0
|
1 /*
|
|
2 * Copyright 2003-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
|
|
20 * CA 95054 USA or visit www.sun.com if you need additional information or
|
|
21 * have any questions.
|
|
22 *
|
|
23 */
|
|
24
|
|
25 #include "incls/_precompiled.incl"
|
|
26 #include "incls/_sharedRuntime_x86_64.cpp.incl"
|
|
27
|
|
28 DeoptimizationBlob *SharedRuntime::_deopt_blob;
|
|
29 #ifdef COMPILER2
|
|
30 UncommonTrapBlob *SharedRuntime::_uncommon_trap_blob;
|
|
31 ExceptionBlob *OptoRuntime::_exception_blob;
|
|
32 #endif // COMPILER2
|
|
33
|
|
34 SafepointBlob *SharedRuntime::_polling_page_safepoint_handler_blob;
|
|
35 SafepointBlob *SharedRuntime::_polling_page_return_handler_blob;
|
|
36 RuntimeStub* SharedRuntime::_wrong_method_blob;
|
|
37 RuntimeStub* SharedRuntime::_ic_miss_blob;
|
|
38 RuntimeStub* SharedRuntime::_resolve_opt_virtual_call_blob;
|
|
39 RuntimeStub* SharedRuntime::_resolve_virtual_call_blob;
|
|
40 RuntimeStub* SharedRuntime::_resolve_static_call_blob;
|
|
41
|
|
42 #define __ masm->
|
|
43
|
|
44 class SimpleRuntimeFrame {
|
|
45
|
|
46 public:
|
|
47
|
|
48 // Most of the runtime stubs have this simple frame layout.
|
|
49 // This class exists to make the layout shared in one place.
|
|
50 // Offsets are for compiler stack slots, which are jints.
|
|
51 enum layout {
|
|
52 // The frame sender code expects that rbp will be in the "natural" place and
|
|
53 // will override any oopMap setting for it. We must therefore force the layout
|
|
54 // so that it agrees with the frame sender code.
|
|
55 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
|
|
56 rbp_off2,
|
|
57 return_off, return_off2,
|
|
58 framesize
|
|
59 };
|
|
60 };
|
|
61
|
|
62 class RegisterSaver {
|
|
63 // Capture info about frame layout. Layout offsets are in jint
|
|
64 // units because compiler frame slots are jints.
|
|
65 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
|
|
66 enum layout {
|
|
67 fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area
|
|
68 xmm_off = fpu_state_off + 160/BytesPerInt, // offset in fxsave save area
|
|
69 DEF_XMM_OFFS(0),
|
|
70 DEF_XMM_OFFS(1),
|
|
71 DEF_XMM_OFFS(2),
|
|
72 DEF_XMM_OFFS(3),
|
|
73 DEF_XMM_OFFS(4),
|
|
74 DEF_XMM_OFFS(5),
|
|
75 DEF_XMM_OFFS(6),
|
|
76 DEF_XMM_OFFS(7),
|
|
77 DEF_XMM_OFFS(8),
|
|
78 DEF_XMM_OFFS(9),
|
|
79 DEF_XMM_OFFS(10),
|
|
80 DEF_XMM_OFFS(11),
|
|
81 DEF_XMM_OFFS(12),
|
|
82 DEF_XMM_OFFS(13),
|
|
83 DEF_XMM_OFFS(14),
|
|
84 DEF_XMM_OFFS(15),
|
|
85 fpu_state_end = fpu_state_off + ((FPUStateSizeInWords-1)*wordSize / BytesPerInt),
|
|
86 fpu_stateH_end,
|
|
87 r15_off, r15H_off,
|
|
88 r14_off, r14H_off,
|
|
89 r13_off, r13H_off,
|
|
90 r12_off, r12H_off,
|
|
91 r11_off, r11H_off,
|
|
92 r10_off, r10H_off,
|
|
93 r9_off, r9H_off,
|
|
94 r8_off, r8H_off,
|
|
95 rdi_off, rdiH_off,
|
|
96 rsi_off, rsiH_off,
|
|
97 ignore_off, ignoreH_off, // extra copy of rbp
|
|
98 rsp_off, rspH_off,
|
|
99 rbx_off, rbxH_off,
|
|
100 rdx_off, rdxH_off,
|
|
101 rcx_off, rcxH_off,
|
|
102 rax_off, raxH_off,
|
|
103 // 16-byte stack alignment fill word: see MacroAssembler::push/pop_IU_state
|
|
104 align_off, alignH_off,
|
|
105 flags_off, flagsH_off,
|
|
106 // The frame sender code expects that rbp will be in the "natural" place and
|
|
107 // will override any oopMap setting for it. We must therefore force the layout
|
|
108 // so that it agrees with the frame sender code.
|
|
109 rbp_off, rbpH_off, // copy of rbp we will restore
|
|
110 return_off, returnH_off, // slot for return address
|
|
111 reg_save_size // size in compiler stack slots
|
|
112 };
|
|
113
|
|
114 public:
|
|
115 static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
|
|
116 static void restore_live_registers(MacroAssembler* masm);
|
|
117
|
|
118 // Offsets into the register save area
|
|
119 // Used by deoptimization when it is managing result register
|
|
120 // values on its own
|
|
121
|
|
122 static int rax_offset_in_bytes(void) { return BytesPerInt * rax_off; }
|
|
123 static int rbx_offset_in_bytes(void) { return BytesPerInt * rbx_off; }
|
|
124 static int xmm0_offset_in_bytes(void) { return BytesPerInt * xmm0_off; }
|
|
125 static int return_offset_in_bytes(void) { return BytesPerInt * return_off; }
|
|
126
|
|
127 // During deoptimization only the result registers need to be restored,
|
|
128 // all the other values have already been extracted.
|
|
129 static void restore_result_registers(MacroAssembler* masm);
|
|
130 };
|
|
131
|
|
132 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
|
|
133
|
|
134 // Always make the frame size 16-byte aligned
|
|
135 int frame_size_in_bytes = round_to(additional_frame_words*wordSize +
|
|
136 reg_save_size*BytesPerInt, 16);
|
|
137 // OopMap frame size is in compiler stack slots (jint's) not bytes or words
|
|
138 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
|
|
139 // The caller will allocate additional_frame_words
|
|
140 int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt;
|
|
141 // CodeBlob frame size is in words.
|
|
142 int frame_size_in_words = frame_size_in_bytes / wordSize;
|
|
143 *total_frame_words = frame_size_in_words;
|
|
144
|
|
145 // Save registers, fpu state, and flags.
|
|
146 // We assume caller has already pushed the return address onto the
|
|
147 // stack, so rsp is 8-byte aligned here.
|
|
148 // We push rpb twice in this sequence because we want the real rbp
|
|
149 // to be under the return like a normal enter.
|
|
150
|
|
151 __ enter(); // rsp becomes 16-byte aligned here
|
|
152 __ push_CPU_state(); // Push a multiple of 16 bytes
|
|
153 if (frame::arg_reg_save_area_bytes != 0) {
|
|
154 // Allocate argument register save area
|
|
155 __ subq(rsp, frame::arg_reg_save_area_bytes);
|
|
156 }
|
|
157
|
|
158 // Set an oopmap for the call site. This oopmap will map all
|
|
159 // oop-registers and debug-info registers as callee-saved. This
|
|
160 // will allow deoptimization at this safepoint to find all possible
|
|
161 // debug-info recordings, as well as let GC find all oops.
|
|
162
|
|
163 OopMapSet *oop_maps = new OopMapSet();
|
|
164 OopMap* map = new OopMap(frame_size_in_slots, 0);
|
|
165 map->set_callee_saved(VMRegImpl::stack2reg( rax_off + additional_frame_slots), rax->as_VMReg());
|
|
166 map->set_callee_saved(VMRegImpl::stack2reg( rcx_off + additional_frame_slots), rcx->as_VMReg());
|
|
167 map->set_callee_saved(VMRegImpl::stack2reg( rdx_off + additional_frame_slots), rdx->as_VMReg());
|
|
168 map->set_callee_saved(VMRegImpl::stack2reg( rbx_off + additional_frame_slots), rbx->as_VMReg());
|
|
169 // rbp location is known implicitly by the frame sender code, needs no oopmap
|
|
170 // and the location where rbp was saved by is ignored
|
|
171 map->set_callee_saved(VMRegImpl::stack2reg( rsi_off + additional_frame_slots), rsi->as_VMReg());
|
|
172 map->set_callee_saved(VMRegImpl::stack2reg( rdi_off + additional_frame_slots), rdi->as_VMReg());
|
|
173 map->set_callee_saved(VMRegImpl::stack2reg( r8_off + additional_frame_slots), r8->as_VMReg());
|
|
174 map->set_callee_saved(VMRegImpl::stack2reg( r9_off + additional_frame_slots), r9->as_VMReg());
|
|
175 map->set_callee_saved(VMRegImpl::stack2reg( r10_off + additional_frame_slots), r10->as_VMReg());
|
|
176 map->set_callee_saved(VMRegImpl::stack2reg( r11_off + additional_frame_slots), r11->as_VMReg());
|
|
177 map->set_callee_saved(VMRegImpl::stack2reg( r12_off + additional_frame_slots), r12->as_VMReg());
|
|
178 map->set_callee_saved(VMRegImpl::stack2reg( r13_off + additional_frame_slots), r13->as_VMReg());
|
|
179 map->set_callee_saved(VMRegImpl::stack2reg( r14_off + additional_frame_slots), r14->as_VMReg());
|
|
180 map->set_callee_saved(VMRegImpl::stack2reg( r15_off + additional_frame_slots), r15->as_VMReg());
|
|
181 map->set_callee_saved(VMRegImpl::stack2reg(xmm0_off + additional_frame_slots), xmm0->as_VMReg());
|
|
182 map->set_callee_saved(VMRegImpl::stack2reg(xmm1_off + additional_frame_slots), xmm1->as_VMReg());
|
|
183 map->set_callee_saved(VMRegImpl::stack2reg(xmm2_off + additional_frame_slots), xmm2->as_VMReg());
|
|
184 map->set_callee_saved(VMRegImpl::stack2reg(xmm3_off + additional_frame_slots), xmm3->as_VMReg());
|
|
185 map->set_callee_saved(VMRegImpl::stack2reg(xmm4_off + additional_frame_slots), xmm4->as_VMReg());
|
|
186 map->set_callee_saved(VMRegImpl::stack2reg(xmm5_off + additional_frame_slots), xmm5->as_VMReg());
|
|
187 map->set_callee_saved(VMRegImpl::stack2reg(xmm6_off + additional_frame_slots), xmm6->as_VMReg());
|
|
188 map->set_callee_saved(VMRegImpl::stack2reg(xmm7_off + additional_frame_slots), xmm7->as_VMReg());
|
|
189 map->set_callee_saved(VMRegImpl::stack2reg(xmm8_off + additional_frame_slots), xmm8->as_VMReg());
|
|
190 map->set_callee_saved(VMRegImpl::stack2reg(xmm9_off + additional_frame_slots), xmm9->as_VMReg());
|
|
191 map->set_callee_saved(VMRegImpl::stack2reg(xmm10_off + additional_frame_slots), xmm10->as_VMReg());
|
|
192 map->set_callee_saved(VMRegImpl::stack2reg(xmm11_off + additional_frame_slots), xmm11->as_VMReg());
|
|
193 map->set_callee_saved(VMRegImpl::stack2reg(xmm12_off + additional_frame_slots), xmm12->as_VMReg());
|
|
194 map->set_callee_saved(VMRegImpl::stack2reg(xmm13_off + additional_frame_slots), xmm13->as_VMReg());
|
|
195 map->set_callee_saved(VMRegImpl::stack2reg(xmm14_off + additional_frame_slots), xmm14->as_VMReg());
|
|
196 map->set_callee_saved(VMRegImpl::stack2reg(xmm15_off + additional_frame_slots), xmm15->as_VMReg());
|
|
197
|
|
198 // %%% These should all be a waste but we'll keep things as they were for now
|
|
199 if (true) {
|
|
200 map->set_callee_saved(VMRegImpl::stack2reg( raxH_off + additional_frame_slots),
|
|
201 rax->as_VMReg()->next());
|
|
202 map->set_callee_saved(VMRegImpl::stack2reg( rcxH_off + additional_frame_slots),
|
|
203 rcx->as_VMReg()->next());
|
|
204 map->set_callee_saved(VMRegImpl::stack2reg( rdxH_off + additional_frame_slots),
|
|
205 rdx->as_VMReg()->next());
|
|
206 map->set_callee_saved(VMRegImpl::stack2reg( rbxH_off + additional_frame_slots),
|
|
207 rbx->as_VMReg()->next());
|
|
208 // rbp location is known implicitly by the frame sender code, needs no oopmap
|
|
209 map->set_callee_saved(VMRegImpl::stack2reg( rsiH_off + additional_frame_slots),
|
|
210 rsi->as_VMReg()->next());
|
|
211 map->set_callee_saved(VMRegImpl::stack2reg( rdiH_off + additional_frame_slots),
|
|
212 rdi->as_VMReg()->next());
|
|
213 map->set_callee_saved(VMRegImpl::stack2reg( r8H_off + additional_frame_slots),
|
|
214 r8->as_VMReg()->next());
|
|
215 map->set_callee_saved(VMRegImpl::stack2reg( r9H_off + additional_frame_slots),
|
|
216 r9->as_VMReg()->next());
|
|
217 map->set_callee_saved(VMRegImpl::stack2reg( r10H_off + additional_frame_slots),
|
|
218 r10->as_VMReg()->next());
|
|
219 map->set_callee_saved(VMRegImpl::stack2reg( r11H_off + additional_frame_slots),
|
|
220 r11->as_VMReg()->next());
|
|
221 map->set_callee_saved(VMRegImpl::stack2reg( r12H_off + additional_frame_slots),
|
|
222 r12->as_VMReg()->next());
|
|
223 map->set_callee_saved(VMRegImpl::stack2reg( r13H_off + additional_frame_slots),
|
|
224 r13->as_VMReg()->next());
|
|
225 map->set_callee_saved(VMRegImpl::stack2reg( r14H_off + additional_frame_slots),
|
|
226 r14->as_VMReg()->next());
|
|
227 map->set_callee_saved(VMRegImpl::stack2reg( r15H_off + additional_frame_slots),
|
|
228 r15->as_VMReg()->next());
|
|
229 map->set_callee_saved(VMRegImpl::stack2reg(xmm0H_off + additional_frame_slots),
|
|
230 xmm0->as_VMReg()->next());
|
|
231 map->set_callee_saved(VMRegImpl::stack2reg(xmm1H_off + additional_frame_slots),
|
|
232 xmm1->as_VMReg()->next());
|
|
233 map->set_callee_saved(VMRegImpl::stack2reg(xmm2H_off + additional_frame_slots),
|
|
234 xmm2->as_VMReg()->next());
|
|
235 map->set_callee_saved(VMRegImpl::stack2reg(xmm3H_off + additional_frame_slots),
|
|
236 xmm3->as_VMReg()->next());
|
|
237 map->set_callee_saved(VMRegImpl::stack2reg(xmm4H_off + additional_frame_slots),
|
|
238 xmm4->as_VMReg()->next());
|
|
239 map->set_callee_saved(VMRegImpl::stack2reg(xmm5H_off + additional_frame_slots),
|
|
240 xmm5->as_VMReg()->next());
|
|
241 map->set_callee_saved(VMRegImpl::stack2reg(xmm6H_off + additional_frame_slots),
|
|
242 xmm6->as_VMReg()->next());
|
|
243 map->set_callee_saved(VMRegImpl::stack2reg(xmm7H_off + additional_frame_slots),
|
|
244 xmm7->as_VMReg()->next());
|
|
245 map->set_callee_saved(VMRegImpl::stack2reg(xmm8H_off + additional_frame_slots),
|
|
246 xmm8->as_VMReg()->next());
|
|
247 map->set_callee_saved(VMRegImpl::stack2reg(xmm9H_off + additional_frame_slots),
|
|
248 xmm9->as_VMReg()->next());
|
|
249 map->set_callee_saved(VMRegImpl::stack2reg(xmm10H_off + additional_frame_slots),
|
|
250 xmm10->as_VMReg()->next());
|
|
251 map->set_callee_saved(VMRegImpl::stack2reg(xmm11H_off + additional_frame_slots),
|
|
252 xmm11->as_VMReg()->next());
|
|
253 map->set_callee_saved(VMRegImpl::stack2reg(xmm12H_off + additional_frame_slots),
|
|
254 xmm12->as_VMReg()->next());
|
|
255 map->set_callee_saved(VMRegImpl::stack2reg(xmm13H_off + additional_frame_slots),
|
|
256 xmm13->as_VMReg()->next());
|
|
257 map->set_callee_saved(VMRegImpl::stack2reg(xmm14H_off + additional_frame_slots),
|
|
258 xmm14->as_VMReg()->next());
|
|
259 map->set_callee_saved(VMRegImpl::stack2reg(xmm15H_off + additional_frame_slots),
|
|
260 xmm15->as_VMReg()->next());
|
|
261 }
|
|
262
|
|
263 return map;
|
|
264 }
|
|
265
|
|
266 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
|
|
267 if (frame::arg_reg_save_area_bytes != 0) {
|
|
268 // Pop arg register save area
|
|
269 __ addq(rsp, frame::arg_reg_save_area_bytes);
|
|
270 }
|
|
271 // Recover CPU state
|
|
272 __ pop_CPU_state();
|
|
273 // Get the rbp described implicitly by the calling convention (no oopMap)
|
|
274 __ popq(rbp);
|
|
275 }
|
|
276
|
|
277 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
|
|
278
|
|
279 // Just restore result register. Only used by deoptimization. By
|
|
280 // now any callee save register that needs to be restored to a c2
|
|
281 // caller of the deoptee has been extracted into the vframeArray
|
|
282 // and will be stuffed into the c2i adapter we create for later
|
|
283 // restoration so only result registers need to be restored here.
|
|
284
|
|
285 // Restore fp result register
|
|
286 __ movdbl(xmm0, Address(rsp, xmm0_offset_in_bytes()));
|
|
287 // Restore integer result register
|
|
288 __ movq(rax, Address(rsp, rax_offset_in_bytes()));
|
|
289 // Pop all of the register save are off the stack except the return address
|
|
290 __ addq(rsp, return_offset_in_bytes());
|
|
291 }
|
|
292
|
|
293 // The java_calling_convention describes stack locations as ideal slots on
|
|
294 // a frame with no abi restrictions. Since we must observe abi restrictions
|
|
295 // (like the placement of the register window) the slots must be biased by
|
|
296 // the following value.
|
|
297 static int reg2offset_in(VMReg r) {
|
|
298 // Account for saved rbp and return address
|
|
299 // This should really be in_preserve_stack_slots
|
|
300 return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
|
|
301 }
|
|
302
|
|
303 static int reg2offset_out(VMReg r) {
|
|
304 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
|
|
305 }
|
|
306
|
|
307 // ---------------------------------------------------------------------------
|
|
308 // Read the array of BasicTypes from a signature, and compute where the
|
|
309 // arguments should go. Values in the VMRegPair regs array refer to 4-byte
|
|
310 // quantities. Values less than VMRegImpl::stack0 are registers, those above
|
|
311 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer
|
|
312 // as framesizes are fixed.
|
|
313 // VMRegImpl::stack0 refers to the first slot 0(sp).
|
|
314 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register
|
|
315 // up to RegisterImpl::number_of_registers) are the 64-bit
|
|
316 // integer registers.
|
|
317
|
|
318 // Note: the INPUTS in sig_bt are in units of Java argument words, which are
|
|
319 // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit
|
|
320 // units regardless of build. Of course for i486 there is no 64 bit build
|
|
321
|
|
322 // The Java calling convention is a "shifted" version of the C ABI.
|
|
323 // By skipping the first C ABI register we can call non-static jni methods
|
|
324 // with small numbers of arguments without having to shuffle the arguments
|
|
325 // at all. Since we control the java ABI we ought to at least get some
|
|
326 // advantage out of it.
|
|
327
|
|
328 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
|
|
329 VMRegPair *regs,
|
|
330 int total_args_passed,
|
|
331 int is_outgoing) {
|
|
332
|
|
333 // Create the mapping between argument positions and
|
|
334 // registers.
|
|
335 static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
|
|
336 j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5
|
|
337 };
|
|
338 static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
|
|
339 j_farg0, j_farg1, j_farg2, j_farg3,
|
|
340 j_farg4, j_farg5, j_farg6, j_farg7
|
|
341 };
|
|
342
|
|
343
|
|
344 uint int_args = 0;
|
|
345 uint fp_args = 0;
|
|
346 uint stk_args = 0; // inc by 2 each time
|
|
347
|
|
348 for (int i = 0; i < total_args_passed; i++) {
|
|
349 switch (sig_bt[i]) {
|
|
350 case T_BOOLEAN:
|
|
351 case T_CHAR:
|
|
352 case T_BYTE:
|
|
353 case T_SHORT:
|
|
354 case T_INT:
|
|
355 if (int_args < Argument::n_int_register_parameters_j) {
|
|
356 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
|
|
357 } else {
|
|
358 regs[i].set1(VMRegImpl::stack2reg(stk_args));
|
|
359 stk_args += 2;
|
|
360 }
|
|
361 break;
|
|
362 case T_VOID:
|
|
363 // halves of T_LONG or T_DOUBLE
|
|
364 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
|
|
365 regs[i].set_bad();
|
|
366 break;
|
|
367 case T_LONG:
|
|
368 assert(sig_bt[i + 1] == T_VOID, "expecting half");
|
|
369 // fall through
|
|
370 case T_OBJECT:
|
|
371 case T_ARRAY:
|
|
372 case T_ADDRESS:
|
|
373 if (int_args < Argument::n_int_register_parameters_j) {
|
|
374 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
|
|
375 } else {
|
|
376 regs[i].set2(VMRegImpl::stack2reg(stk_args));
|
|
377 stk_args += 2;
|
|
378 }
|
|
379 break;
|
|
380 case T_FLOAT:
|
|
381 if (fp_args < Argument::n_float_register_parameters_j) {
|
|
382 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
|
|
383 } else {
|
|
384 regs[i].set1(VMRegImpl::stack2reg(stk_args));
|
|
385 stk_args += 2;
|
|
386 }
|
|
387 break;
|
|
388 case T_DOUBLE:
|
|
389 assert(sig_bt[i + 1] == T_VOID, "expecting half");
|
|
390 if (fp_args < Argument::n_float_register_parameters_j) {
|
|
391 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
|
|
392 } else {
|
|
393 regs[i].set2(VMRegImpl::stack2reg(stk_args));
|
|
394 stk_args += 2;
|
|
395 }
|
|
396 break;
|
|
397 default:
|
|
398 ShouldNotReachHere();
|
|
399 break;
|
|
400 }
|
|
401 }
|
|
402
|
|
403 return round_to(stk_args, 2);
|
|
404 }
|
|
405
|
|
406 // Patch the callers callsite with entry to compiled code if it exists.
|
|
407 static void patch_callers_callsite(MacroAssembler *masm) {
|
|
408 Label L;
|
|
409 __ verify_oop(rbx);
|
|
410 __ cmpq(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int)NULL_WORD);
|
|
411 __ jcc(Assembler::equal, L);
|
|
412
|
|
413 // Save the current stack pointer
|
|
414 __ movq(r13, rsp);
|
|
415 // Schedule the branch target address early.
|
|
416 // Call into the VM to patch the caller, then jump to compiled callee
|
|
417 // rax isn't live so capture return address while we easily can
|
|
418 __ movq(rax, Address(rsp, 0));
|
|
419
|
|
420 // align stack so push_CPU_state doesn't fault
|
|
421 __ andq(rsp, -(StackAlignmentInBytes));
|
|
422 __ push_CPU_state();
|
|
423
|
|
424
|
|
425 __ verify_oop(rbx);
|
|
426 // VM needs caller's callsite
|
|
427 // VM needs target method
|
|
428 // This needs to be a long call since we will relocate this adapter to
|
|
429 // the codeBuffer and it may not reach
|
|
430
|
|
431 // Allocate argument register save area
|
|
432 if (frame::arg_reg_save_area_bytes != 0) {
|
|
433 __ subq(rsp, frame::arg_reg_save_area_bytes);
|
|
434 }
|
|
435 __ movq(c_rarg0, rbx);
|
|
436 __ movq(c_rarg1, rax);
|
|
437 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
|
|
438
|
|
439 // De-allocate argument register save area
|
|
440 if (frame::arg_reg_save_area_bytes != 0) {
|
|
441 __ addq(rsp, frame::arg_reg_save_area_bytes);
|
|
442 }
|
|
443
|
|
444 __ pop_CPU_state();
|
|
445 // restore sp
|
|
446 __ movq(rsp, r13);
|
|
447 __ bind(L);
|
|
448 }
|
|
449
|
|
450 // Helper function to put tags in interpreter stack.
|
|
451 static void tag_stack(MacroAssembler *masm, const BasicType sig, int st_off) {
|
|
452 if (TaggedStackInterpreter) {
|
|
453 int tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(0);
|
|
454 if (sig == T_OBJECT || sig == T_ARRAY) {
|
|
455 __ mov64(Address(rsp, tag_offset), frame::TagReference);
|
|
456 } else if (sig == T_LONG || sig == T_DOUBLE) {
|
|
457 int next_tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(1);
|
|
458 __ mov64(Address(rsp, next_tag_offset), frame::TagValue);
|
|
459 __ mov64(Address(rsp, tag_offset), frame::TagValue);
|
|
460 } else {
|
|
461 __ mov64(Address(rsp, tag_offset), frame::TagValue);
|
|
462 }
|
|
463 }
|
|
464 }
|
|
465
|
|
466
|
|
467 static void gen_c2i_adapter(MacroAssembler *masm,
|
|
468 int total_args_passed,
|
|
469 int comp_args_on_stack,
|
|
470 const BasicType *sig_bt,
|
|
471 const VMRegPair *regs,
|
|
472 Label& skip_fixup) {
|
|
473 // Before we get into the guts of the C2I adapter, see if we should be here
|
|
474 // at all. We've come from compiled code and are attempting to jump to the
|
|
475 // interpreter, which means the caller made a static call to get here
|
|
476 // (vcalls always get a compiled target if there is one). Check for a
|
|
477 // compiled target. If there is one, we need to patch the caller's call.
|
|
478 patch_callers_callsite(masm);
|
|
479
|
|
480 __ bind(skip_fixup);
|
|
481
|
|
482 // Since all args are passed on the stack, total_args_passed *
|
|
483 // Interpreter::stackElementSize is the space we need. Plus 1 because
|
|
484 // we also account for the return address location since
|
|
485 // we store it first rather than hold it in rax across all the shuffling
|
|
486
|
|
487 int extraspace = (total_args_passed * Interpreter::stackElementSize()) + wordSize;
|
|
488
|
|
489 // stack is aligned, keep it that way
|
|
490 extraspace = round_to(extraspace, 2*wordSize);
|
|
491
|
|
492 // Get return address
|
|
493 __ popq(rax);
|
|
494
|
|
495 // set senderSP value
|
|
496 __ movq(r13, rsp);
|
|
497
|
|
498 __ subq(rsp, extraspace);
|
|
499
|
|
500 // Store the return address in the expected location
|
|
501 __ movq(Address(rsp, 0), rax);
|
|
502
|
|
503 // Now write the args into the outgoing interpreter space
|
|
504 for (int i = 0; i < total_args_passed; i++) {
|
|
505 if (sig_bt[i] == T_VOID) {
|
|
506 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
|
|
507 continue;
|
|
508 }
|
|
509
|
|
510 // offset to start parameters
|
|
511 int st_off = (total_args_passed - i) * Interpreter::stackElementSize() +
|
|
512 Interpreter::value_offset_in_bytes();
|
|
513 int next_off = st_off - Interpreter::stackElementSize();
|
|
514
|
|
515 // Say 4 args:
|
|
516 // i st_off
|
|
517 // 0 32 T_LONG
|
|
518 // 1 24 T_VOID
|
|
519 // 2 16 T_OBJECT
|
|
520 // 3 8 T_BOOL
|
|
521 // - 0 return address
|
|
522 //
|
|
523 // However to make thing extra confusing. Because we can fit a long/double in
|
|
524 // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
|
|
525 // leaves one slot empty and only stores to a single slot. In this case the
|
|
526 // slot that is occupied is the T_VOID slot. See I said it was confusing.
|
|
527
|
|
528 VMReg r_1 = regs[i].first();
|
|
529 VMReg r_2 = regs[i].second();
|
|
530 if (!r_1->is_valid()) {
|
|
531 assert(!r_2->is_valid(), "");
|
|
532 continue;
|
|
533 }
|
|
534 if (r_1->is_stack()) {
|
|
535 // memory to memory use rax
|
|
536 int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
|
|
537 if (!r_2->is_valid()) {
|
|
538 // sign extend??
|
|
539 __ movl(rax, Address(rsp, ld_off));
|
|
540 __ movq(Address(rsp, st_off), rax);
|
|
541 tag_stack(masm, sig_bt[i], st_off);
|
|
542
|
|
543 } else {
|
|
544
|
|
545 __ movq(rax, Address(rsp, ld_off));
|
|
546
|
|
547 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
|
|
548 // T_DOUBLE and T_LONG use two slots in the interpreter
|
|
549 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
|
|
550 // ld_off == LSW, ld_off+wordSize == MSW
|
|
551 // st_off == MSW, next_off == LSW
|
|
552 __ movq(Address(rsp, next_off), rax);
|
|
553 #ifdef ASSERT
|
|
554 // Overwrite the unused slot with known junk
|
|
555 __ mov64(rax, CONST64(0xdeadffffdeadaaaa));
|
|
556 __ movq(Address(rsp, st_off), rax);
|
|
557 #endif /* ASSERT */
|
|
558 tag_stack(masm, sig_bt[i], next_off);
|
|
559 } else {
|
|
560 __ movq(Address(rsp, st_off), rax);
|
|
561 tag_stack(masm, sig_bt[i], st_off);
|
|
562 }
|
|
563 }
|
|
564 } else if (r_1->is_Register()) {
|
|
565 Register r = r_1->as_Register();
|
|
566 if (!r_2->is_valid()) {
|
|
567 // must be only an int (or less ) so move only 32bits to slot
|
|
568 // why not sign extend??
|
|
569 __ movl(Address(rsp, st_off), r);
|
|
570 tag_stack(masm, sig_bt[i], st_off);
|
|
571 } else {
|
|
572 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
|
|
573 // T_DOUBLE and T_LONG use two slots in the interpreter
|
|
574 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
|
|
575 // long/double in gpr
|
|
576 #ifdef ASSERT
|
|
577 // Overwrite the unused slot with known junk
|
|
578 __ mov64(rax, CONST64(0xdeadffffdeadaaab));
|
|
579 __ movq(Address(rsp, st_off), rax);
|
|
580 #endif /* ASSERT */
|
|
581 __ movq(Address(rsp, next_off), r);
|
|
582 tag_stack(masm, sig_bt[i], next_off);
|
|
583 } else {
|
|
584 __ movq(Address(rsp, st_off), r);
|
|
585 tag_stack(masm, sig_bt[i], st_off);
|
|
586 }
|
|
587 }
|
|
588 } else {
|
|
589 assert(r_1->is_XMMRegister(), "");
|
|
590 if (!r_2->is_valid()) {
|
|
591 // only a float use just part of the slot
|
|
592 __ movflt(Address(rsp, st_off), r_1->as_XMMRegister());
|
|
593 tag_stack(masm, sig_bt[i], st_off);
|
|
594 } else {
|
|
595 #ifdef ASSERT
|
|
596 // Overwrite the unused slot with known junk
|
|
597 __ mov64(rax, CONST64(0xdeadffffdeadaaac));
|
|
598 __ movq(Address(rsp, st_off), rax);
|
|
599 #endif /* ASSERT */
|
|
600 __ movdbl(Address(rsp, next_off), r_1->as_XMMRegister());
|
|
601 tag_stack(masm, sig_bt[i], next_off);
|
|
602 }
|
|
603 }
|
|
604 }
|
|
605
|
|
606 // Schedule the branch target address early.
|
|
607 __ movq(rcx, Address(rbx, in_bytes(methodOopDesc::interpreter_entry_offset())));
|
|
608 __ jmp(rcx);
|
|
609 }
|
|
610
|
|
611 static void gen_i2c_adapter(MacroAssembler *masm,
|
|
612 int total_args_passed,
|
|
613 int comp_args_on_stack,
|
|
614 const BasicType *sig_bt,
|
|
615 const VMRegPair *regs) {
|
|
616
|
|
617 //
|
|
618 // We will only enter here from an interpreted frame and never from after
|
|
619 // passing thru a c2i. Azul allowed this but we do not. If we lose the
|
|
620 // race and use a c2i we will remain interpreted for the race loser(s).
|
|
621 // This removes all sorts of headaches on the x86 side and also eliminates
|
|
622 // the possibility of having c2i -> i2c -> c2i -> ... endless transitions.
|
|
623
|
|
624
|
|
625 // Note: r13 contains the senderSP on entry. We must preserve it since
|
|
626 // we may do a i2c -> c2i transition if we lose a race where compiled
|
|
627 // code goes non-entrant while we get args ready.
|
|
628 // In addition we use r13 to locate all the interpreter args as
|
|
629 // we must align the stack to 16 bytes on an i2c entry else we
|
|
630 // lose alignment we expect in all compiled code and register
|
|
631 // save code can segv when fxsave instructions find improperly
|
|
632 // aligned stack pointer.
|
|
633
|
|
634 __ movq(rax, Address(rsp, 0));
|
|
635
|
|
636 // Cut-out for having no stack args. Since up to 2 int/oop args are passed
|
|
637 // in registers, we will occasionally have no stack args.
|
|
638 int comp_words_on_stack = 0;
|
|
639 if (comp_args_on_stack) {
|
|
640 // Sig words on the stack are greater-than VMRegImpl::stack0. Those in
|
|
641 // registers are below. By subtracting stack0, we either get a negative
|
|
642 // number (all values in registers) or the maximum stack slot accessed.
|
|
643
|
|
644 // Convert 4-byte c2 stack slots to words.
|
|
645 comp_words_on_stack = round_to(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
|
|
646 // Round up to miminum stack alignment, in wordSize
|
|
647 comp_words_on_stack = round_to(comp_words_on_stack, 2);
|
|
648 __ subq(rsp, comp_words_on_stack * wordSize);
|
|
649 }
|
|
650
|
|
651
|
|
652 // Ensure compiled code always sees stack at proper alignment
|
|
653 __ andq(rsp, -16);
|
|
654
|
|
655 // push the return address and misalign the stack that youngest frame always sees
|
|
656 // as far as the placement of the call instruction
|
|
657 __ pushq(rax);
|
|
658
|
|
659 // Will jump to the compiled code just as if compiled code was doing it.
|
|
660 // Pre-load the register-jump target early, to schedule it better.
|
|
661 __ movq(r11, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset())));
|
|
662
|
|
663 // Now generate the shuffle code. Pick up all register args and move the
|
|
664 // rest through the floating point stack top.
|
|
665 for (int i = 0; i < total_args_passed; i++) {
|
|
666 if (sig_bt[i] == T_VOID) {
|
|
667 // Longs and doubles are passed in native word order, but misaligned
|
|
668 // in the 32-bit build.
|
|
669 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
|
|
670 continue;
|
|
671 }
|
|
672
|
|
673 // Pick up 0, 1 or 2 words from SP+offset.
|
|
674
|
|
675 assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
|
|
676 "scrambled load targets?");
|
|
677 // Load in argument order going down.
|
|
678 // int ld_off = (total_args_passed + comp_words_on_stack -i)*wordSize;
|
|
679 // base ld_off on r13 (sender_sp) as the stack alignment makes offsets from rsp
|
|
680 // unpredictable
|
|
681 int ld_off = ((total_args_passed - 1) - i)*Interpreter::stackElementSize();
|
|
682
|
|
683 // Point to interpreter value (vs. tag)
|
|
684 int next_off = ld_off - Interpreter::stackElementSize();
|
|
685 //
|
|
686 //
|
|
687 //
|
|
688 VMReg r_1 = regs[i].first();
|
|
689 VMReg r_2 = regs[i].second();
|
|
690 if (!r_1->is_valid()) {
|
|
691 assert(!r_2->is_valid(), "");
|
|
692 continue;
|
|
693 }
|
|
694 if (r_1->is_stack()) {
|
|
695 // Convert stack slot to an SP offset (+ wordSize to account for return address )
|
|
696 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
|
|
697 if (!r_2->is_valid()) {
|
|
698 // sign extend???
|
|
699 __ movl(rax, Address(r13, ld_off));
|
|
700 __ movq(Address(rsp, st_off), rax);
|
|
701 } else {
|
|
702 //
|
|
703 // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
|
|
704 // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
|
|
705 // So we must adjust where to pick up the data to match the interpreter.
|
|
706 //
|
|
707 // Interpreter local[n] == MSW, local[n+1] == LSW however locals
|
|
708 // are accessed as negative so LSW is at LOW address
|
|
709
|
|
710 // ld_off is MSW so get LSW
|
|
711 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
|
|
712 next_off : ld_off;
|
|
713 __ movq(rax, Address(r13, offset));
|
|
714 // st_off is LSW (i.e. reg.first())
|
|
715 __ movq(Address(rsp, st_off), rax);
|
|
716 }
|
|
717 } else if (r_1->is_Register()) { // Register argument
|
|
718 Register r = r_1->as_Register();
|
|
719 assert(r != rax, "must be different");
|
|
720 if (r_2->is_valid()) {
|
|
721 //
|
|
722 // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
|
|
723 // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
|
|
724 // So we must adjust where to pick up the data to match the interpreter.
|
|
725
|
|
726 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
|
|
727 next_off : ld_off;
|
|
728
|
|
729 // this can be a misaligned move
|
|
730 __ movq(r, Address(r13, offset));
|
|
731 } else {
|
|
732 // sign extend and use a full word?
|
|
733 __ movl(r, Address(r13, ld_off));
|
|
734 }
|
|
735 } else {
|
|
736 if (!r_2->is_valid()) {
|
|
737 __ movflt(r_1->as_XMMRegister(), Address(r13, ld_off));
|
|
738 } else {
|
|
739 __ movdbl(r_1->as_XMMRegister(), Address(r13, next_off));
|
|
740 }
|
|
741 }
|
|
742 }
|
|
743
|
|
744 // 6243940 We might end up in handle_wrong_method if
|
|
745 // the callee is deoptimized as we race thru here. If that
|
|
746 // happens we don't want to take a safepoint because the
|
|
747 // caller frame will look interpreted and arguments are now
|
|
748 // "compiled" so it is much better to make this transition
|
|
749 // invisible to the stack walking code. Unfortunately if
|
|
750 // we try and find the callee by normal means a safepoint
|
|
751 // is possible. So we stash the desired callee in the thread
|
|
752 // and the vm will find there should this case occur.
|
|
753
|
|
754 __ movq(Address(r15_thread, JavaThread::callee_target_offset()), rbx);
|
|
755
|
|
756 // put methodOop where a c2i would expect should we end up there
|
|
757 // only needed becaus eof c2 resolve stubs return methodOop as a result in
|
|
758 // rax
|
|
759 __ movq(rax, rbx);
|
|
760 __ jmp(r11);
|
|
761 }
|
|
762
|
|
763 // ---------------------------------------------------------------
|
|
764 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
|
|
765 int total_args_passed,
|
|
766 int comp_args_on_stack,
|
|
767 const BasicType *sig_bt,
|
|
768 const VMRegPair *regs) {
|
|
769 address i2c_entry = __ pc();
|
|
770
|
|
771 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
|
|
772
|
|
773 // -------------------------------------------------------------------------
|
|
774 // Generate a C2I adapter. On entry we know rbx holds the methodOop during calls
|
|
775 // to the interpreter. The args start out packed in the compiled layout. They
|
|
776 // need to be unpacked into the interpreter layout. This will almost always
|
|
777 // require some stack space. We grow the current (compiled) stack, then repack
|
|
778 // the args. We finally end in a jump to the generic interpreter entry point.
|
|
779 // On exit from the interpreter, the interpreter will restore our SP (lest the
|
|
780 // compiled code, which relys solely on SP and not RBP, get sick).
|
|
781
|
|
782 address c2i_unverified_entry = __ pc();
|
|
783 Label skip_fixup;
|
|
784 Label ok;
|
|
785
|
|
786 Register holder = rax;
|
|
787 Register receiver = j_rarg0;
|
|
788 Register temp = rbx;
|
|
789
|
|
790 {
|
|
791 __ verify_oop(holder);
|
|
792 __ movq(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
|
|
793 __ verify_oop(temp);
|
|
794
|
|
795 __ cmpq(temp, Address(holder, compiledICHolderOopDesc::holder_klass_offset()));
|
|
796 __ movq(rbx, Address(holder, compiledICHolderOopDesc::holder_method_offset()));
|
|
797 __ jcc(Assembler::equal, ok);
|
|
798 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
|
|
799
|
|
800 __ bind(ok);
|
|
801 // Method might have been compiled since the call site was patched to
|
|
802 // interpreted if that is the case treat it as a miss so we can get
|
|
803 // the call site corrected.
|
|
804 __ cmpq(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int)NULL_WORD);
|
|
805 __ jcc(Assembler::equal, skip_fixup);
|
|
806 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
|
|
807 }
|
|
808
|
|
809 address c2i_entry = __ pc();
|
|
810
|
|
811 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
|
|
812
|
|
813 __ flush();
|
|
814 return new AdapterHandlerEntry(i2c_entry, c2i_entry, c2i_unverified_entry);
|
|
815 }
|
|
816
|
|
817 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
|
|
818 VMRegPair *regs,
|
|
819 int total_args_passed) {
|
|
820 // We return the amount of VMRegImpl stack slots we need to reserve for all
|
|
821 // the arguments NOT counting out_preserve_stack_slots.
|
|
822
|
|
823 // NOTE: These arrays will have to change when c1 is ported
|
|
824 #ifdef _WIN64
|
|
825 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
|
|
826 c_rarg0, c_rarg1, c_rarg2, c_rarg3
|
|
827 };
|
|
828 static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
|
|
829 c_farg0, c_farg1, c_farg2, c_farg3
|
|
830 };
|
|
831 #else
|
|
832 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
|
|
833 c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5
|
|
834 };
|
|
835 static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
|
|
836 c_farg0, c_farg1, c_farg2, c_farg3,
|
|
837 c_farg4, c_farg5, c_farg6, c_farg7
|
|
838 };
|
|
839 #endif // _WIN64
|
|
840
|
|
841
|
|
842 uint int_args = 0;
|
|
843 uint fp_args = 0;
|
|
844 uint stk_args = 0; // inc by 2 each time
|
|
845
|
|
846 for (int i = 0; i < total_args_passed; i++) {
|
|
847 switch (sig_bt[i]) {
|
|
848 case T_BOOLEAN:
|
|
849 case T_CHAR:
|
|
850 case T_BYTE:
|
|
851 case T_SHORT:
|
|
852 case T_INT:
|
|
853 if (int_args < Argument::n_int_register_parameters_c) {
|
|
854 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
|
|
855 #ifdef _WIN64
|
|
856 fp_args++;
|
|
857 // Allocate slots for callee to stuff register args the stack.
|
|
858 stk_args += 2;
|
|
859 #endif
|
|
860 } else {
|
|
861 regs[i].set1(VMRegImpl::stack2reg(stk_args));
|
|
862 stk_args += 2;
|
|
863 }
|
|
864 break;
|
|
865 case T_LONG:
|
|
866 assert(sig_bt[i + 1] == T_VOID, "expecting half");
|
|
867 // fall through
|
|
868 case T_OBJECT:
|
|
869 case T_ARRAY:
|
|
870 case T_ADDRESS:
|
|
871 if (int_args < Argument::n_int_register_parameters_c) {
|
|
872 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
|
|
873 #ifdef _WIN64
|
|
874 fp_args++;
|
|
875 stk_args += 2;
|
|
876 #endif
|
|
877 } else {
|
|
878 regs[i].set2(VMRegImpl::stack2reg(stk_args));
|
|
879 stk_args += 2;
|
|
880 }
|
|
881 break;
|
|
882 case T_FLOAT:
|
|
883 if (fp_args < Argument::n_float_register_parameters_c) {
|
|
884 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
|
|
885 #ifdef _WIN64
|
|
886 int_args++;
|
|
887 // Allocate slots for callee to stuff register args the stack.
|
|
888 stk_args += 2;
|
|
889 #endif
|
|
890 } else {
|
|
891 regs[i].set1(VMRegImpl::stack2reg(stk_args));
|
|
892 stk_args += 2;
|
|
893 }
|
|
894 break;
|
|
895 case T_DOUBLE:
|
|
896 assert(sig_bt[i + 1] == T_VOID, "expecting half");
|
|
897 if (fp_args < Argument::n_float_register_parameters_c) {
|
|
898 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
|
|
899 #ifdef _WIN64
|
|
900 int_args++;
|
|
901 // Allocate slots for callee to stuff register args the stack.
|
|
902 stk_args += 2;
|
|
903 #endif
|
|
904 } else {
|
|
905 regs[i].set2(VMRegImpl::stack2reg(stk_args));
|
|
906 stk_args += 2;
|
|
907 }
|
|
908 break;
|
|
909 case T_VOID: // Halves of longs and doubles
|
|
910 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
|
|
911 regs[i].set_bad();
|
|
912 break;
|
|
913 default:
|
|
914 ShouldNotReachHere();
|
|
915 break;
|
|
916 }
|
|
917 }
|
|
918 #ifdef _WIN64
|
|
919 // windows abi requires that we always allocate enough stack space
|
|
920 // for 4 64bit registers to be stored down.
|
|
921 if (stk_args < 8) {
|
|
922 stk_args = 8;
|
|
923 }
|
|
924 #endif // _WIN64
|
|
925
|
|
926 return stk_args;
|
|
927 }
|
|
928
|
|
929 // On 64 bit we will store integer like items to the stack as
|
|
930 // 64 bits items (sparc abi) even though java would only store
|
|
931 // 32bits for a parameter. On 32bit it will simply be 32 bits
|
|
932 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
|
|
933 static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
|
|
934 if (src.first()->is_stack()) {
|
|
935 if (dst.first()->is_stack()) {
|
|
936 // stack to stack
|
|
937 __ movslq(rax, Address(rbp, reg2offset_in(src.first())));
|
|
938 __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
|
|
939 } else {
|
|
940 // stack to reg
|
|
941 __ movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
|
|
942 }
|
|
943 } else if (dst.first()->is_stack()) {
|
|
944 // reg to stack
|
|
945 // Do we really have to sign extend???
|
|
946 // __ movslq(src.first()->as_Register(), src.first()->as_Register());
|
|
947 __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
|
|
948 } else {
|
|
949 // Do we really have to sign extend???
|
|
950 // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
|
|
951 if (dst.first() != src.first()) {
|
|
952 __ movq(dst.first()->as_Register(), src.first()->as_Register());
|
|
953 }
|
|
954 }
|
|
955 }
|
|
956
|
|
957
|
|
958 // An oop arg. Must pass a handle not the oop itself
|
|
959 static void object_move(MacroAssembler* masm,
|
|
960 OopMap* map,
|
|
961 int oop_handle_offset,
|
|
962 int framesize_in_slots,
|
|
963 VMRegPair src,
|
|
964 VMRegPair dst,
|
|
965 bool is_receiver,
|
|
966 int* receiver_offset) {
|
|
967
|
|
968 // must pass a handle. First figure out the location we use as a handle
|
|
969
|
|
970 Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
|
|
971
|
|
972 // See if oop is NULL if it is we need no handle
|
|
973
|
|
974 if (src.first()->is_stack()) {
|
|
975
|
|
976 // Oop is already on the stack as an argument
|
|
977 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
|
|
978 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
|
|
979 if (is_receiver) {
|
|
980 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
|
|
981 }
|
|
982
|
|
983 __ cmpq(Address(rbp, reg2offset_in(src.first())), (int)NULL_WORD);
|
|
984 __ leaq(rHandle, Address(rbp, reg2offset_in(src.first())));
|
|
985 // conditionally move a NULL
|
|
986 __ cmovq(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
|
|
987 } else {
|
|
988
|
|
989 // Oop is in an a register we must store it to the space we reserve
|
|
990 // on the stack for oop_handles and pass a handle if oop is non-NULL
|
|
991
|
|
992 const Register rOop = src.first()->as_Register();
|
|
993 int oop_slot;
|
|
994 if (rOop == j_rarg0)
|
|
995 oop_slot = 0;
|
|
996 else if (rOop == j_rarg1)
|
|
997 oop_slot = 1;
|
|
998 else if (rOop == j_rarg2)
|
|
999 oop_slot = 2;
|
|
1000 else if (rOop == j_rarg3)
|
|
1001 oop_slot = 3;
|
|
1002 else if (rOop == j_rarg4)
|
|
1003 oop_slot = 4;
|
|
1004 else {
|
|
1005 assert(rOop == j_rarg5, "wrong register");
|
|
1006 oop_slot = 5;
|
|
1007 }
|
|
1008
|
|
1009 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
|
|
1010 int offset = oop_slot*VMRegImpl::stack_slot_size;
|
|
1011
|
|
1012 map->set_oop(VMRegImpl::stack2reg(oop_slot));
|
|
1013 // Store oop in handle area, may be NULL
|
|
1014 __ movq(Address(rsp, offset), rOop);
|
|
1015 if (is_receiver) {
|
|
1016 *receiver_offset = offset;
|
|
1017 }
|
|
1018
|
|
1019 __ cmpq(rOop, (int)NULL);
|
|
1020 __ leaq(rHandle, Address(rsp, offset));
|
|
1021 // conditionally move a NULL from the handle area where it was just stored
|
|
1022 __ cmovq(Assembler::equal, rHandle, Address(rsp, offset));
|
|
1023 }
|
|
1024
|
|
1025 // If arg is on the stack then place it otherwise it is already in correct reg.
|
|
1026 if (dst.first()->is_stack()) {
|
|
1027 __ movq(Address(rsp, reg2offset_out(dst.first())), rHandle);
|
|
1028 }
|
|
1029 }
|
|
1030
|
|
1031 // A float arg may have to do float reg int reg conversion
|
|
1032 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
|
|
1033 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
|
|
1034
|
|
1035 // The calling conventions assures us that each VMregpair is either
|
|
1036 // all really one physical register or adjacent stack slots.
|
|
1037 // This greatly simplifies the cases here compared to sparc.
|
|
1038
|
|
1039 if (src.first()->is_stack()) {
|
|
1040 if (dst.first()->is_stack()) {
|
|
1041 __ movl(rax, Address(rbp, reg2offset_in(src.first())));
|
|
1042 __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
|
|
1043 } else {
|
|
1044 // stack to reg
|
|
1045 assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
|
|
1046 __ movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first())));
|
|
1047 }
|
|
1048 } else if (dst.first()->is_stack()) {
|
|
1049 // reg to stack
|
|
1050 assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
|
|
1051 __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
|
|
1052 } else {
|
|
1053 // reg to reg
|
|
1054 // In theory these overlap but the ordering is such that this is likely a nop
|
|
1055 if ( src.first() != dst.first()) {
|
|
1056 __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
|
|
1057 }
|
|
1058 }
|
|
1059 }
|
|
1060
|
|
1061 // A long move
|
|
1062 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
|
|
1063
|
|
1064 // The calling conventions assures us that each VMregpair is either
|
|
1065 // all really one physical register or adjacent stack slots.
|
|
1066 // This greatly simplifies the cases here compared to sparc.
|
|
1067
|
|
1068 if (src.is_single_phys_reg() ) {
|
|
1069 if (dst.is_single_phys_reg()) {
|
|
1070 if (dst.first() != src.first()) {
|
|
1071 __ movq(dst.first()->as_Register(), src.first()->as_Register());
|
|
1072 }
|
|
1073 } else {
|
|
1074 assert(dst.is_single_reg(), "not a stack pair");
|
|
1075 __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
|
|
1076 }
|
|
1077 } else if (dst.is_single_phys_reg()) {
|
|
1078 assert(src.is_single_reg(), "not a stack pair");
|
|
1079 __ movq(dst.first()->as_Register(), Address(rbp, reg2offset_out(src.first())));
|
|
1080 } else {
|
|
1081 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
|
|
1082 __ movq(rax, Address(rbp, reg2offset_in(src.first())));
|
|
1083 __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
|
|
1084 }
|
|
1085 }
|
|
1086
|
|
1087 // A double move
|
|
1088 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
|
|
1089
|
|
1090 // The calling conventions assures us that each VMregpair is either
|
|
1091 // all really one physical register or adjacent stack slots.
|
|
1092 // This greatly simplifies the cases here compared to sparc.
|
|
1093
|
|
1094 if (src.is_single_phys_reg() ) {
|
|
1095 if (dst.is_single_phys_reg()) {
|
|
1096 // In theory these overlap but the ordering is such that this is likely a nop
|
|
1097 if ( src.first() != dst.first()) {
|
|
1098 __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
|
|
1099 }
|
|
1100 } else {
|
|
1101 assert(dst.is_single_reg(), "not a stack pair");
|
|
1102 __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
|
|
1103 }
|
|
1104 } else if (dst.is_single_phys_reg()) {
|
|
1105 assert(src.is_single_reg(), "not a stack pair");
|
|
1106 __ movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_out(src.first())));
|
|
1107 } else {
|
|
1108 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
|
|
1109 __ movq(rax, Address(rbp, reg2offset_in(src.first())));
|
|
1110 __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
|
|
1111 }
|
|
1112 }
|
|
1113
|
|
1114
|
|
1115 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
|
|
1116 // We always ignore the frame_slots arg and just use the space just below frame pointer
|
|
1117 // which by this time is free to use
|
|
1118 switch (ret_type) {
|
|
1119 case T_FLOAT:
|
|
1120 __ movflt(Address(rbp, -wordSize), xmm0);
|
|
1121 break;
|
|
1122 case T_DOUBLE:
|
|
1123 __ movdbl(Address(rbp, -wordSize), xmm0);
|
|
1124 break;
|
|
1125 case T_VOID: break;
|
|
1126 default: {
|
|
1127 __ movq(Address(rbp, -wordSize), rax);
|
|
1128 }
|
|
1129 }
|
|
1130 }
|
|
1131
|
|
1132 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
|
|
1133 // We always ignore the frame_slots arg and just use the space just below frame pointer
|
|
1134 // which by this time is free to use
|
|
1135 switch (ret_type) {
|
|
1136 case T_FLOAT:
|
|
1137 __ movflt(xmm0, Address(rbp, -wordSize));
|
|
1138 break;
|
|
1139 case T_DOUBLE:
|
|
1140 __ movdbl(xmm0, Address(rbp, -wordSize));
|
|
1141 break;
|
|
1142 case T_VOID: break;
|
|
1143 default: {
|
|
1144 __ movq(rax, Address(rbp, -wordSize));
|
|
1145 }
|
|
1146 }
|
|
1147 }
|
|
1148
|
|
1149 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
|
|
1150 for ( int i = first_arg ; i < arg_count ; i++ ) {
|
|
1151 if (args[i].first()->is_Register()) {
|
|
1152 __ pushq(args[i].first()->as_Register());
|
|
1153 } else if (args[i].first()->is_XMMRegister()) {
|
|
1154 __ subq(rsp, 2*wordSize);
|
|
1155 __ movdbl(Address(rsp, 0), args[i].first()->as_XMMRegister());
|
|
1156 }
|
|
1157 }
|
|
1158 }
|
|
1159
|
|
1160 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
|
|
1161 for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
|
|
1162 if (args[i].first()->is_Register()) {
|
|
1163 __ popq(args[i].first()->as_Register());
|
|
1164 } else if (args[i].first()->is_XMMRegister()) {
|
|
1165 __ movdbl(args[i].first()->as_XMMRegister(), Address(rsp, 0));
|
|
1166 __ addq(rsp, 2*wordSize);
|
|
1167 }
|
|
1168 }
|
|
1169 }
|
|
1170
|
|
1171 // ---------------------------------------------------------------------------
|
|
1172 // Generate a native wrapper for a given method. The method takes arguments
|
|
1173 // in the Java compiled code convention, marshals them to the native
|
|
1174 // convention (handlizes oops, etc), transitions to native, makes the call,
|
|
1175 // returns to java state (possibly blocking), unhandlizes any result and
|
|
1176 // returns.
|
|
1177 nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
|
|
1178 methodHandle method,
|
|
1179 int total_in_args,
|
|
1180 int comp_args_on_stack,
|
|
1181 BasicType *in_sig_bt,
|
|
1182 VMRegPair *in_regs,
|
|
1183 BasicType ret_type) {
|
|
1184 // Native nmethod wrappers never take possesion of the oop arguments.
|
|
1185 // So the caller will gc the arguments. The only thing we need an
|
|
1186 // oopMap for is if the call is static
|
|
1187 //
|
|
1188 // An OopMap for lock (and class if static)
|
|
1189 OopMapSet *oop_maps = new OopMapSet();
|
|
1190 intptr_t start = (intptr_t)__ pc();
|
|
1191
|
|
1192 // We have received a description of where all the java arg are located
|
|
1193 // on entry to the wrapper. We need to convert these args to where
|
|
1194 // the jni function will expect them. To figure out where they go
|
|
1195 // we convert the java signature to a C signature by inserting
|
|
1196 // the hidden arguments as arg[0] and possibly arg[1] (static method)
|
|
1197
|
|
1198 int total_c_args = total_in_args + 1;
|
|
1199 if (method->is_static()) {
|
|
1200 total_c_args++;
|
|
1201 }
|
|
1202
|
|
1203 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
|
|
1204 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
|
|
1205
|
|
1206 int argc = 0;
|
|
1207 out_sig_bt[argc++] = T_ADDRESS;
|
|
1208 if (method->is_static()) {
|
|
1209 out_sig_bt[argc++] = T_OBJECT;
|
|
1210 }
|
|
1211
|
|
1212 for (int i = 0; i < total_in_args ; i++ ) {
|
|
1213 out_sig_bt[argc++] = in_sig_bt[i];
|
|
1214 }
|
|
1215
|
|
1216 // Now figure out where the args must be stored and how much stack space
|
|
1217 // they require.
|
|
1218 //
|
|
1219 int out_arg_slots;
|
|
1220 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args);
|
|
1221
|
|
1222 // Compute framesize for the wrapper. We need to handlize all oops in
|
|
1223 // incoming registers
|
|
1224
|
|
1225 // Calculate the total number of stack slots we will need.
|
|
1226
|
|
1227 // First count the abi requirement plus all of the outgoing args
|
|
1228 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
|
|
1229
|
|
1230 // Now the space for the inbound oop handle area
|
|
1231
|
|
1232 int oop_handle_offset = stack_slots;
|
|
1233 stack_slots += 6*VMRegImpl::slots_per_word;
|
|
1234
|
|
1235 // Now any space we need for handlizing a klass if static method
|
|
1236
|
|
1237 int oop_temp_slot_offset = 0;
|
|
1238 int klass_slot_offset = 0;
|
|
1239 int klass_offset = -1;
|
|
1240 int lock_slot_offset = 0;
|
|
1241 bool is_static = false;
|
|
1242
|
|
1243 if (method->is_static()) {
|
|
1244 klass_slot_offset = stack_slots;
|
|
1245 stack_slots += VMRegImpl::slots_per_word;
|
|
1246 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
|
|
1247 is_static = true;
|
|
1248 }
|
|
1249
|
|
1250 // Plus a lock if needed
|
|
1251
|
|
1252 if (method->is_synchronized()) {
|
|
1253 lock_slot_offset = stack_slots;
|
|
1254 stack_slots += VMRegImpl::slots_per_word;
|
|
1255 }
|
|
1256
|
|
1257 // Now a place (+2) to save return values or temp during shuffling
|
|
1258 // + 4 for return address (which we own) and saved rbp
|
|
1259 stack_slots += 6;
|
|
1260
|
|
1261 // Ok The space we have allocated will look like:
|
|
1262 //
|
|
1263 //
|
|
1264 // FP-> | |
|
|
1265 // |---------------------|
|
|
1266 // | 2 slots for moves |
|
|
1267 // |---------------------|
|
|
1268 // | lock box (if sync) |
|
|
1269 // |---------------------| <- lock_slot_offset
|
|
1270 // | klass (if static) |
|
|
1271 // |---------------------| <- klass_slot_offset
|
|
1272 // | oopHandle area |
|
|
1273 // |---------------------| <- oop_handle_offset (6 java arg registers)
|
|
1274 // | outbound memory |
|
|
1275 // | based arguments |
|
|
1276 // | |
|
|
1277 // |---------------------|
|
|
1278 // | |
|
|
1279 // SP-> | out_preserved_slots |
|
|
1280 //
|
|
1281 //
|
|
1282
|
|
1283
|
|
1284 // Now compute actual number of stack words we need rounding to make
|
|
1285 // stack properly aligned.
|
|
1286 stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word);
|
|
1287
|
|
1288 int stack_size = stack_slots * VMRegImpl::stack_slot_size;
|
|
1289
|
|
1290
|
|
1291 // First thing make an ic check to see if we should even be here
|
|
1292
|
|
1293 // We are free to use all registers as temps without saving them and
|
|
1294 // restoring them except rbp. rbp is the only callee save register
|
|
1295 // as far as the interpreter and the compiler(s) are concerned.
|
|
1296
|
|
1297
|
|
1298 const Register ic_reg = rax;
|
|
1299 const Register receiver = j_rarg0;
|
|
1300
|
|
1301 Label ok;
|
|
1302 Label exception_pending;
|
|
1303
|
|
1304 __ verify_oop(receiver);
|
|
1305 __ cmpq(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes()));
|
|
1306 __ jcc(Assembler::equal, ok);
|
|
1307
|
|
1308 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
|
|
1309
|
|
1310 // Verified entry point must be aligned
|
|
1311 __ align(8);
|
|
1312
|
|
1313 __ bind(ok);
|
|
1314
|
|
1315 int vep_offset = ((intptr_t)__ pc()) - start;
|
|
1316
|
|
1317 // The instruction at the verified entry point must be 5 bytes or longer
|
|
1318 // because it can be patched on the fly by make_non_entrant. The stack bang
|
|
1319 // instruction fits that requirement.
|
|
1320
|
|
1321 // Generate stack overflow check
|
|
1322
|
|
1323 if (UseStackBanging) {
|
|
1324 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size());
|
|
1325 } else {
|
|
1326 // need a 5 byte instruction to allow MT safe patching to non-entrant
|
|
1327 __ fat_nop();
|
|
1328 }
|
|
1329
|
|
1330 // Generate a new frame for the wrapper.
|
|
1331 __ enter();
|
|
1332 // -2 because return address is already present and so is saved rbp
|
|
1333 __ subq(rsp, stack_size - 2*wordSize);
|
|
1334
|
|
1335 // Frame is now completed as far as size and linkage.
|
|
1336
|
|
1337 int frame_complete = ((intptr_t)__ pc()) - start;
|
|
1338
|
|
1339 #ifdef ASSERT
|
|
1340 {
|
|
1341 Label L;
|
|
1342 __ movq(rax, rsp);
|
|
1343 __ andq(rax, -16); // must be 16 byte boundry (see amd64 ABI)
|
|
1344 __ cmpq(rax, rsp);
|
|
1345 __ jcc(Assembler::equal, L);
|
|
1346 __ stop("improperly aligned stack");
|
|
1347 __ bind(L);
|
|
1348 }
|
|
1349 #endif /* ASSERT */
|
|
1350
|
|
1351
|
|
1352 // We use r14 as the oop handle for the receiver/klass
|
|
1353 // It is callee save so it survives the call to native
|
|
1354
|
|
1355 const Register oop_handle_reg = r14;
|
|
1356
|
|
1357
|
|
1358
|
|
1359 //
|
|
1360 // We immediately shuffle the arguments so that any vm call we have to
|
|
1361 // make from here on out (sync slow path, jvmti, etc.) we will have
|
|
1362 // captured the oops from our caller and have a valid oopMap for
|
|
1363 // them.
|
|
1364
|
|
1365 // -----------------
|
|
1366 // The Grand Shuffle
|
|
1367
|
|
1368 // The Java calling convention is either equal (linux) or denser (win64) than the
|
|
1369 // c calling convention. However the because of the jni_env argument the c calling
|
|
1370 // convention always has at least one more (and two for static) arguments than Java.
|
|
1371 // Therefore if we move the args from java -> c backwards then we will never have
|
|
1372 // a register->register conflict and we don't have to build a dependency graph
|
|
1373 // and figure out how to break any cycles.
|
|
1374 //
|
|
1375
|
|
1376 // Record esp-based slot for receiver on stack for non-static methods
|
|
1377 int receiver_offset = -1;
|
|
1378
|
|
1379 // This is a trick. We double the stack slots so we can claim
|
|
1380 // the oops in the caller's frame. Since we are sure to have
|
|
1381 // more args than the caller doubling is enough to make
|
|
1382 // sure we can capture all the incoming oop args from the
|
|
1383 // caller.
|
|
1384 //
|
|
1385 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
|
|
1386
|
|
1387 // Mark location of rbp (someday)
|
|
1388 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rbp));
|
|
1389
|
|
1390 // Use eax, ebx as temporaries during any memory-memory moves we have to do
|
|
1391 // All inbound args are referenced based on rbp and all outbound args via rsp.
|
|
1392
|
|
1393
|
|
1394 #ifdef ASSERT
|
|
1395 bool reg_destroyed[RegisterImpl::number_of_registers];
|
|
1396 bool freg_destroyed[XMMRegisterImpl::number_of_registers];
|
|
1397 for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) {
|
|
1398 reg_destroyed[r] = false;
|
|
1399 }
|
|
1400 for ( int f = 0 ; f < XMMRegisterImpl::number_of_registers ; f++ ) {
|
|
1401 freg_destroyed[f] = false;
|
|
1402 }
|
|
1403
|
|
1404 #endif /* ASSERT */
|
|
1405
|
|
1406
|
|
1407 int c_arg = total_c_args - 1;
|
|
1408 for ( int i = total_in_args - 1; i >= 0 ; i--, c_arg-- ) {
|
|
1409 #ifdef ASSERT
|
|
1410 if (in_regs[i].first()->is_Register()) {
|
|
1411 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
|
|
1412 } else if (in_regs[i].first()->is_XMMRegister()) {
|
|
1413 assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!");
|
|
1414 }
|
|
1415 if (out_regs[c_arg].first()->is_Register()) {
|
|
1416 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
|
|
1417 } else if (out_regs[c_arg].first()->is_XMMRegister()) {
|
|
1418 freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true;
|
|
1419 }
|
|
1420 #endif /* ASSERT */
|
|
1421 switch (in_sig_bt[i]) {
|
|
1422 case T_ARRAY:
|
|
1423 case T_OBJECT:
|
|
1424 object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
|
|
1425 ((i == 0) && (!is_static)),
|
|
1426 &receiver_offset);
|
|
1427 break;
|
|
1428 case T_VOID:
|
|
1429 break;
|
|
1430
|
|
1431 case T_FLOAT:
|
|
1432 float_move(masm, in_regs[i], out_regs[c_arg]);
|
|
1433 break;
|
|
1434
|
|
1435 case T_DOUBLE:
|
|
1436 assert( i + 1 < total_in_args &&
|
|
1437 in_sig_bt[i + 1] == T_VOID &&
|
|
1438 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
|
|
1439 double_move(masm, in_regs[i], out_regs[c_arg]);
|
|
1440 break;
|
|
1441
|
|
1442 case T_LONG :
|
|
1443 long_move(masm, in_regs[i], out_regs[c_arg]);
|
|
1444 break;
|
|
1445
|
|
1446 case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
|
|
1447
|
|
1448 default:
|
|
1449 move32_64(masm, in_regs[i], out_regs[c_arg]);
|
|
1450 }
|
|
1451 }
|
|
1452
|
|
1453 // point c_arg at the first arg that is already loaded in case we
|
|
1454 // need to spill before we call out
|
|
1455 c_arg++;
|
|
1456
|
|
1457 // Pre-load a static method's oop into r14. Used both by locking code and
|
|
1458 // the normal JNI call code.
|
|
1459 if (method->is_static()) {
|
|
1460
|
|
1461 // load oop into a register
|
|
1462 __ movoop(oop_handle_reg, JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror()));
|
|
1463
|
|
1464 // Now handlize the static class mirror it's known not-null.
|
|
1465 __ movq(Address(rsp, klass_offset), oop_handle_reg);
|
|
1466 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
|
|
1467
|
|
1468 // Now get the handle
|
|
1469 __ leaq(oop_handle_reg, Address(rsp, klass_offset));
|
|
1470 // store the klass handle as second argument
|
|
1471 __ movq(c_rarg1, oop_handle_reg);
|
|
1472 // and protect the arg if we must spill
|
|
1473 c_arg--;
|
|
1474 }
|
|
1475
|
|
1476 // Change state to native (we save the return address in the thread, since it might not
|
|
1477 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
|
|
1478 // points into the right code segment. It does not have to be the correct return pc.
|
|
1479 // We use the same pc/oopMap repeatedly when we call out
|
|
1480
|
|
1481 intptr_t the_pc = (intptr_t) __ pc();
|
|
1482 oop_maps->add_gc_map(the_pc - start, map);
|
|
1483
|
|
1484 __ set_last_Java_frame(rsp, noreg, (address)the_pc);
|
|
1485
|
|
1486
|
|
1487 // We have all of the arguments setup at this point. We must not touch any register
|
|
1488 // argument registers at this point (what if we save/restore them there are no oop?
|
|
1489
|
|
1490 {
|
|
1491 SkipIfEqual skip(masm, &DTraceMethodProbes, false);
|
|
1492 // protect the args we've loaded
|
|
1493 save_args(masm, total_c_args, c_arg, out_regs);
|
|
1494 __ movoop(c_rarg1, JNIHandles::make_local(method()));
|
|
1495 __ call_VM_leaf(
|
|
1496 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
|
|
1497 r15_thread, c_rarg1);
|
|
1498 restore_args(masm, total_c_args, c_arg, out_regs);
|
|
1499 }
|
|
1500
|
|
1501 // Lock a synchronized method
|
|
1502
|
|
1503 // Register definitions used by locking and unlocking
|
|
1504
|
|
1505 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
|
|
1506 const Register obj_reg = rbx; // Will contain the oop
|
|
1507 const Register lock_reg = r13; // Address of compiler lock object (BasicLock)
|
|
1508 const Register old_hdr = r13; // value of old header at unlock time
|
|
1509
|
|
1510 Label slow_path_lock;
|
|
1511 Label lock_done;
|
|
1512
|
|
1513 if (method->is_synchronized()) {
|
|
1514
|
|
1515
|
|
1516 const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
|
|
1517
|
|
1518 // Get the handle (the 2nd argument)
|
|
1519 __ movq(oop_handle_reg, c_rarg1);
|
|
1520
|
|
1521 // Get address of the box
|
|
1522
|
|
1523 __ leaq(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
|
|
1524
|
|
1525 // Load the oop from the handle
|
|
1526 __ movq(obj_reg, Address(oop_handle_reg, 0));
|
|
1527
|
|
1528 if (UseBiasedLocking) {
|
|
1529 __ biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, lock_done, &slow_path_lock);
|
|
1530 }
|
|
1531
|
|
1532 // Load immediate 1 into swap_reg %rax
|
|
1533 __ movl(swap_reg, 1);
|
|
1534
|
|
1535 // Load (object->mark() | 1) into swap_reg %rax
|
|
1536 __ orq(swap_reg, Address(obj_reg, 0));
|
|
1537
|
|
1538 // Save (object->mark() | 1) into BasicLock's displaced header
|
|
1539 __ movq(Address(lock_reg, mark_word_offset), swap_reg);
|
|
1540
|
|
1541 if (os::is_MP()) {
|
|
1542 __ lock();
|
|
1543 }
|
|
1544
|
|
1545 // src -> dest iff dest == rax else rax <- dest
|
|
1546 __ cmpxchgq(lock_reg, Address(obj_reg, 0));
|
|
1547 __ jcc(Assembler::equal, lock_done);
|
|
1548
|
|
1549 // Hmm should this move to the slow path code area???
|
|
1550
|
|
1551 // Test if the oopMark is an obvious stack pointer, i.e.,
|
|
1552 // 1) (mark & 3) == 0, and
|
|
1553 // 2) rsp <= mark < mark + os::pagesize()
|
|
1554 // These 3 tests can be done by evaluating the following
|
|
1555 // expression: ((mark - rsp) & (3 - os::vm_page_size())),
|
|
1556 // assuming both stack pointer and pagesize have their
|
|
1557 // least significant 2 bits clear.
|
|
1558 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
|
|
1559
|
|
1560 __ subq(swap_reg, rsp);
|
|
1561 __ andq(swap_reg, 3 - os::vm_page_size());
|
|
1562
|
|
1563 // Save the test result, for recursive case, the result is zero
|
|
1564 __ movq(Address(lock_reg, mark_word_offset), swap_reg);
|
|
1565 __ jcc(Assembler::notEqual, slow_path_lock);
|
|
1566
|
|
1567 // Slow path will re-enter here
|
|
1568
|
|
1569 __ bind(lock_done);
|
|
1570 }
|
|
1571
|
|
1572
|
|
1573 // Finally just about ready to make the JNI call
|
|
1574
|
|
1575
|
|
1576 // get JNIEnv* which is first argument to native
|
|
1577
|
|
1578 __ leaq(c_rarg0, Address(r15_thread, in_bytes(JavaThread::jni_environment_offset())));
|
|
1579
|
|
1580 // Now set thread in native
|
|
1581 __ mov64(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native);
|
|
1582
|
|
1583 __ call(RuntimeAddress(method->native_function()));
|
|
1584
|
|
1585 // Either restore the MXCSR register after returning from the JNI Call
|
|
1586 // or verify that it wasn't changed.
|
|
1587 if (RestoreMXCSROnJNICalls) {
|
|
1588 __ ldmxcsr(ExternalAddress(StubRoutines::amd64::mxcsr_std()));
|
|
1589
|
|
1590 }
|
|
1591 else if (CheckJNICalls ) {
|
|
1592 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::amd64::verify_mxcsr_entry())));
|
|
1593 }
|
|
1594
|
|
1595
|
|
1596 // Unpack native results.
|
|
1597 switch (ret_type) {
|
|
1598 case T_BOOLEAN: __ c2bool(rax); break;
|
|
1599 case T_CHAR : __ movzwl(rax, rax); break;
|
|
1600 case T_BYTE : __ sign_extend_byte (rax); break;
|
|
1601 case T_SHORT : __ sign_extend_short(rax); break;
|
|
1602 case T_INT : /* nothing to do */ break;
|
|
1603 case T_DOUBLE :
|
|
1604 case T_FLOAT :
|
|
1605 // Result is in xmm0 we'll save as needed
|
|
1606 break;
|
|
1607 case T_ARRAY: // Really a handle
|
|
1608 case T_OBJECT: // Really a handle
|
|
1609 break; // can't de-handlize until after safepoint check
|
|
1610 case T_VOID: break;
|
|
1611 case T_LONG: break;
|
|
1612 default : ShouldNotReachHere();
|
|
1613 }
|
|
1614
|
|
1615 // Switch thread to "native transition" state before reading the synchronization state.
|
|
1616 // This additional state is necessary because reading and testing the synchronization
|
|
1617 // state is not atomic w.r.t. GC, as this scenario demonstrates:
|
|
1618 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
|
|
1619 // VM thread changes sync state to synchronizing and suspends threads for GC.
|
|
1620 // Thread A is resumed to finish this native method, but doesn't block here since it
|
|
1621 // didn't see any synchronization is progress, and escapes.
|
|
1622 __ mov64(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
|
|
1623
|
|
1624 if(os::is_MP()) {
|
|
1625 if (UseMembar) {
|
|
1626 // Force this write out before the read below
|
|
1627 __ membar(Assembler::Membar_mask_bits(
|
|
1628 Assembler::LoadLoad | Assembler::LoadStore |
|
|
1629 Assembler::StoreLoad | Assembler::StoreStore));
|
|
1630 } else {
|
|
1631 // Write serialization page so VM thread can do a pseudo remote membar.
|
|
1632 // We use the current thread pointer to calculate a thread specific
|
|
1633 // offset to write to within the page. This minimizes bus traffic
|
|
1634 // due to cache line collision.
|
|
1635 __ serialize_memory(r15_thread, rcx);
|
|
1636 }
|
|
1637 }
|
|
1638
|
|
1639
|
|
1640 // check for safepoint operation in progress and/or pending suspend requests
|
|
1641 {
|
|
1642 Label Continue;
|
|
1643
|
|
1644 __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()),
|
|
1645 SafepointSynchronize::_not_synchronized);
|
|
1646
|
|
1647 Label L;
|
|
1648 __ jcc(Assembler::notEqual, L);
|
|
1649 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
|
|
1650 __ jcc(Assembler::equal, Continue);
|
|
1651 __ bind(L);
|
|
1652
|
|
1653 // Don't use call_VM as it will see a possible pending exception and forward it
|
|
1654 // and never return here preventing us from clearing _last_native_pc down below.
|
|
1655 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
|
|
1656 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
|
|
1657 // by hand.
|
|
1658 //
|
|
1659 save_native_result(masm, ret_type, stack_slots);
|
|
1660 __ movq(c_rarg0, r15_thread);
|
|
1661 __ movq(r12, rsp); // remember sp
|
|
1662 __ subq(rsp, frame::arg_reg_save_area_bytes); // windows
|
|
1663 __ andq(rsp, -16); // align stack as required by ABI
|
|
1664 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
|
|
1665 __ movq(rsp, r12); // restore sp
|
|
1666 // Restore any method result value
|
|
1667 restore_native_result(masm, ret_type, stack_slots);
|
|
1668 __ bind(Continue);
|
|
1669 }
|
|
1670
|
|
1671 // change thread state
|
|
1672 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
|
|
1673
|
|
1674 Label reguard;
|
|
1675 Label reguard_done;
|
|
1676 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
|
|
1677 __ jcc(Assembler::equal, reguard);
|
|
1678 __ bind(reguard_done);
|
|
1679
|
|
1680 // native result if any is live
|
|
1681
|
|
1682 // Unlock
|
|
1683 Label unlock_done;
|
|
1684 Label slow_path_unlock;
|
|
1685 if (method->is_synchronized()) {
|
|
1686
|
|
1687 // Get locked oop from the handle we passed to jni
|
|
1688 __ movq(obj_reg, Address(oop_handle_reg, 0));
|
|
1689
|
|
1690 Label done;
|
|
1691
|
|
1692 if (UseBiasedLocking) {
|
|
1693 __ biased_locking_exit(obj_reg, old_hdr, done);
|
|
1694 }
|
|
1695
|
|
1696 // Simple recursive lock?
|
|
1697
|
|
1698 __ cmpq(Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size), (int)NULL_WORD);
|
|
1699 __ jcc(Assembler::equal, done);
|
|
1700
|
|
1701 // Must save rax if if it is live now because cmpxchg must use it
|
|
1702 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
|
|
1703 save_native_result(masm, ret_type, stack_slots);
|
|
1704 }
|
|
1705
|
|
1706
|
|
1707 // get address of the stack lock
|
|
1708 __ leaq(rax, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
|
|
1709 // get old displaced header
|
|
1710 __ movq(old_hdr, Address(rax, 0));
|
|
1711
|
|
1712 // Atomic swap old header if oop still contains the stack lock
|
|
1713 if (os::is_MP()) {
|
|
1714 __ lock();
|
|
1715 }
|
|
1716 __ cmpxchgq(old_hdr, Address(obj_reg, 0));
|
|
1717 __ jcc(Assembler::notEqual, slow_path_unlock);
|
|
1718
|
|
1719 // slow path re-enters here
|
|
1720 __ bind(unlock_done);
|
|
1721 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
|
|
1722 restore_native_result(masm, ret_type, stack_slots);
|
|
1723 }
|
|
1724
|
|
1725 __ bind(done);
|
|
1726
|
|
1727 }
|
|
1728
|
|
1729 {
|
|
1730 SkipIfEqual skip(masm, &DTraceMethodProbes, false);
|
|
1731 save_native_result(masm, ret_type, stack_slots);
|
|
1732 __ movoop(c_rarg1, JNIHandles::make_local(method()));
|
|
1733 __ call_VM_leaf(
|
|
1734 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
|
|
1735 r15_thread, c_rarg1);
|
|
1736 restore_native_result(masm, ret_type, stack_slots);
|
|
1737 }
|
|
1738
|
|
1739 __ reset_last_Java_frame(false, true);
|
|
1740
|
|
1741 // Unpack oop result
|
|
1742 if (ret_type == T_OBJECT || ret_type == T_ARRAY) {
|
|
1743 Label L;
|
|
1744 __ testq(rax, rax);
|
|
1745 __ jcc(Assembler::zero, L);
|
|
1746 __ movq(rax, Address(rax, 0));
|
|
1747 __ bind(L);
|
|
1748 __ verify_oop(rax);
|
|
1749 }
|
|
1750
|
|
1751 // reset handle block
|
|
1752 __ movq(rcx, Address(r15_thread, JavaThread::active_handles_offset()));
|
|
1753 __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), (int)NULL_WORD);
|
|
1754
|
|
1755 // pop our frame
|
|
1756
|
|
1757 __ leave();
|
|
1758
|
|
1759 // Any exception pending?
|
|
1760 __ cmpq(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
|
|
1761 __ jcc(Assembler::notEqual, exception_pending);
|
|
1762
|
|
1763 // Return
|
|
1764
|
|
1765 __ ret(0);
|
|
1766
|
|
1767 // Unexpected paths are out of line and go here
|
|
1768
|
|
1769 // forward the exception
|
|
1770 __ bind(exception_pending);
|
|
1771
|
|
1772 // and forward the exception
|
|
1773 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
|
|
1774
|
|
1775
|
|
1776 // Slow path locking & unlocking
|
|
1777 if (method->is_synchronized()) {
|
|
1778
|
|
1779 // BEGIN Slow path lock
|
|
1780 __ bind(slow_path_lock);
|
|
1781
|
|
1782 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
|
|
1783 // args are (oop obj, BasicLock* lock, JavaThread* thread)
|
|
1784
|
|
1785 // protect the args we've loaded
|
|
1786 save_args(masm, total_c_args, c_arg, out_regs);
|
|
1787
|
|
1788 __ movq(c_rarg0, obj_reg);
|
|
1789 __ movq(c_rarg1, lock_reg);
|
|
1790 __ movq(c_rarg2, r15_thread);
|
|
1791
|
|
1792 // Not a leaf but we have last_Java_frame setup as we want
|
|
1793 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
|
|
1794 restore_args(masm, total_c_args, c_arg, out_regs);
|
|
1795
|
|
1796 #ifdef ASSERT
|
|
1797 { Label L;
|
|
1798 __ cmpq(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
|
|
1799 __ jcc(Assembler::equal, L);
|
|
1800 __ stop("no pending exception allowed on exit from monitorenter");
|
|
1801 __ bind(L);
|
|
1802 }
|
|
1803 #endif
|
|
1804 __ jmp(lock_done);
|
|
1805
|
|
1806 // END Slow path lock
|
|
1807
|
|
1808 // BEGIN Slow path unlock
|
|
1809 __ bind(slow_path_unlock);
|
|
1810
|
|
1811 // If we haven't already saved the native result we must save it now as xmm registers
|
|
1812 // are still exposed.
|
|
1813
|
|
1814 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
|
|
1815 save_native_result(masm, ret_type, stack_slots);
|
|
1816 }
|
|
1817
|
|
1818 __ leaq(c_rarg1, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
|
|
1819
|
|
1820 __ movq(c_rarg0, obj_reg);
|
|
1821 __ movq(r12, rsp); // remember sp
|
|
1822 __ subq(rsp, frame::arg_reg_save_area_bytes); // windows
|
|
1823 __ andq(rsp, -16); // align stack as required by ABI
|
|
1824
|
|
1825 // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
|
|
1826 // NOTE that obj_reg == rbx currently
|
|
1827 __ movq(rbx, Address(r15_thread, in_bytes(Thread::pending_exception_offset())));
|
|
1828 __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
|
|
1829
|
|
1830 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C)));
|
|
1831 __ movq(rsp, r12); // restore sp
|
|
1832 #ifdef ASSERT
|
|
1833 {
|
|
1834 Label L;
|
|
1835 __ cmpq(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
|
|
1836 __ jcc(Assembler::equal, L);
|
|
1837 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
|
|
1838 __ bind(L);
|
|
1839 }
|
|
1840 #endif /* ASSERT */
|
|
1841
|
|
1842 __ movq(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), rbx);
|
|
1843
|
|
1844 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
|
|
1845 restore_native_result(masm, ret_type, stack_slots);
|
|
1846 }
|
|
1847 __ jmp(unlock_done);
|
|
1848
|
|
1849 // END Slow path unlock
|
|
1850
|
|
1851 } // synchronized
|
|
1852
|
|
1853 // SLOW PATH Reguard the stack if needed
|
|
1854
|
|
1855 __ bind(reguard);
|
|
1856 save_native_result(masm, ret_type, stack_slots);
|
|
1857 __ movq(r12, rsp); // remember sp
|
|
1858 __ subq(rsp, frame::arg_reg_save_area_bytes); // windows
|
|
1859 __ andq(rsp, -16); // align stack as required by ABI
|
|
1860 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
|
|
1861 __ movq(rsp, r12); // restore sp
|
|
1862 restore_native_result(masm, ret_type, stack_slots);
|
|
1863 // and continue
|
|
1864 __ jmp(reguard_done);
|
|
1865
|
|
1866
|
|
1867
|
|
1868 __ flush();
|
|
1869
|
|
1870 nmethod *nm = nmethod::new_native_nmethod(method,
|
|
1871 masm->code(),
|
|
1872 vep_offset,
|
|
1873 frame_complete,
|
|
1874 stack_slots / VMRegImpl::slots_per_word,
|
|
1875 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
|
|
1876 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
|
|
1877 oop_maps);
|
|
1878 return nm;
|
|
1879
|
|
1880 }
|
|
1881
|
|
1882 // this function returns the adjust size (in number of words) to a c2i adapter
|
|
1883 // activation for use during deoptimization
|
|
1884 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) {
|
|
1885 return (callee_locals - callee_parameters) * Interpreter::stackElementWords();
|
|
1886 }
|
|
1887
|
|
1888
|
|
1889 uint SharedRuntime::out_preserve_stack_slots() {
|
|
1890 return 0;
|
|
1891 }
|
|
1892
|
|
1893
|
|
1894 //------------------------------generate_deopt_blob----------------------------
|
|
1895 void SharedRuntime::generate_deopt_blob() {
|
|
1896 // Allocate space for the code
|
|
1897 ResourceMark rm;
|
|
1898 // Setup code generation tools
|
|
1899 CodeBuffer buffer("deopt_blob", 2048, 1024);
|
|
1900 MacroAssembler* masm = new MacroAssembler(&buffer);
|
|
1901 int frame_size_in_words;
|
|
1902 OopMap* map = NULL;
|
|
1903 OopMapSet *oop_maps = new OopMapSet();
|
|
1904
|
|
1905 // -------------
|
|
1906 // This code enters when returning to a de-optimized nmethod. A return
|
|
1907 // address has been pushed on the the stack, and return values are in
|
|
1908 // registers.
|
|
1909 // If we are doing a normal deopt then we were called from the patched
|
|
1910 // nmethod from the point we returned to the nmethod. So the return
|
|
1911 // address on the stack is wrong by NativeCall::instruction_size
|
|
1912 // We will adjust the value so it looks like we have the original return
|
|
1913 // address on the stack (like when we eagerly deoptimized).
|
|
1914 // In the case of an exception pending when deoptimizing, we enter
|
|
1915 // with a return address on the stack that points after the call we patched
|
|
1916 // into the exception handler. We have the following register state from,
|
|
1917 // e.g., the forward exception stub (see stubGenerator_x86_64.cpp).
|
|
1918 // rax: exception oop
|
|
1919 // rbx: exception handler
|
|
1920 // rdx: throwing pc
|
|
1921 // So in this case we simply jam rdx into the useless return address and
|
|
1922 // the stack looks just like we want.
|
|
1923 //
|
|
1924 // At this point we need to de-opt. We save the argument return
|
|
1925 // registers. We call the first C routine, fetch_unroll_info(). This
|
|
1926 // routine captures the return values and returns a structure which
|
|
1927 // describes the current frame size and the sizes of all replacement frames.
|
|
1928 // The current frame is compiled code and may contain many inlined
|
|
1929 // functions, each with their own JVM state. We pop the current frame, then
|
|
1930 // push all the new frames. Then we call the C routine unpack_frames() to
|
|
1931 // populate these frames. Finally unpack_frames() returns us the new target
|
|
1932 // address. Notice that callee-save registers are BLOWN here; they have
|
|
1933 // already been captured in the vframeArray at the time the return PC was
|
|
1934 // patched.
|
|
1935 address start = __ pc();
|
|
1936 Label cont;
|
|
1937
|
|
1938 // Prolog for non exception case!
|
|
1939
|
|
1940 // Save everything in sight.
|
|
1941 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
|
|
1942
|
|
1943 // Normal deoptimization. Save exec mode for unpack_frames.
|
|
1944 __ movl(r12, Deoptimization::Unpack_deopt); // callee-saved
|
|
1945 __ jmp(cont);
|
|
1946
|
|
1947 int exception_offset = __ pc() - start;
|
|
1948
|
|
1949 // Prolog for exception case
|
|
1950
|
|
1951 // Push throwing pc as return address
|
|
1952 __ pushq(rdx);
|
|
1953
|
|
1954 // Save everything in sight.
|
|
1955 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
|
|
1956
|
|
1957 // Deopt during an exception. Save exec mode for unpack_frames.
|
|
1958 __ movl(r12, Deoptimization::Unpack_exception); // callee-saved
|
|
1959
|
|
1960 __ bind(cont);
|
|
1961
|
|
1962 // Call C code. Need thread and this frame, but NOT official VM entry
|
|
1963 // crud. We cannot block on this call, no GC can happen.
|
|
1964 //
|
|
1965 // UnrollBlock* fetch_unroll_info(JavaThread* thread)
|
|
1966
|
|
1967 // fetch_unroll_info needs to call last_java_frame().
|
|
1968
|
|
1969 __ set_last_Java_frame(noreg, noreg, NULL);
|
|
1970 #ifdef ASSERT
|
|
1971 { Label L;
|
|
1972 __ cmpq(Address(r15_thread,
|
|
1973 JavaThread::last_Java_fp_offset()),
|
|
1974 0);
|
|
1975 __ jcc(Assembler::equal, L);
|
|
1976 __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
|
|
1977 __ bind(L);
|
|
1978 }
|
|
1979 #endif // ASSERT
|
|
1980 __ movq(c_rarg0, r15_thread);
|
|
1981 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
|
|
1982
|
|
1983 // Need to have an oopmap that tells fetch_unroll_info where to
|
|
1984 // find any register it might need.
|
|
1985 oop_maps->add_gc_map(__ pc() - start, map);
|
|
1986
|
|
1987 __ reset_last_Java_frame(false, false);
|
|
1988
|
|
1989 // Load UnrollBlock* into rdi
|
|
1990 __ movq(rdi, rax);
|
|
1991
|
|
1992 // Only register save data is on the stack.
|
|
1993 // Now restore the result registers. Everything else is either dead
|
|
1994 // or captured in the vframeArray.
|
|
1995 RegisterSaver::restore_result_registers(masm);
|
|
1996
|
|
1997 // All of the register save area has been popped of the stack. Only the
|
|
1998 // return address remains.
|
|
1999
|
|
2000 // Pop all the frames we must move/replace.
|
|
2001 //
|
|
2002 // Frame picture (youngest to oldest)
|
|
2003 // 1: self-frame (no frame link)
|
|
2004 // 2: deopting frame (no frame link)
|
|
2005 // 3: caller of deopting frame (could be compiled/interpreted).
|
|
2006 //
|
|
2007 // Note: by leaving the return address of self-frame on the stack
|
|
2008 // and using the size of frame 2 to adjust the stack
|
|
2009 // when we are done the return to frame 3 will still be on the stack.
|
|
2010
|
|
2011 // Pop deoptimized frame
|
|
2012 __ movl(rcx, Address(rdi, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()));
|
|
2013 __ addq(rsp, rcx);
|
|
2014
|
|
2015 // rsp should be pointing at the return address to the caller (3)
|
|
2016
|
|
2017 // Stack bang to make sure there's enough room for these interpreter frames.
|
|
2018 if (UseStackBanging) {
|
|
2019 __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
|
|
2020 __ bang_stack_size(rbx, rcx);
|
|
2021 }
|
|
2022
|
|
2023 // Load address of array of frame pcs into rcx
|
|
2024 __ movq(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
|
|
2025
|
|
2026 // Trash the old pc
|
|
2027 __ addq(rsp, wordSize);
|
|
2028
|
|
2029 // Load address of array of frame sizes into rsi
|
|
2030 __ movq(rsi, Address(rdi, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()));
|
|
2031
|
|
2032 // Load counter into rdx
|
|
2033 __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()));
|
|
2034
|
|
2035 // Pick up the initial fp we should save
|
|
2036 __ movq(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes()));
|
|
2037
|
|
2038 // Now adjust the caller's stack to make up for the extra locals
|
|
2039 // but record the original sp so that we can save it in the skeletal interpreter
|
|
2040 // frame and the stack walking of interpreter_sender will get the unextended sp
|
|
2041 // value and not the "real" sp value.
|
|
2042
|
|
2043 const Register sender_sp = r8;
|
|
2044
|
|
2045 __ movq(sender_sp, rsp);
|
|
2046 __ movl(rbx, Address(rdi,
|
|
2047 Deoptimization::UnrollBlock::
|
|
2048 caller_adjustment_offset_in_bytes()));
|
|
2049 __ subq(rsp, rbx);
|
|
2050
|
|
2051 // Push interpreter frames in a loop
|
|
2052 Label loop;
|
|
2053 __ bind(loop);
|
|
2054 __ movq(rbx, Address(rsi, 0)); // Load frame size
|
|
2055 __ subq(rbx, 2*wordSize); // We'll push pc and ebp by hand
|
|
2056 __ pushq(Address(rcx, 0)); // Save return address
|
|
2057 __ enter(); // Save old & set new ebp
|
|
2058 __ subq(rsp, rbx); // Prolog
|
|
2059 __ movq(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize),
|
|
2060 sender_sp); // Make it walkable
|
|
2061 // This value is corrected by layout_activation_impl
|
|
2062 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int)NULL_WORD );
|
|
2063 __ movq(sender_sp, rsp); // Pass sender_sp to next frame
|
|
2064 __ addq(rsi, wordSize); // Bump array pointer (sizes)
|
|
2065 __ addq(rcx, wordSize); // Bump array pointer (pcs)
|
|
2066 __ decrementl(rdx); // Decrement counter
|
|
2067 __ jcc(Assembler::notZero, loop);
|
|
2068 __ pushq(Address(rcx, 0)); // Save final return address
|
|
2069
|
|
2070 // Re-push self-frame
|
|
2071 __ enter(); // Save old & set new ebp
|
|
2072
|
|
2073 // Allocate a full sized register save area.
|
|
2074 // Return address and rbp are in place, so we allocate two less words.
|
|
2075 __ subq(rsp, (frame_size_in_words - 2) * wordSize);
|
|
2076
|
|
2077 // Restore frame locals after moving the frame
|
|
2078 __ movdbl(Address(rsp, RegisterSaver::xmm0_offset_in_bytes()), xmm0);
|
|
2079 __ movq(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
|
|
2080
|
|
2081 // Call C code. Need thread but NOT official VM entry
|
|
2082 // crud. We cannot block on this call, no GC can happen. Call should
|
|
2083 // restore return values to their stack-slots with the new SP.
|
|
2084 //
|
|
2085 // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
|
|
2086
|
|
2087 // Use rbp because the frames look interpreted now
|
|
2088 __ set_last_Java_frame(noreg, rbp, NULL);
|
|
2089
|
|
2090 __ movq(c_rarg0, r15_thread);
|
|
2091 __ movl(c_rarg1, r12); // second arg: exec_mode
|
|
2092 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
|
|
2093
|
|
2094 // Set an oopmap for the call site
|
|
2095 oop_maps->add_gc_map(__ pc() - start,
|
|
2096 new OopMap( frame_size_in_words, 0 ));
|
|
2097
|
|
2098 __ reset_last_Java_frame(true, false);
|
|
2099
|
|
2100 // Collect return values
|
|
2101 __ movdbl(xmm0, Address(rsp, RegisterSaver::xmm0_offset_in_bytes()));
|
|
2102 __ movq(rax, Address(rsp, RegisterSaver::rax_offset_in_bytes()));
|
|
2103
|
|
2104 // Pop self-frame.
|
|
2105 __ leave(); // Epilog
|
|
2106
|
|
2107 // Jump to interpreter
|
|
2108 __ ret(0);
|
|
2109
|
|
2110 // Make sure all code is generated
|
|
2111 masm->flush();
|
|
2112
|
|
2113 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, 0, frame_size_in_words);
|
|
2114 }
|
|
2115
|
|
2116 #ifdef COMPILER2
|
|
2117 //------------------------------generate_uncommon_trap_blob--------------------
|
|
2118 void SharedRuntime::generate_uncommon_trap_blob() {
|
|
2119 // Allocate space for the code
|
|
2120 ResourceMark rm;
|
|
2121 // Setup code generation tools
|
|
2122 CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
|
|
2123 MacroAssembler* masm = new MacroAssembler(&buffer);
|
|
2124
|
|
2125 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
|
|
2126
|
|
2127 address start = __ pc();
|
|
2128
|
|
2129 // Push self-frame. We get here with a return address on the
|
|
2130 // stack, so rsp is 8-byte aligned until we allocate our frame.
|
|
2131 __ subq(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog!
|
|
2132
|
|
2133 // No callee saved registers. rbp is assumed implicitly saved
|
|
2134 __ movq(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
|
|
2135
|
|
2136 // compiler left unloaded_class_index in j_rarg0 move to where the
|
|
2137 // runtime expects it.
|
|
2138 __ movl(c_rarg1, j_rarg0);
|
|
2139
|
|
2140 __ set_last_Java_frame(noreg, noreg, NULL);
|
|
2141
|
|
2142 // Call C code. Need thread but NOT official VM entry
|
|
2143 // crud. We cannot block on this call, no GC can happen. Call should
|
|
2144 // capture callee-saved registers as well as return values.
|
|
2145 // Thread is in rdi already.
|
|
2146 //
|
|
2147 // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
|
|
2148
|
|
2149 __ movq(c_rarg0, r15_thread);
|
|
2150 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)));
|
|
2151
|
|
2152 // Set an oopmap for the call site
|
|
2153 OopMapSet* oop_maps = new OopMapSet();
|
|
2154 OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
|
|
2155
|
|
2156 // location of rbp is known implicitly by the frame sender code
|
|
2157
|
|
2158 oop_maps->add_gc_map(__ pc() - start, map);
|
|
2159
|
|
2160 __ reset_last_Java_frame(false, false);
|
|
2161
|
|
2162 // Load UnrollBlock* into rdi
|
|
2163 __ movq(rdi, rax);
|
|
2164
|
|
2165 // Pop all the frames we must move/replace.
|
|
2166 //
|
|
2167 // Frame picture (youngest to oldest)
|
|
2168 // 1: self-frame (no frame link)
|
|
2169 // 2: deopting frame (no frame link)
|
|
2170 // 3: caller of deopting frame (could be compiled/interpreted).
|
|
2171
|
|
2172 // Pop self-frame. We have no frame, and must rely only on rax and rsp.
|
|
2173 __ addq(rsp, (SimpleRuntimeFrame::framesize - 2) << LogBytesPerInt); // Epilog!
|
|
2174
|
|
2175 // Pop deoptimized frame (int)
|
|
2176 __ movl(rcx, Address(rdi,
|
|
2177 Deoptimization::UnrollBlock::
|
|
2178 size_of_deoptimized_frame_offset_in_bytes()));
|
|
2179 __ addq(rsp, rcx);
|
|
2180
|
|
2181 // rsp should be pointing at the return address to the caller (3)
|
|
2182
|
|
2183 // Stack bang to make sure there's enough room for these interpreter frames.
|
|
2184 if (UseStackBanging) {
|
|
2185 __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
|
|
2186 __ bang_stack_size(rbx, rcx);
|
|
2187 }
|
|
2188
|
|
2189 // Load address of array of frame pcs into rcx (address*)
|
|
2190 __ movq(rcx,
|
|
2191 Address(rdi,
|
|
2192 Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
|
|
2193
|
|
2194 // Trash the return pc
|
|
2195 __ addq(rsp, wordSize);
|
|
2196
|
|
2197 // Load address of array of frame sizes into rsi (intptr_t*)
|
|
2198 __ movq(rsi, Address(rdi,
|
|
2199 Deoptimization::UnrollBlock::
|
|
2200 frame_sizes_offset_in_bytes()));
|
|
2201
|
|
2202 // Counter
|
|
2203 __ movl(rdx, Address(rdi,
|
|
2204 Deoptimization::UnrollBlock::
|
|
2205 number_of_frames_offset_in_bytes())); // (int)
|
|
2206
|
|
2207 // Pick up the initial fp we should save
|
|
2208 __ movq(rbp,
|
|
2209 Address(rdi,
|
|
2210 Deoptimization::UnrollBlock::initial_fp_offset_in_bytes()));
|
|
2211
|
|
2212 // Now adjust the caller's stack to make up for the extra locals but
|
|
2213 // record the original sp so that we can save it in the skeletal
|
|
2214 // interpreter frame and the stack walking of interpreter_sender
|
|
2215 // will get the unextended sp value and not the "real" sp value.
|
|
2216
|
|
2217 const Register sender_sp = r8;
|
|
2218
|
|
2219 __ movq(sender_sp, rsp);
|
|
2220 __ movl(rbx, Address(rdi,
|
|
2221 Deoptimization::UnrollBlock::
|
|
2222 caller_adjustment_offset_in_bytes())); // (int)
|
|
2223 __ subq(rsp, rbx);
|
|
2224
|
|
2225 // Push interpreter frames in a loop
|
|
2226 Label loop;
|
|
2227 __ bind(loop);
|
|
2228 __ movq(rbx, Address(rsi, 0)); // Load frame size
|
|
2229 __ subq(rbx, 2 * wordSize); // We'll push pc and rbp by hand
|
|
2230 __ pushq(Address(rcx, 0)); // Save return address
|
|
2231 __ enter(); // Save old & set new rbp
|
|
2232 __ subq(rsp, rbx); // Prolog
|
|
2233 __ movq(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize),
|
|
2234 sender_sp); // Make it walkable
|
|
2235 // This value is corrected by layout_activation_impl
|
|
2236 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int)NULL_WORD );
|
|
2237 __ movq(sender_sp, rsp); // Pass sender_sp to next frame
|
|
2238 __ addq(rsi, wordSize); // Bump array pointer (sizes)
|
|
2239 __ addq(rcx, wordSize); // Bump array pointer (pcs)
|
|
2240 __ decrementl(rdx); // Decrement counter
|
|
2241 __ jcc(Assembler::notZero, loop);
|
|
2242 __ pushq(Address(rcx, 0)); // Save final return address
|
|
2243
|
|
2244 // Re-push self-frame
|
|
2245 __ enter(); // Save old & set new rbp
|
|
2246 __ subq(rsp, (SimpleRuntimeFrame::framesize - 4) << LogBytesPerInt);
|
|
2247 // Prolog
|
|
2248
|
|
2249 // Use rbp because the frames look interpreted now
|
|
2250 __ set_last_Java_frame(noreg, rbp, NULL);
|
|
2251
|
|
2252 // Call C code. Need thread but NOT official VM entry
|
|
2253 // crud. We cannot block on this call, no GC can happen. Call should
|
|
2254 // restore return values to their stack-slots with the new SP.
|
|
2255 // Thread is in rdi already.
|
|
2256 //
|
|
2257 // BasicType unpack_frames(JavaThread* thread, int exec_mode);
|
|
2258
|
|
2259 __ movq(c_rarg0, r15_thread);
|
|
2260 __ movl(c_rarg1, Deoptimization::Unpack_uncommon_trap);
|
|
2261 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
|
|
2262
|
|
2263 // Set an oopmap for the call site
|
|
2264 oop_maps->add_gc_map(__ pc() - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
|
|
2265
|
|
2266 __ reset_last_Java_frame(true, false);
|
|
2267
|
|
2268 // Pop self-frame.
|
|
2269 __ leave(); // Epilog
|
|
2270
|
|
2271 // Jump to interpreter
|
|
2272 __ ret(0);
|
|
2273
|
|
2274 // Make sure all code is generated
|
|
2275 masm->flush();
|
|
2276
|
|
2277 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps,
|
|
2278 SimpleRuntimeFrame::framesize >> 1);
|
|
2279 }
|
|
2280 #endif // COMPILER2
|
|
2281
|
|
2282
|
|
2283 //------------------------------generate_handler_blob------
|
|
2284 //
|
|
2285 // Generate a special Compile2Runtime blob that saves all registers,
|
|
2286 // and setup oopmap.
|
|
2287 //
|
|
2288 static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) {
|
|
2289 assert(StubRoutines::forward_exception_entry() != NULL,
|
|
2290 "must be generated before");
|
|
2291
|
|
2292 ResourceMark rm;
|
|
2293 OopMapSet *oop_maps = new OopMapSet();
|
|
2294 OopMap* map;
|
|
2295
|
|
2296 // Allocate space for the code. Setup code generation tools.
|
|
2297 CodeBuffer buffer("handler_blob", 2048, 1024);
|
|
2298 MacroAssembler* masm = new MacroAssembler(&buffer);
|
|
2299
|
|
2300 address start = __ pc();
|
|
2301 address call_pc = NULL;
|
|
2302 int frame_size_in_words;
|
|
2303
|
|
2304 // Make room for return address (or push it again)
|
|
2305 if (!cause_return) {
|
|
2306 __ pushq(rbx);
|
|
2307 }
|
|
2308
|
|
2309 // Save registers, fpu state, and flags
|
|
2310 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
|
|
2311
|
|
2312 // The following is basically a call_VM. However, we need the precise
|
|
2313 // address of the call in order to generate an oopmap. Hence, we do all the
|
|
2314 // work outselves.
|
|
2315
|
|
2316 __ set_last_Java_frame(noreg, noreg, NULL);
|
|
2317
|
|
2318 // The return address must always be correct so that frame constructor never
|
|
2319 // sees an invalid pc.
|
|
2320
|
|
2321 if (!cause_return) {
|
|
2322 // overwrite the dummy value we pushed on entry
|
|
2323 __ movq(c_rarg0, Address(r15_thread, JavaThread::saved_exception_pc_offset()));
|
|
2324 __ movq(Address(rbp, wordSize), c_rarg0);
|
|
2325 }
|
|
2326
|
|
2327 // Do the call
|
|
2328 __ movq(c_rarg0, r15_thread);
|
|
2329 __ call(RuntimeAddress(call_ptr));
|
|
2330
|
|
2331 // Set an oopmap for the call site. This oopmap will map all
|
|
2332 // oop-registers and debug-info registers as callee-saved. This
|
|
2333 // will allow deoptimization at this safepoint to find all possible
|
|
2334 // debug-info recordings, as well as let GC find all oops.
|
|
2335
|
|
2336 oop_maps->add_gc_map( __ pc() - start, map);
|
|
2337
|
|
2338 Label noException;
|
|
2339
|
|
2340 __ reset_last_Java_frame(false, false);
|
|
2341
|
|
2342 __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD);
|
|
2343 __ jcc(Assembler::equal, noException);
|
|
2344
|
|
2345 // Exception pending
|
|
2346
|
|
2347 RegisterSaver::restore_live_registers(masm);
|
|
2348
|
|
2349 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
|
|
2350
|
|
2351 // No exception case
|
|
2352 __ bind(noException);
|
|
2353
|
|
2354 // Normal exit, restore registers and exit.
|
|
2355 RegisterSaver::restore_live_registers(masm);
|
|
2356
|
|
2357 __ ret(0);
|
|
2358
|
|
2359 // Make sure all code is generated
|
|
2360 masm->flush();
|
|
2361
|
|
2362 // Fill-out other meta info
|
|
2363 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
|
|
2364 }
|
|
2365
|
|
2366 //
|
|
2367 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
|
|
2368 //
|
|
2369 // Generate a stub that calls into vm to find out the proper destination
|
|
2370 // of a java call. All the argument registers are live at this point
|
|
2371 // but since this is generic code we don't know what they are and the caller
|
|
2372 // must do any gc of the args.
|
|
2373 //
|
|
2374 static RuntimeStub* generate_resolve_blob(address destination, const char* name) {
|
|
2375 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
|
|
2376
|
|
2377 // allocate space for the code
|
|
2378 ResourceMark rm;
|
|
2379
|
|
2380 CodeBuffer buffer(name, 1000, 512);
|
|
2381 MacroAssembler* masm = new MacroAssembler(&buffer);
|
|
2382
|
|
2383 int frame_size_in_words;
|
|
2384
|
|
2385 OopMapSet *oop_maps = new OopMapSet();
|
|
2386 OopMap* map = NULL;
|
|
2387
|
|
2388 int start = __ offset();
|
|
2389
|
|
2390 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
|
|
2391
|
|
2392 int frame_complete = __ offset();
|
|
2393
|
|
2394 __ set_last_Java_frame(noreg, noreg, NULL);
|
|
2395
|
|
2396 __ movq(c_rarg0, r15_thread);
|
|
2397
|
|
2398 __ call(RuntimeAddress(destination));
|
|
2399
|
|
2400
|
|
2401 // Set an oopmap for the call site.
|
|
2402 // We need this not only for callee-saved registers, but also for volatile
|
|
2403 // registers that the compiler might be keeping live across a safepoint.
|
|
2404
|
|
2405 oop_maps->add_gc_map( __ offset() - start, map);
|
|
2406
|
|
2407 // rax contains the address we are going to jump to assuming no exception got installed
|
|
2408
|
|
2409 // clear last_Java_sp
|
|
2410 __ reset_last_Java_frame(false, false);
|
|
2411 // check for pending exceptions
|
|
2412 Label pending;
|
|
2413 __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD);
|
|
2414 __ jcc(Assembler::notEqual, pending);
|
|
2415
|
|
2416 // get the returned methodOop
|
|
2417 __ movq(rbx, Address(r15_thread, JavaThread::vm_result_offset()));
|
|
2418 __ movq(Address(rsp, RegisterSaver::rbx_offset_in_bytes()), rbx);
|
|
2419
|
|
2420 __ movq(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
|
|
2421
|
|
2422 RegisterSaver::restore_live_registers(masm);
|
|
2423
|
|
2424 // We are back the the original state on entry and ready to go.
|
|
2425
|
|
2426 __ jmp(rax);
|
|
2427
|
|
2428 // Pending exception after the safepoint
|
|
2429
|
|
2430 __ bind(pending);
|
|
2431
|
|
2432 RegisterSaver::restore_live_registers(masm);
|
|
2433
|
|
2434 // exception pending => remove activation and forward to exception handler
|
|
2435
|
|
2436 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int)NULL_WORD);
|
|
2437
|
|
2438 __ movq(rax, Address(r15_thread, Thread::pending_exception_offset()));
|
|
2439 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
|
|
2440
|
|
2441 // -------------
|
|
2442 // make sure all code is generated
|
|
2443 masm->flush();
|
|
2444
|
|
2445 // return the blob
|
|
2446 // frame_size_words or bytes??
|
|
2447 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
|
|
2448 }
|
|
2449
|
|
2450
|
|
2451 void SharedRuntime::generate_stubs() {
|
|
2452
|
|
2453 _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method),
|
|
2454 "wrong_method_stub");
|
|
2455 _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss),
|
|
2456 "ic_miss_stub");
|
|
2457 _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C),
|
|
2458 "resolve_opt_virtual_call");
|
|
2459
|
|
2460 _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C),
|
|
2461 "resolve_virtual_call");
|
|
2462
|
|
2463 _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C),
|
|
2464 "resolve_static_call");
|
|
2465 _polling_page_safepoint_handler_blob =
|
|
2466 generate_handler_blob(CAST_FROM_FN_PTR(address,
|
|
2467 SafepointSynchronize::handle_polling_page_exception), false);
|
|
2468
|
|
2469 _polling_page_return_handler_blob =
|
|
2470 generate_handler_blob(CAST_FROM_FN_PTR(address,
|
|
2471 SafepointSynchronize::handle_polling_page_exception), true);
|
|
2472
|
|
2473 generate_deopt_blob();
|
|
2474
|
|
2475 #ifdef COMPILER2
|
|
2476 generate_uncommon_trap_blob();
|
|
2477 #endif // COMPILER2
|
|
2478 }
|
|
2479
|
|
2480
|
|
2481 #ifdef COMPILER2
|
|
2482 // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame
|
|
2483 //
|
|
2484 //------------------------------generate_exception_blob---------------------------
|
|
2485 // creates exception blob at the end
|
|
2486 // Using exception blob, this code is jumped from a compiled method.
|
|
2487 // (see emit_exception_handler in x86_64.ad file)
|
|
2488 //
|
|
2489 // Given an exception pc at a call we call into the runtime for the
|
|
2490 // handler in this method. This handler might merely restore state
|
|
2491 // (i.e. callee save registers) unwind the frame and jump to the
|
|
2492 // exception handler for the nmethod if there is no Java level handler
|
|
2493 // for the nmethod.
|
|
2494 //
|
|
2495 // This code is entered with a jmp.
|
|
2496 //
|
|
2497 // Arguments:
|
|
2498 // rax: exception oop
|
|
2499 // rdx: exception pc
|
|
2500 //
|
|
2501 // Results:
|
|
2502 // rax: exception oop
|
|
2503 // rdx: exception pc in caller or ???
|
|
2504 // destination: exception handler of caller
|
|
2505 //
|
|
2506 // Note: the exception pc MUST be at a call (precise debug information)
|
|
2507 // Registers rax, rdx, rcx, rsi, rdi, r8-r11 are not callee saved.
|
|
2508 //
|
|
2509
|
|
2510 void OptoRuntime::generate_exception_blob() {
|
|
2511 assert(!OptoRuntime::is_callee_saved_register(RDX_num), "");
|
|
2512 assert(!OptoRuntime::is_callee_saved_register(RAX_num), "");
|
|
2513 assert(!OptoRuntime::is_callee_saved_register(RCX_num), "");
|
|
2514
|
|
2515 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
|
|
2516
|
|
2517 // Allocate space for the code
|
|
2518 ResourceMark rm;
|
|
2519 // Setup code generation tools
|
|
2520 CodeBuffer buffer("exception_blob", 2048, 1024);
|
|
2521 MacroAssembler* masm = new MacroAssembler(&buffer);
|
|
2522
|
|
2523
|
|
2524 address start = __ pc();
|
|
2525
|
|
2526 // Exception pc is 'return address' for stack walker
|
|
2527 __ pushq(rdx);
|
|
2528 __ subq(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Prolog
|
|
2529
|
|
2530 // Save callee-saved registers. See x86_64.ad.
|
|
2531
|
|
2532 // rbp is an implicitly saved callee saved register (i.e. the calling
|
|
2533 // convention will save restore it in prolog/epilog) Other than that
|
|
2534 // there are no callee save registers now that adapter frames are gone.
|
|
2535
|
|
2536 __ movq(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
|
|
2537
|
|
2538 // Store exception in Thread object. We cannot pass any arguments to the
|
|
2539 // handle_exception call, since we do not want to make any assumption
|
|
2540 // about the size of the frame where the exception happened in.
|
|
2541 // c_rarg0 is either rdi (Linux) or rcx (Windows).
|
|
2542 __ movq(Address(r15_thread, JavaThread::exception_oop_offset()),rax);
|
|
2543 __ movq(Address(r15_thread, JavaThread::exception_pc_offset()), rdx);
|
|
2544
|
|
2545 // This call does all the hard work. It checks if an exception handler
|
|
2546 // exists in the method.
|
|
2547 // If so, it returns the handler address.
|
|
2548 // If not, it prepares for stack-unwinding, restoring the callee-save
|
|
2549 // registers of the frame being removed.
|
|
2550 //
|
|
2551 // address OptoRuntime::handle_exception_C(JavaThread* thread)
|
|
2552
|
|
2553 __ set_last_Java_frame(noreg, noreg, NULL);
|
|
2554 __ movq(c_rarg0, r15_thread);
|
|
2555 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
|
|
2556
|
|
2557 // Set an oopmap for the call site. This oopmap will only be used if we
|
|
2558 // are unwinding the stack. Hence, all locations will be dead.
|
|
2559 // Callee-saved registers will be the same as the frame above (i.e.,
|
|
2560 // handle_exception_stub), since they were restored when we got the
|
|
2561 // exception.
|
|
2562
|
|
2563 OopMapSet* oop_maps = new OopMapSet();
|
|
2564
|
|
2565 oop_maps->add_gc_map( __ pc()-start, new OopMap(SimpleRuntimeFrame::framesize, 0));
|
|
2566
|
|
2567 __ reset_last_Java_frame(false, false);
|
|
2568
|
|
2569 // Restore callee-saved registers
|
|
2570
|
|
2571 // rbp is an implicitly saved callee saved register (i.e. the calling
|
|
2572 // convention will save restore it in prolog/epilog) Other than that
|
|
2573 // there are no callee save registers no that adapter frames are gone.
|
|
2574
|
|
2575 __ movq(rbp, Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt));
|
|
2576
|
|
2577 __ addq(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog
|
|
2578 __ popq(rdx); // No need for exception pc anymore
|
|
2579
|
|
2580 // rax: exception handler
|
|
2581
|
|
2582 // We have a handler in rax (could be deopt blob).
|
|
2583 __ movq(r8, rax);
|
|
2584
|
|
2585 // Get the exception oop
|
|
2586 __ movq(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
|
|
2587 // Get the exception pc in case we are deoptimized
|
|
2588 __ movq(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
|
|
2589 #ifdef ASSERT
|
|
2590 __ movptr(Address(r15_thread, JavaThread::exception_handler_pc_offset()), (int)NULL_WORD);
|
|
2591 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int)NULL_WORD);
|
|
2592 #endif
|
|
2593 // Clear the exception oop so GC no longer processes it as a root.
|
|
2594 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int)NULL_WORD);
|
|
2595
|
|
2596 // rax: exception oop
|
|
2597 // r8: exception handler
|
|
2598 // rdx: exception pc
|
|
2599 // Jump to handler
|
|
2600
|
|
2601 __ jmp(r8);
|
|
2602
|
|
2603 // Make sure all code is generated
|
|
2604 masm->flush();
|
|
2605
|
|
2606 // Set exception blob
|
|
2607 _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
|
|
2608 }
|
|
2609 #endif // COMPILER2
|