1 /* |
|
2 * Copyright (c) 1997, 2015, 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 "interpreter/interpreter.hpp" |
|
27 #include "oops/constMethod.hpp" |
|
28 #include "oops/method.hpp" |
|
29 #include "runtime/arguments.hpp" |
|
30 #include "runtime/frame.inline.hpp" |
|
31 #include "runtime/synchronizer.hpp" |
|
32 #include "utilities/macros.hpp" |
|
33 |
|
34 // Size of interpreter code. Increase if too small. Interpreter will |
|
35 // fail with a guarantee ("not enough space for interpreter generation"); |
|
36 // if too small. |
|
37 // Run with +PrintInterpreter to get the VM to print out the size. |
|
38 // Max size with JVMTI |
|
39 #ifdef _LP64 |
|
40 // The sethi() instruction generates lots more instructions when shell |
|
41 // stack limit is unlimited, so that's why this is much bigger. |
|
42 int TemplateInterpreter::InterpreterCodeSize = 260 * K; |
|
43 #else |
|
44 int TemplateInterpreter::InterpreterCodeSize = 230 * K; |
|
45 #endif |
|
46 |
|
47 int AbstractInterpreter::BasicType_as_index(BasicType type) { |
|
48 int i = 0; |
|
49 switch (type) { |
|
50 case T_BOOLEAN: i = 0; break; |
|
51 case T_CHAR : i = 1; break; |
|
52 case T_BYTE : i = 2; break; |
|
53 case T_SHORT : i = 3; break; |
|
54 case T_INT : i = 4; break; |
|
55 case T_LONG : i = 5; break; |
|
56 case T_VOID : i = 6; break; |
|
57 case T_FLOAT : i = 7; break; |
|
58 case T_DOUBLE : i = 8; break; |
|
59 case T_OBJECT : i = 9; break; |
|
60 case T_ARRAY : i = 9; break; |
|
61 default : ShouldNotReachHere(); |
|
62 } |
|
63 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); |
|
64 return i; |
|
65 } |
|
66 |
|
67 bool AbstractInterpreter::can_be_compiled(methodHandle m) { |
|
68 // No special entry points that preclude compilation |
|
69 return true; |
|
70 } |
|
71 |
|
72 static int size_activation_helper(int callee_extra_locals, int max_stack, int monitor_size) { |
|
73 |
|
74 // Figure out the size of an interpreter frame (in words) given that we have a fully allocated |
|
75 // expression stack, the callee will have callee_extra_locals (so we can account for |
|
76 // frame extension) and monitor_size for monitors. Basically we need to calculate |
|
77 // this exactly like generate_fixed_frame/generate_compute_interpreter_state. |
|
78 // |
|
79 // |
|
80 // The big complicating thing here is that we must ensure that the stack stays properly |
|
81 // aligned. This would be even uglier if monitor size wasn't modulo what the stack |
|
82 // needs to be aligned for). We are given that the sp (fp) is already aligned by |
|
83 // the caller so we must ensure that it is properly aligned for our callee. |
|
84 // |
|
85 const int rounded_vm_local_words = |
|
86 round_to(frame::interpreter_frame_vm_local_words,WordsPerLong); |
|
87 // callee_locals and max_stack are counts, not the size in frame. |
|
88 const int locals_size = |
|
89 round_to(callee_extra_locals * Interpreter::stackElementWords, WordsPerLong); |
|
90 const int max_stack_words = max_stack * Interpreter::stackElementWords; |
|
91 return (round_to((max_stack_words |
|
92 + rounded_vm_local_words |
|
93 + frame::memory_parameter_word_sp_offset), WordsPerLong) |
|
94 // already rounded |
|
95 + locals_size + monitor_size); |
|
96 } |
|
97 |
|
98 // How much stack a method top interpreter activation needs in words. |
|
99 int AbstractInterpreter::size_top_interpreter_activation(Method* method) { |
|
100 |
|
101 // See call_stub code |
|
102 int call_stub_size = round_to(7 + frame::memory_parameter_word_sp_offset, |
|
103 WordsPerLong); // 7 + register save area |
|
104 |
|
105 // Save space for one monitor to get into the interpreted method in case |
|
106 // the method is synchronized |
|
107 int monitor_size = method->is_synchronized() ? |
|
108 1*frame::interpreter_frame_monitor_size() : 0; |
|
109 return size_activation_helper(method->max_locals(), method->max_stack(), |
|
110 monitor_size) + call_stub_size; |
|
111 } |
|
112 |
|
113 int AbstractInterpreter::size_activation(int max_stack, |
|
114 int temps, |
|
115 int extra_args, |
|
116 int monitors, |
|
117 int callee_params, |
|
118 int callee_locals, |
|
119 bool is_top_frame) { |
|
120 // Note: This calculation must exactly parallel the frame setup |
|
121 // in TemplateInterpreterGenerator::generate_fixed_frame. |
|
122 |
|
123 int monitor_size = monitors * frame::interpreter_frame_monitor_size(); |
|
124 |
|
125 assert(monitor_size == round_to(monitor_size, WordsPerLong), "must align"); |
|
126 |
|
127 // |
|
128 // Note: if you look closely this appears to be doing something much different |
|
129 // than generate_fixed_frame. What is happening is this. On sparc we have to do |
|
130 // this dance with interpreter_sp_adjustment because the window save area would |
|
131 // appear just below the bottom (tos) of the caller's java expression stack. Because |
|
132 // the interpreter want to have the locals completely contiguous generate_fixed_frame |
|
133 // will adjust the caller's sp for the "extra locals" (max_locals - parameter_size). |
|
134 // Now in generate_fixed_frame the extension of the caller's sp happens in the callee. |
|
135 // In this code the opposite occurs the caller adjusts it's own stack base on the callee. |
|
136 // This is mostly ok but it does cause a problem when we get to the initial frame (the oldest) |
|
137 // because the oldest frame would have adjust its callers frame and yet that frame |
|
138 // already exists and isn't part of this array of frames we are unpacking. So at first |
|
139 // glance this would seem to mess up that frame. However Deoptimization::fetch_unroll_info_helper() |
|
140 // will after it calculates all of the frame's on_stack_size()'s will then figure out the |
|
141 // amount to adjust the caller of the initial (oldest) frame and the calculation will all |
|
142 // add up. It does seem like it simpler to account for the adjustment here (and remove the |
|
143 // callee... parameters here). However this would mean that this routine would have to take |
|
144 // the caller frame as input so we could adjust its sp (and set it's interpreter_sp_adjustment) |
|
145 // and run the calling loop in the reverse order. This would also would appear to mean making |
|
146 // this code aware of what the interactions are when that initial caller fram was an osr or |
|
147 // other adapter frame. deoptimization is complicated enough and hard enough to debug that |
|
148 // there is no sense in messing working code. |
|
149 // |
|
150 |
|
151 int rounded_cls = round_to((callee_locals - callee_params), WordsPerLong); |
|
152 assert(rounded_cls == round_to(rounded_cls, WordsPerLong), "must align"); |
|
153 |
|
154 int raw_frame_size = size_activation_helper(rounded_cls, max_stack, monitor_size); |
|
155 |
|
156 return raw_frame_size; |
|
157 } |
|
158 |
|
159 void AbstractInterpreter::layout_activation(Method* method, |
|
160 int tempcount, |
|
161 int popframe_extra_args, |
|
162 int moncount, |
|
163 int caller_actual_parameters, |
|
164 int callee_param_count, |
|
165 int callee_local_count, |
|
166 frame* caller, |
|
167 frame* interpreter_frame, |
|
168 bool is_top_frame, |
|
169 bool is_bottom_frame) { |
|
170 // Set up the following variables: |
|
171 // - Lmethod |
|
172 // - Llocals |
|
173 // - Lmonitors (to the indicated number of monitors) |
|
174 // - Lesp (to the indicated number of temps) |
|
175 // The frame caller on entry is a description of the caller of the |
|
176 // frame we are about to layout. We are guaranteed that we will be |
|
177 // able to fill in a new interpreter frame as its callee (i.e. the |
|
178 // stack space is allocated and the amount was determined by an |
|
179 // earlier call to the size_activation() method). On return caller |
|
180 // while describe the interpreter frame we just layed out. |
|
181 |
|
182 // The skeleton frame must already look like an interpreter frame |
|
183 // even if not fully filled out. |
|
184 assert(interpreter_frame->is_interpreted_frame(), "Must be interpreted frame"); |
|
185 |
|
186 int rounded_vm_local_words = round_to(frame::interpreter_frame_vm_local_words,WordsPerLong); |
|
187 int monitor_size = moncount * frame::interpreter_frame_monitor_size(); |
|
188 assert(monitor_size == round_to(monitor_size, WordsPerLong), "must align"); |
|
189 |
|
190 intptr_t* fp = interpreter_frame->fp(); |
|
191 |
|
192 JavaThread* thread = JavaThread::current(); |
|
193 RegisterMap map(thread, false); |
|
194 // More verification that skeleton frame is properly walkable |
|
195 assert(fp == caller->sp(), "fp must match"); |
|
196 |
|
197 intptr_t* montop = fp - rounded_vm_local_words; |
|
198 |
|
199 // preallocate monitors (cf. __ add_monitor_to_stack) |
|
200 intptr_t* monitors = montop - monitor_size; |
|
201 |
|
202 // preallocate stack space |
|
203 intptr_t* esp = monitors - 1 - |
|
204 (tempcount * Interpreter::stackElementWords) - |
|
205 popframe_extra_args; |
|
206 |
|
207 int local_words = method->max_locals() * Interpreter::stackElementWords; |
|
208 NEEDS_CLEANUP; |
|
209 intptr_t* locals; |
|
210 if (caller->is_interpreted_frame()) { |
|
211 // Can force the locals area to end up properly overlapping the top of the expression stack. |
|
212 intptr_t* Lesp_ptr = caller->interpreter_frame_tos_address() - 1; |
|
213 // Note that this computation means we replace size_of_parameters() values from the caller |
|
214 // interpreter frame's expression stack with our argument locals |
|
215 int parm_words = caller_actual_parameters * Interpreter::stackElementWords; |
|
216 locals = Lesp_ptr + parm_words; |
|
217 int delta = local_words - parm_words; |
|
218 int computed_sp_adjustment = (delta > 0) ? round_to(delta, WordsPerLong) : 0; |
|
219 *interpreter_frame->register_addr(I5_savedSP) = (intptr_t) (fp + computed_sp_adjustment) - STACK_BIAS; |
|
220 if (!is_bottom_frame) { |
|
221 // Llast_SP is set below for the current frame to SP (with the |
|
222 // extra space for the callee's locals). Here we adjust |
|
223 // Llast_SP for the caller's frame, removing the extra space |
|
224 // for the current method's locals. |
|
225 *caller->register_addr(Llast_SP) = *interpreter_frame->register_addr(I5_savedSP); |
|
226 } else { |
|
227 assert(*caller->register_addr(Llast_SP) >= *interpreter_frame->register_addr(I5_savedSP), "strange Llast_SP"); |
|
228 } |
|
229 } else { |
|
230 assert(caller->is_compiled_frame() || caller->is_entry_frame(), "only possible cases"); |
|
231 // Don't have Lesp available; lay out locals block in the caller |
|
232 // adjacent to the register window save area. |
|
233 // |
|
234 // Compiled frames do not allocate a varargs area which is why this if |
|
235 // statement is needed. |
|
236 // |
|
237 if (caller->is_compiled_frame()) { |
|
238 locals = fp + frame::register_save_words + local_words - 1; |
|
239 } else { |
|
240 locals = fp + frame::memory_parameter_word_sp_offset + local_words - 1; |
|
241 } |
|
242 if (!caller->is_entry_frame()) { |
|
243 // Caller wants his own SP back |
|
244 int caller_frame_size = caller->cb()->frame_size(); |
|
245 *interpreter_frame->register_addr(I5_savedSP) = (intptr_t)(caller->fp() - caller_frame_size) - STACK_BIAS; |
|
246 } |
|
247 } |
|
248 if (TraceDeoptimization) { |
|
249 if (caller->is_entry_frame()) { |
|
250 // make sure I5_savedSP and the entry frames notion of saved SP |
|
251 // agree. This assertion duplicate a check in entry frame code |
|
252 // but catches the failure earlier. |
|
253 assert(*caller->register_addr(Lscratch) == *interpreter_frame->register_addr(I5_savedSP), |
|
254 "would change callers SP"); |
|
255 } |
|
256 if (caller->is_entry_frame()) { |
|
257 tty->print("entry "); |
|
258 } |
|
259 if (caller->is_compiled_frame()) { |
|
260 tty->print("compiled "); |
|
261 if (caller->is_deoptimized_frame()) { |
|
262 tty->print("(deopt) "); |
|
263 } |
|
264 } |
|
265 if (caller->is_interpreted_frame()) { |
|
266 tty->print("interpreted "); |
|
267 } |
|
268 tty->print_cr("caller fp=" INTPTR_FORMAT " sp=" INTPTR_FORMAT, p2i(caller->fp()), p2i(caller->sp())); |
|
269 tty->print_cr("save area = " INTPTR_FORMAT ", " INTPTR_FORMAT, p2i(caller->sp()), p2i(caller->sp() + 16)); |
|
270 tty->print_cr("save area = " INTPTR_FORMAT ", " INTPTR_FORMAT, p2i(caller->fp()), p2i(caller->fp() + 16)); |
|
271 tty->print_cr("interpreter fp=" INTPTR_FORMAT ", " INTPTR_FORMAT, p2i(interpreter_frame->fp()), p2i(interpreter_frame->sp())); |
|
272 tty->print_cr("save area = " INTPTR_FORMAT ", " INTPTR_FORMAT, p2i(interpreter_frame->sp()), p2i(interpreter_frame->sp() + 16)); |
|
273 tty->print_cr("save area = " INTPTR_FORMAT ", " INTPTR_FORMAT, p2i(interpreter_frame->fp()), p2i(interpreter_frame->fp() + 16)); |
|
274 tty->print_cr("Llocals = " INTPTR_FORMAT, p2i(locals)); |
|
275 tty->print_cr("Lesp = " INTPTR_FORMAT, p2i(esp)); |
|
276 tty->print_cr("Lmonitors = " INTPTR_FORMAT, p2i(monitors)); |
|
277 } |
|
278 |
|
279 if (method->max_locals() > 0) { |
|
280 assert(locals < caller->sp() || locals >= (caller->sp() + 16), "locals in save area"); |
|
281 assert(locals < caller->fp() || locals > (caller->fp() + 16), "locals in save area"); |
|
282 assert(locals < interpreter_frame->sp() || locals > (interpreter_frame->sp() + 16), "locals in save area"); |
|
283 assert(locals < interpreter_frame->fp() || locals >= (interpreter_frame->fp() + 16), "locals in save area"); |
|
284 } |
|
285 #ifdef _LP64 |
|
286 assert(*interpreter_frame->register_addr(I5_savedSP) & 1, "must be odd"); |
|
287 #endif |
|
288 |
|
289 *interpreter_frame->register_addr(Lmethod) = (intptr_t) method; |
|
290 *interpreter_frame->register_addr(Llocals) = (intptr_t) locals; |
|
291 *interpreter_frame->register_addr(Lmonitors) = (intptr_t) monitors; |
|
292 *interpreter_frame->register_addr(Lesp) = (intptr_t) esp; |
|
293 // Llast_SP will be same as SP as there is no adapter space |
|
294 *interpreter_frame->register_addr(Llast_SP) = (intptr_t) interpreter_frame->sp() - STACK_BIAS; |
|
295 *interpreter_frame->register_addr(LcpoolCache) = (intptr_t) method->constants()->cache(); |
|
296 #ifdef FAST_DISPATCH |
|
297 *interpreter_frame->register_addr(IdispatchTables) = (intptr_t) Interpreter::dispatch_table(); |
|
298 #endif |
|
299 |
|
300 |
|
301 #ifdef ASSERT |
|
302 BasicObjectLock* mp = (BasicObjectLock*)monitors; |
|
303 |
|
304 assert(interpreter_frame->interpreter_frame_method() == method, "method matches"); |
|
305 assert(interpreter_frame->interpreter_frame_local_at(9) == (intptr_t *)((intptr_t)locals - (9 * Interpreter::stackElementSize)), "locals match"); |
|
306 assert(interpreter_frame->interpreter_frame_monitor_end() == mp, "monitor_end matches"); |
|
307 assert(((intptr_t *)interpreter_frame->interpreter_frame_monitor_begin()) == ((intptr_t *)mp)+monitor_size, "monitor_begin matches"); |
|
308 assert(interpreter_frame->interpreter_frame_tos_address()-1 == esp, "esp matches"); |
|
309 |
|
310 // check bounds |
|
311 intptr_t* lo = interpreter_frame->sp() + (frame::memory_parameter_word_sp_offset - 1); |
|
312 intptr_t* hi = interpreter_frame->fp() - rounded_vm_local_words; |
|
313 assert(lo < monitors && montop <= hi, "monitors in bounds"); |
|
314 assert(lo <= esp && esp < monitors, "esp in bounds"); |
|
315 #endif // ASSERT |
|
316 } |
|