1 /*
2  * Copyright (c) 2004, 2020, Oracle and/or its affiliates. All rights reserved.
3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4  *
5  * This code is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 only, as
7  * published by the Free Software Foundation.
8  *
9  * This code is distributed in the hope that it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12  * version 2 for more details (a copy is included in the LICENSE file that
13  * accompanied this code).
14  *
15  * You should have received a copy of the GNU General Public License version
16  * 2 along with this work; if not, write to the Free Software Foundation,
17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18  *
19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20  * or visit www.oracle.com if you need additional information or have any
21  * questions.
22  *
23  */
24 
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "memory/resourceArea.hpp"
28 #include "prims/jniFastGetField.hpp"
29 #include "prims/jvm_misc.hpp"
30 #include "prims/jvmtiExport.hpp"
31 #include "runtime/safepoint.hpp"
32 
33 #define __ masm->
34 
35 #define BUFFER_SIZE 30
36 
37 #ifdef _WINDOWS
38 GetBooleanField_t JNI_FastGetField::jni_fast_GetBooleanField_fp;
39 GetByteField_t    JNI_FastGetField::jni_fast_GetByteField_fp;
40 GetCharField_t    JNI_FastGetField::jni_fast_GetCharField_fp;
41 GetShortField_t   JNI_FastGetField::jni_fast_GetShortField_fp;
42 GetIntField_t     JNI_FastGetField::jni_fast_GetIntField_fp;
43 GetLongField_t    JNI_FastGetField::jni_fast_GetLongField_fp;
44 GetFloatField_t   JNI_FastGetField::jni_fast_GetFloatField_fp;
45 GetDoubleField_t  JNI_FastGetField::jni_fast_GetDoubleField_fp;
46 #endif
47 
48 // Instead of issuing lfence for LoadLoad barrier, we create data dependency
49 // between loads, which is much more efficient than lfence.
50 
generate_fast_get_int_field0(BasicType type)51 address JNI_FastGetField::generate_fast_get_int_field0(BasicType type) {
52   const char *name = NULL;
53   switch (type) {
54     case T_BOOLEAN: name = "jni_fast_GetBooleanField"; break;
55     case T_BYTE:    name = "jni_fast_GetByteField";    break;
56     case T_CHAR:    name = "jni_fast_GetCharField";    break;
57     case T_SHORT:   name = "jni_fast_GetShortField";   break;
58     case T_INT:     name = "jni_fast_GetIntField";     break;
59     default:        ShouldNotReachHere();
60   }
61   ResourceMark rm;
62   BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
63   CodeBuffer cbuf(blob);
64   MacroAssembler* masm = new MacroAssembler(&cbuf);
65   address fast_entry = __ pc();
66 
67   Label slow;
68 
69   // stack layout:    offset from rsp (in words):
70   //  return pc        0
71   //  jni env          1
72   //  obj              2
73   //  jfieldID         3
74 
75   ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
76   __ mov32 (rcx, counter);
77   __ testb (rcx, 1);
78   __ jcc (Assembler::notZero, slow);
79 
80   if (JvmtiExport::can_post_field_access()) {
81     // Check to see if a field access watch has been set before we
82     // take the fast path.
83     __ cmp32(ExternalAddress((address) JvmtiExport::get_field_access_count_addr()), 0);
84     __ jcc(Assembler::notZero, slow);
85   }
86 
87   __ mov(rax, rcx);
88   __ andptr(rax, 1);                         // rax, must end up 0
89   __ movptr(rdx, Address(rsp, rax, Address::times_1, 2*wordSize));
90                                             // obj, notice rax, is 0.
91                                             // rdx is data dependent on rcx.
92   __ movptr(rax, Address(rsp, 3*wordSize));  // jfieldID
93 
94   __ clear_jweak_tag(rdx);
95 
96   __ movptr(rdx, Address(rdx, 0));           // *obj
97   __ shrptr (rax, 2);                         // offset
98 
99   assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
100   speculative_load_pclist[count] = __ pc();
101   switch (type) {
102     case T_BOOLEAN: __ movzbl (rax, Address(rdx, rax, Address::times_1)); break;
103     case T_BYTE:    __ movsbl (rax, Address(rdx, rax, Address::times_1)); break;
104     case T_CHAR:    __ movzwl (rax, Address(rdx, rax, Address::times_1)); break;
105     case T_SHORT:   __ movswl (rax, Address(rdx, rax, Address::times_1)); break;
106     case T_INT:     __ movl   (rax, Address(rdx, rax, Address::times_1)); break;
107     default:        ShouldNotReachHere();
108   }
109 
110   Address ca1;
111   __ lea(rdx, counter);
112   __ xorptr(rdx, rax);
113   __ xorptr(rdx, rax);
114   __ cmp32(rcx, Address(rdx, 0));
115   // ca1 is the same as ca because
116   // rax, ^ counter_addr ^ rax, = address
117   // ca1 is data dependent on rax,.
118   __ jcc (Assembler::notEqual, slow);
119 
120 #ifndef _WINDOWS
121   __ ret (0);
122 #else
123   // __stdcall calling convention
124   __ ret (3*wordSize);
125 #endif
126 
127   slowcase_entry_pclist[count++] = __ pc();
128   __ bind (slow);
129   address slow_case_addr = NULL;
130   switch (type) {
131     case T_BOOLEAN: slow_case_addr = jni_GetBooleanField_addr(); break;
132     case T_BYTE:    slow_case_addr = jni_GetByteField_addr();    break;
133     case T_CHAR:    slow_case_addr = jni_GetCharField_addr();    break;
134     case T_SHORT:   slow_case_addr = jni_GetShortField_addr();   break;
135     case T_INT:     slow_case_addr = jni_GetIntField_addr();     break;
136     default:        ShouldNotReachHere();
137   }
138   // tail call
139   __ jump (ExternalAddress(slow_case_addr));
140 
141   __ flush ();
142 
143 #ifndef _WINDOWS
144   return fast_entry;
145 #else
146   switch (type) {
147   case T_BOOLEAN: jni_fast_GetBooleanField_fp = (GetBooleanField_t) fast_entry; break;
148   case T_BYTE:    jni_fast_GetByteField_fp    = (GetByteField_t)    fast_entry; break;
149   case T_CHAR:    jni_fast_GetCharField_fp    = (GetCharField_t)    fast_entry; break;
150   case T_SHORT:   jni_fast_GetShortField_fp   = (GetShortField_t)   fast_entry; break;
151   case T_INT:     jni_fast_GetIntField_fp     = (GetIntField_t)     fast_entry; break;
152   }
153   return os::win32::fast_jni_accessor_wrapper(type);
154 #endif
155 }
156 
generate_fast_get_boolean_field()157 address JNI_FastGetField::generate_fast_get_boolean_field() {
158   return generate_fast_get_int_field0(T_BOOLEAN);
159 }
160 
generate_fast_get_byte_field()161 address JNI_FastGetField::generate_fast_get_byte_field() {
162   return generate_fast_get_int_field0(T_BYTE);
163 }
164 
generate_fast_get_char_field()165 address JNI_FastGetField::generate_fast_get_char_field() {
166   return generate_fast_get_int_field0(T_CHAR);
167 }
168 
generate_fast_get_short_field()169 address JNI_FastGetField::generate_fast_get_short_field() {
170   return generate_fast_get_int_field0(T_SHORT);
171 }
172 
generate_fast_get_int_field()173 address JNI_FastGetField::generate_fast_get_int_field() {
174   return generate_fast_get_int_field0(T_INT);
175 }
176 
generate_fast_get_long_field()177 address JNI_FastGetField::generate_fast_get_long_field() {
178   const char *name = "jni_fast_GetLongField";
179   ResourceMark rm;
180   BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
181   CodeBuffer cbuf(blob);
182   MacroAssembler* masm = new MacroAssembler(&cbuf);
183   address fast_entry = __ pc();
184 
185   Label slow;
186 
187   // stack layout:    offset from rsp (in words):
188   //  old rsi          0
189   //  return pc        1
190   //  jni env          2
191   //  obj              3
192   //  jfieldID         4
193 
194   ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
195 
196   __ push  (rsi);
197   __ mov32 (rcx, counter);
198   __ testb (rcx, 1);
199   __ jcc (Assembler::notZero, slow);
200 
201   if (JvmtiExport::can_post_field_access()) {
202     // Check to see if a field access watch has been set before we
203     // take the fast path.
204     __ cmp32(ExternalAddress((address) JvmtiExport::get_field_access_count_addr()), 0);
205     __ jcc(Assembler::notZero, slow);
206   }
207 
208   __ mov(rax, rcx);
209   __ andptr(rax, 1);                         // rax, must end up 0
210   __ movptr(rdx, Address(rsp, rax, Address::times_1, 3*wordSize));
211                                             // obj, notice rax, is 0.
212                                             // rdx is data dependent on rcx.
213   __ movptr(rsi, Address(rsp, 4*wordSize));  // jfieldID
214 
215   __ clear_jweak_tag(rdx);
216 
217   __ movptr(rdx, Address(rdx, 0));           // *obj
218   __ shrptr(rsi, 2);                         // offset
219 
220   assert(count < LIST_CAPACITY-1, "LIST_CAPACITY too small");
221   speculative_load_pclist[count++] = __ pc();
222   __ movptr(rax, Address(rdx, rsi, Address::times_1));
223 #ifndef _LP64
224   speculative_load_pclist[count] = __ pc();
225   __ movl(rdx, Address(rdx, rsi, Address::times_1, 4));
226 #endif // _LP64
227 
228   __ lea(rsi, counter);
229   __ xorptr(rsi, rdx);
230   __ xorptr(rsi, rax);
231   __ xorptr(rsi, rdx);
232   __ xorptr(rsi, rax);
233   __ cmp32(rcx, Address(rsi, 0));
234   // ca1 is the same as ca because
235   // rax, ^ rdx ^ counter_addr ^ rax, ^ rdx = address
236   // ca1 is data dependent on both rax, and rdx.
237   __ jcc (Assembler::notEqual, slow);
238 
239   __ pop (rsi);
240 
241 #ifndef _WINDOWS
242   __ ret (0);
243 #else
244   // __stdcall calling convention
245   __ ret (3*wordSize);
246 #endif
247 
248   slowcase_entry_pclist[count-1] = __ pc();
249   slowcase_entry_pclist[count++] = __ pc();
250   __ bind (slow);
251   __ pop  (rsi);
252   address slow_case_addr = jni_GetLongField_addr();;
253   // tail call
254   __ jump (ExternalAddress(slow_case_addr));
255 
256   __ flush ();
257 
258 #ifndef _WINDOWS
259   return fast_entry;
260 #else
261   jni_fast_GetLongField_fp = (GetLongField_t) fast_entry;
262   return os::win32::fast_jni_accessor_wrapper(T_LONG);
263 #endif
264 }
265 
generate_fast_get_float_field0(BasicType type)266 address JNI_FastGetField::generate_fast_get_float_field0(BasicType type) {
267   const char *name = NULL;
268   switch (type) {
269     case T_FLOAT:  name = "jni_fast_GetFloatField";  break;
270     case T_DOUBLE: name = "jni_fast_GetDoubleField"; break;
271     default:       ShouldNotReachHere();
272   }
273   ResourceMark rm;
274   BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
275   CodeBuffer cbuf(blob);
276   MacroAssembler* masm = new MacroAssembler(&cbuf);
277   address fast_entry = __ pc();
278 
279   Label slow_with_pop, slow;
280 
281   // stack layout:    offset from rsp (in words):
282   //  return pc        0
283   //  jni env          1
284   //  obj              2
285   //  jfieldID         3
286 
287   ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
288 
289   __ mov32 (rcx, counter);
290   __ testb (rcx, 1);
291   __ jcc (Assembler::notZero, slow);
292 
293   if (JvmtiExport::can_post_field_access()) {
294     // Check to see if a field access watch has been set before we
295     // take the fast path.
296     __ cmp32(ExternalAddress((address) JvmtiExport::get_field_access_count_addr()), 0);
297     __ jcc(Assembler::notZero, slow);
298   }
299 
300   __ mov(rax, rcx);
301   __ andptr(rax, 1);                         // rax, must end up 0
302   __ movptr(rdx, Address(rsp, rax, Address::times_1, 2*wordSize));
303                                             // obj, notice rax, is 0.
304                                             // rdx is data dependent on rcx.
305   __ movptr(rax, Address(rsp, 3*wordSize));  // jfieldID
306 
307   __ clear_jweak_tag(rdx);
308 
309   __ movptr(rdx, Address(rdx, 0));           // *obj
310   __ shrptr(rax, 2);                         // offset
311 
312   assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
313   speculative_load_pclist[count] = __ pc();
314   switch (type) {
315 #ifndef _LP64
316     case T_FLOAT:  __ fld_s (Address(rdx, rax, Address::times_1)); break;
317     case T_DOUBLE: __ fld_d (Address(rdx, rax, Address::times_1)); break;
318 #else
319     case T_FLOAT:  __ movflt (xmm0, Address(robj, roffset, Address::times_1)); break;
320     case T_DOUBLE: __ movdbl (xmm0, Address(robj, roffset, Address::times_1)); break;
321 #endif // _LP64
322     default:       ShouldNotReachHere();
323   }
324 
325   Address ca1;
326   __ fst_s (Address(rsp, -4));
327   __ lea(rdx, counter);
328   __ movl (rax, Address(rsp, -4));
329   // garbage hi-order bits on 64bit are harmless.
330   __ xorptr(rdx, rax);
331   __ xorptr(rdx, rax);
332   __ cmp32(rcx, Address(rdx, 0));
333   // rax, ^ counter_addr ^ rax, = address
334   // ca1 is data dependent on the field
335   // access.
336   __ jcc (Assembler::notEqual, slow_with_pop);
337 
338 #ifndef _WINDOWS
339   __ ret (0);
340 #else
341   // __stdcall calling convention
342   __ ret (3*wordSize);
343 #endif
344 
345   __ bind (slow_with_pop);
346   // invalid load. pop FPU stack.
347   __ fstp_d (0);
348 
349   slowcase_entry_pclist[count++] = __ pc();
350   __ bind (slow);
351   address slow_case_addr = NULL;
352   switch (type) {
353     case T_FLOAT:  slow_case_addr = jni_GetFloatField_addr();  break;
354     case T_DOUBLE: slow_case_addr = jni_GetDoubleField_addr(); break;
355     default:       ShouldNotReachHere();
356   }
357   // tail call
358   __ jump (ExternalAddress(slow_case_addr));
359 
360   __ flush ();
361 
362 #ifndef _WINDOWS
363   return fast_entry;
364 #else
365   switch (type) {
366   case T_FLOAT:  jni_fast_GetFloatField_fp  = (GetFloatField_t)  fast_entry; break;
367   case T_DOUBLE: jni_fast_GetDoubleField_fp = (GetDoubleField_t) fast_entry; break;
368   }
369   return os::win32::fast_jni_accessor_wrapper(type);
370 #endif
371 }
372 
generate_fast_get_float_field()373 address JNI_FastGetField::generate_fast_get_float_field() {
374   return generate_fast_get_float_field0(T_FLOAT);
375 }
376 
generate_fast_get_double_field()377 address JNI_FastGetField::generate_fast_get_double_field() {
378   return generate_fast_get_float_field0(T_DOUBLE);
379 }
380