/klassVtable.hpp"
32 #include "runtime/sharedRuntime.hpp"
33 #include "vmreg_x86.inline.hpp"
34 #ifdef COMPILER2
35 #include "opto/runtime.hpp"
36 #endif
38 // machine-dependent part of VtableStubs: create VtableStub of correct size and
39 // initialize its code
41 #define __ masm->
43 #ifndef PRODUCT
44 extern "C" void bad_compiled_vtable_index(JavaThread* thread,
45 oop receiver,
46 int index);
47 #endif
49 VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
50 const int amd64_code_length = VtableStub::pd_code_size_limit(true);
51 VtableStub* s = new(amd64_code_length) VtableStub(true, vtable_index);
52 ResourceMark rm;
53 CodeBuffer cb(s->entry_point(), amd64_code_length);
54 MacroAssembler* masm = new MacroAssembler(&cb);
56 #ifndef PRODUCT
57 if (CountCompiledCalls) {
58 __ incrementl(ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
59 }
60 #endif
62 // get receiver (need to skip return address on top of stack)
63 assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
65 // Free registers (non-args) are rax, rbx
67 // get receiver klass
68 address npe_addr = __ pc();
69 __ load_klass(rax, j_rarg0);
71 #ifndef PRODUCT
72 if (DebugVtables) {
73 Label L;
74 // check offset vs vtable length
75 __ cmpl(Address(rax, InstanceKlass::vtable_length_offset() * wordSize),
76 vtable_index * vtableEntry::size());
77 __ jcc(Assembler::greater, L);
78 __ movl(rbx, vtable_index);
79 __ call_VM(noreg,
80 CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), j_rarg0, rbx);
81 __ bind(L);
82 }
83 #endif // PRODUCT
85 // load Method* and target address
86 const Register method = rbx;
88 __ lookup_virtual_method(rax, vtable_index, method);
90 if (DebugVtables) {
91 Label L;
92 __ cmpptr(method, (int32_t)NULL_WORD);
93 __ jcc(Assembler::equal, L);
94 __ cmpptr(Address(method, Method::from_compiled_offset()), (int32_t)NULL_WORD);
95 __ jcc(Assembler::notZero, L);
96 __ stop("Vtable entry is NULL");
97 __ bind(L);
98 }
99 // rax: receiver klass
100 // rbx: Method*
101 // rcx: receiver
102 address ame_addr = __ pc();
103 __ jmp( Address(rbx, Method::from_compiled_offset()));
105 __ flush();
107 if (PrintMiscellaneous && (WizardMode || Verbose)) {
108 tty->print_cr("vtable #%d at "PTR_FORMAT"[%d] left over: %d",
109 vtable_index, s->entry_point(),
110 (int)(s->code_end() - s->entry_point()),
111 (int)(s->code_end() - __ pc()));
112 }
113 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
114 // shut the door on sizing bugs
115 int slop = 3; // 32-bit offset is this much larger than an 8-bit one
116 assert(vtable_index > 10 || __ pc() + slop <= s->code_end(), "room for 32-bit offset");
118 s->set_exception_points(npe_addr, ame_addr);
119 return s;
120 }
123 VtableStub* VtableStubs::create_itable_stub(int itable_index) {
124 // Note well: pd_code_size_limit is the absolute minimum we can get
125 // away with. If you add code here, bump the code stub size
126 // returned by pd_code_size_limit!
127 const int amd64_code_length = VtableStub::pd_code_size_limit(false);
128 VtableStub* s = new(amd64_code_length) VtableStub(false, itable_index);
129 ResourceMark rm;
130 CodeBuffer cb(s->entry_point(), amd64_code_length);
131 MacroAssembler* masm = new MacroAssembler(&cb);
133 #ifndef PRODUCT
134 if (CountCompiledCalls) {
135 __ incrementl(ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
136 }
137 #endif
139 // Entry arguments:
140 // rax: Interface
141 // j_rarg0: Receiver
143 // Free registers (non-args) are rax (interface), rbx
145 // get receiver (need to skip return address on top of stack)
147 assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
148 // get receiver klass (also an implicit null-check)
149 address npe_addr = __ pc();
151 // Most registers are in use; we'll use rax, rbx, r10, r11
152 // (various calling sequences use r[cd]x, r[sd]i, r[89]; stay away from them)
153 __ load_klass(r10, j_rarg0);
155 // If we take a trap while this arg is on the stack we will not
156 // be able to walk the stack properly. This is not an issue except
157 // when there are mistakes in this assembly code that could generate
158 // a spurious fault. Ask me how I know...
160 const Register method = rbx;
161 Label throw_icce;
163 // Get Method* and entrypoint for compiler
164 __ lookup_interface_method(// inputs: rec. class, interface, itable index
165 r10, rax, itable_index,
166 // outputs: method, scan temp. reg
167 method, r11,
168 throw_icce);
170 // method (rbx): Method*
171 // j_rarg0: receiver
173 #ifdef ASSERT
174 if (DebugVtables) {
175 Label L2;
176 __ cmpptr(method, (int32_t)NULL_WORD);
177 __ jcc(Assembler::equal, L2);
178 __ cmpptr(Address(method, Method::from_compiled_offset()), (int32_t)NULL_WORD);
179 __ jcc(Assembler::notZero, L2);
180 __ stop("compiler entrypoint is null");
181 __ bind(L2);
182 }
183 #endif // ASSERT
185 // rbx: Method*
186 // j_rarg0: receiver
187 address ame_addr = __ pc();
188 __ jmp(Address(method, Method::from_compiled_offset()));
190 __ bind(throw_icce);
191 __ jump(RuntimeAddress(StubRoutines::throw_IncompatibleClassChangeError_entry()));
193 __ flush();
195 if (PrintMiscellaneous && (WizardMode || Verbose)) {
196 tty->print_cr("itable #%d at "PTR_FORMAT"[%d] left over: %d",
197 itable_index, s->entry_point(),
198 (int)(s->code_end() - s->entry_point()),
199 (int)(s->code_end() - __ pc()));
200 }
201 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
202 // shut the door on sizing bugs
203 int slop = 3; // 32-bit offset is this much larger than an 8-bit one
204 assert(itable_index > 10 || __ pc() + slop <= s->code_end(), "room for 32-bit offset");
206 s->set_exception_points(npe_addr, ame_addr);
207 return s;
208 }
210 int VtableStub::pd_code_size_limit(bool is_vtable_stub) {
211 if (is_vtable_stub) {
212 // Vtable stub size
213 return (DebugVtables ? 512 : 24) + (CountCompiledCalls ? 13 : 0) +
214 (UseCompressedKlassPointers ? MacroAssembler::instr_size_for_decode_klass_not_null() : 0);
215 } else {
216 // Itable stub size
217 return (DebugVtables ? 512 : 74) + (CountCompiledCalls ? 13 : 0) +
218 (UseCompressedKlassPointers ? MacroAssembler::instr_size_for_decode_klass_not_null() : 0);
219 }
220 // In order to tune these parameters, run the JVM with VM options
221 // +PrintMiscellaneous and +WizardMode to see information about
222 // actual itable stubs. Look for lines like this:
223 // itable #1 at 0x5551212[71] left over: 3
224 // Reduce the constants so that the "left over" number is >=3
225 // for the common cases.
226 // Do not aim at a left-over number of zero, because a
227 // large vtable or itable index (>= 32) will require a 32-bit
228 // immediate displacement instead of an 8-bit one.
229 //
230 // The JVM98 app. _202_jess has a megamorphic interface call.
231 // The itable code looks like this:
232 // Decoding VtableStub itbl[1]@12
233 // mov 0x8(%rsi),%r10
234 // mov 0x198(%r10),%r11d
235 // lea 0x218(%r10,%r11,8),%r11
236 // lea 0x8(%r10),%r10
237 // mov (%r11),%rbx
238 // cmp %rbx,%rax
239 // je success
240 // loop:
241 // test %rbx,%rbx
242 // je throw_icce
243 // add $0x10,%r11
244 // mov (%r11),%rbx
245 // cmp %rbx,%rax
246 // jne loop
247 // success:
248 // mov 0x8(%r11),%r11d
249 // mov (%r10,%r11,1),%rbx
250 // jmpq *0x60(%rbx)
251 // throw_icce:
252 // jmpq throw_ICCE_entry
253 }
255 int VtableStub::pd_code_alignment() {
256 return wordSize;
257 }