Fri, 07 Jan 2011 10:42:32 -0500
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
Summary: Track allocated bytes in Thread's, update on TLAB retirement and direct allocation in Eden and tenured, add JNI methods for ThreadMXBean.
Reviewed-by: coleenp, kvn, dholmes, ysr
1 /*
2 * Copyright (c) 1997, 2010, 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/assembler.hpp"
27 #include "assembler_x86.inline.hpp"
28 #include "code/vtableStubs.hpp"
29 #include "interp_masm_x86_32.hpp"
30 #include "memory/resourceArea.hpp"
31 #include "oops/instanceKlass.hpp"
32 #include "oops/klassVtable.hpp"
33 #include "runtime/sharedRuntime.hpp"
34 #include "vmreg_x86.inline.hpp"
35 #ifdef COMPILER2
36 #include "opto/runtime.hpp"
37 #endif
39 // machine-dependent part of VtableStubs: create VtableStub of correct size and
40 // initialize its code
42 #define __ masm->
44 #ifndef PRODUCT
45 extern "C" void bad_compiled_vtable_index(JavaThread* thread, oop receiver, int index);
46 #endif
48 // These stubs are used by the compiler only.
49 // Argument registers, which must be preserved:
50 // rcx - receiver (always first argument)
51 // rdx - second argument (if any)
52 // Other registers that might be usable:
53 // rax - inline cache register (is interface for itable stub)
54 // rbx - method (used when calling out to interpreter)
55 // Available now, but may become callee-save at some point:
56 // rsi, rdi
57 // Note that rax and rdx are also used for return values.
58 //
59 VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
60 const int i486_code_length = VtableStub::pd_code_size_limit(true);
61 VtableStub* s = new(i486_code_length) VtableStub(true, vtable_index);
62 ResourceMark rm;
63 CodeBuffer cb(s->entry_point(), i486_code_length);
64 MacroAssembler* masm = new MacroAssembler(&cb);
66 #ifndef PRODUCT
68 if (CountCompiledCalls) {
69 __ incrementl(ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
70 }
71 #endif /* PRODUCT */
73 // get receiver (need to skip return address on top of stack)
74 assert(VtableStub::receiver_location() == rcx->as_VMReg(), "receiver expected in rcx");
76 // get receiver klass
77 address npe_addr = __ pc();
78 __ movptr(rax, Address(rcx, oopDesc::klass_offset_in_bytes()));
79 // compute entry offset (in words)
80 int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
81 #ifndef PRODUCT
82 if (DebugVtables) {
83 Label L;
84 // check offset vs vtable length
85 __ cmpl(Address(rax, instanceKlass::vtable_length_offset()*wordSize), vtable_index*vtableEntry::size());
86 __ jcc(Assembler::greater, L);
87 __ movl(rbx, vtable_index);
88 __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), rcx, rbx);
89 __ bind(L);
90 }
91 #endif // PRODUCT
93 const Register method = rbx;
95 // load methodOop and target address
96 __ movptr(method, Address(rax, entry_offset*wordSize + vtableEntry::method_offset_in_bytes()));
97 if (DebugVtables) {
98 Label L;
99 __ cmpptr(method, (int32_t)NULL_WORD);
100 __ jcc(Assembler::equal, L);
101 __ cmpptr(Address(method, methodOopDesc::from_compiled_offset()), (int32_t)NULL_WORD);
102 __ jcc(Assembler::notZero, L);
103 __ stop("Vtable entry is NULL");
104 __ bind(L);
105 }
107 // rax,: receiver klass
108 // method (rbx): methodOop
109 // rcx: receiver
110 address ame_addr = __ pc();
111 __ jmp( Address(method, methodOopDesc::from_compiled_offset()));
113 masm->flush();
115 if (PrintMiscellaneous && (WizardMode || Verbose)) {
116 tty->print_cr("vtable #%d at "PTR_FORMAT"[%d] left over: %d",
117 vtable_index, s->entry_point(),
118 (int)(s->code_end() - s->entry_point()),
119 (int)(s->code_end() - __ pc()));
120 }
121 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
122 // shut the door on sizing bugs
123 int slop = 3; // 32-bit offset is this much larger than an 8-bit one
124 assert(vtable_index > 10 || __ pc() + slop <= s->code_end(), "room for 32-bit offset");
126 s->set_exception_points(npe_addr, ame_addr);
127 return s;
128 }
131 VtableStub* VtableStubs::create_itable_stub(int itable_index) {
132 // Note well: pd_code_size_limit is the absolute minimum we can get away with. If you
133 // add code here, bump the code stub size returned by pd_code_size_limit!
134 const int i486_code_length = VtableStub::pd_code_size_limit(false);
135 VtableStub* s = new(i486_code_length) VtableStub(false, itable_index);
136 ResourceMark rm;
137 CodeBuffer cb(s->entry_point(), i486_code_length);
138 MacroAssembler* masm = new MacroAssembler(&cb);
140 // Entry arguments:
141 // rax,: Interface
142 // rcx: Receiver
144 #ifndef PRODUCT
145 if (CountCompiledCalls) {
146 __ incrementl(ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
147 }
148 #endif /* PRODUCT */
149 // get receiver (need to skip return address on top of stack)
151 assert(VtableStub::receiver_location() == rcx->as_VMReg(), "receiver expected in rcx");
153 // get receiver klass (also an implicit null-check)
154 address npe_addr = __ pc();
155 __ movptr(rsi, Address(rcx, oopDesc::klass_offset_in_bytes()));
157 // Most registers are in use; we'll use rax, rbx, rsi, rdi
158 // (If we need to make rsi, rdi callee-save, do a push/pop here.)
159 const Register method = rbx;
160 Label throw_icce;
162 // Get methodOop and entrypoint for compiler
163 __ lookup_interface_method(// inputs: rec. class, interface, itable index
164 rsi, rax, itable_index,
165 // outputs: method, scan temp. reg
166 method, rdi,
167 throw_icce);
169 // method (rbx): methodOop
170 // rcx: receiver
172 #ifdef ASSERT
173 if (DebugVtables) {
174 Label L1;
175 __ cmpptr(method, (int32_t)NULL_WORD);
176 __ jcc(Assembler::equal, L1);
177 __ cmpptr(Address(method, methodOopDesc::from_compiled_offset()), (int32_t)NULL_WORD);
178 __ jcc(Assembler::notZero, L1);
179 __ stop("methodOop is null");
180 __ bind(L1);
181 }
182 #endif // ASSERT
184 address ame_addr = __ pc();
185 __ jmp(Address(method, methodOopDesc::from_compiled_offset()));
187 __ bind(throw_icce);
188 __ jump(RuntimeAddress(StubRoutines::throw_IncompatibleClassChangeError_entry()));
189 masm->flush();
191 if (PrintMiscellaneous && (WizardMode || Verbose)) {
192 tty->print_cr("itable #%d at "PTR_FORMAT"[%d] left over: %d",
193 itable_index, s->entry_point(),
194 (int)(s->code_end() - s->entry_point()),
195 (int)(s->code_end() - __ pc()));
196 }
197 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
198 // shut the door on sizing bugs
199 int slop = 3; // 32-bit offset is this much larger than an 8-bit one
200 assert(itable_index > 10 || __ pc() + slop <= s->code_end(), "room for 32-bit offset");
202 s->set_exception_points(npe_addr, ame_addr);
203 return s;
204 }
208 int VtableStub::pd_code_size_limit(bool is_vtable_stub) {
209 if (is_vtable_stub) {
210 // Vtable stub size
211 return (DebugVtables ? 210 : 16) + (CountCompiledCalls ? 6 : 0);
212 } else {
213 // Itable stub size
214 return (DebugVtables ? 256 : 66) + (CountCompiledCalls ? 6 : 0);
215 }
216 // In order to tune these parameters, run the JVM with VM options
217 // +PrintMiscellaneous and +WizardMode to see information about
218 // actual itable stubs. Look for lines like this:
219 // itable #1 at 0x5551212[65] left over: 3
220 // Reduce the constants so that the "left over" number is >=3
221 // for the common cases.
222 // Do not aim at a left-over number of zero, because a
223 // large vtable or itable index (> 16) will require a 32-bit
224 // immediate displacement instead of an 8-bit one.
225 //
226 // The JVM98 app. _202_jess has a megamorphic interface call.
227 // The itable code looks like this:
228 // Decoding VtableStub itbl[1]@1
229 // mov 0x4(%ecx),%esi
230 // mov 0xe8(%esi),%edi
231 // lea 0x130(%esi,%edi,4),%edi
232 // add $0x7,%edi
233 // and $0xfffffff8,%edi
234 // lea 0x4(%esi),%esi
235 // mov (%edi),%ebx
236 // cmp %ebx,%eax
237 // je success
238 // loop:
239 // test %ebx,%ebx
240 // je throw_icce
241 // add $0x8,%edi
242 // mov (%edi),%ebx
243 // cmp %ebx,%eax
244 // jne loop
245 // success:
246 // mov 0x4(%edi),%edi
247 // mov (%esi,%edi,1),%ebx
248 // jmp *0x44(%ebx)
249 // throw_icce:
250 // jmp throw_ICCE_entry
251 }
253 int VtableStub::pd_code_alignment() {
254 return wordSize;
255 }