1 /*
2 * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016, 2024 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "registerSaver_s390.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/barrierSetNMethod.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "memory/universe.hpp"
35 #include "nativeInst_s390.hpp"
36 #include "oops/instanceOop.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/methodHandles.hpp"
40 #include "prims/upcallLinker.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/handles.inline.hpp"
43 #include "runtime/javaThread.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/stubCodeGenerator.hpp"
46 #include "runtime/stubRoutines.hpp"
47 #include "utilities/formatBuffer.hpp"
48 #include "utilities/macros.hpp"
49 #include "utilities/powerOfTwo.hpp"
50
51 // Declaration and definition of StubGenerator (no .hpp file).
52 // For a more detailed description of the stub routine structure
53 // see the comment in stubRoutines.hpp.
54
55 #ifdef PRODUCT
56 #define __ _masm->
57 #else
58 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
59 #endif
60
61 #define BLOCK_COMMENT(str) if (PrintAssembly || PrintStubCode) __ block_comment(str)
62 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
63
64
65 // These static, partially const, variables are for the AES intrinsics.
66 // They are declared/initialized here to make them available across function bodies.
67
68 static const int AES_parmBlk_align = 32; // octoword alignment.
69 static const int AES_stackSpace_incr = AES_parmBlk_align; // add'l stack space is allocated in such increments.
70 // Must be multiple of AES_parmBlk_align.
71
72 static int AES_ctrVal_len = 0; // ctr init value len (in bytes), expected: length of dataBlk (16)
73 static int AES_ctrVec_len = 0; // # of ctr vector elements. That many block can be ciphered with one instruction execution
74 static int AES_ctrArea_len = 0; // reserved stack space (in bytes) for ctr (= ctrVal_len * ctrVec_len)
75
76 static int AES_parmBlk_addspace = 0; // Must be multiple of AES_parmblk_align.
77 // Will be set by stub generator to stub specific value.
78 static int AES_dataBlk_space = 0; // Must be multiple of AES_parmblk_align.
79 // Will be set by stub generator to stub specific value.
80 static int AES_dataBlk_offset = 0; // offset of the local src and dst dataBlk buffers
81 // Will be set by stub generator to stub specific value.
82
83 // These offsets are relative to the parameter block address (Register parmBlk = Z_R1)
84 static const int keylen_offset = -1;
85 static const int fCode_offset = -2;
86 static const int ctrVal_len_offset = -4;
87 static const int msglen_offset = -8;
88 static const int unextSP_offset = -16;
89 static const int rem_msgblk_offset = -20;
90 static const int argsave_offset = -2*AES_parmBlk_align;
91 static const int regsave_offset = -4*AES_parmBlk_align; // save space for work regs (Z_R10..13)
92 static const int msglen_red_offset = regsave_offset + AES_parmBlk_align; // reduced len after preLoop;
93 static const int counter_offset = msglen_red_offset+8; // current counter vector position.
94 static const int localSpill_offset = argsave_offset + 24; // arg2..arg4 are saved
95
96
97 // -----------------------------------------------------------------------
98 // Stub Code definitions
99
100 class StubGenerator: public StubCodeGenerator {
101 private:
102
103 //----------------------------------------------------------------------
104 // Call stubs are used to call Java from C.
105
106 //
107 // Arguments:
108 //
109 // R2 - call wrapper address : address
110 // R3 - result : intptr_t*
111 // R4 - result type : BasicType
112 // R5 - method : method
113 // R6 - frame mgr entry point : address
114 // [SP+160] - parameter block : intptr_t*
115 // [SP+172] - parameter count in words : int
116 // [SP+176] - thread : Thread*
117 //
118 address generate_call_stub(address& return_address) {
119 // Set up a new C frame, copy Java arguments, call frame manager
120 // or native_entry, and process result.
121
122 StubCodeMark mark(this, "StubRoutines", "call_stub");
123 address start = __ pc();
124
125 Register r_arg_call_wrapper_addr = Z_ARG1;
126 Register r_arg_result_addr = Z_ARG2;
127 Register r_arg_result_type = Z_ARG3;
128 Register r_arg_method = Z_ARG4;
129 Register r_arg_entry = Z_ARG5;
130
131 // offsets to fp
132 #define d_arg_thread 176
133 #define d_arg_argument_addr 160
134 #define d_arg_argument_count 168+4
135
136 Register r_entryframe_fp = Z_tmp_1;
137 Register r_top_of_arguments_addr = Z_ARG4;
138 Register r_new_arg_entry = Z_R14;
139
140 // macros for frame offsets
141 #define call_wrapper_address_offset \
142 _z_entry_frame_locals_neg(call_wrapper_address)
143 #define result_address_offset \
144 _z_entry_frame_locals_neg(result_address)
145 #define result_type_offset \
146 _z_entry_frame_locals_neg(result_type)
147 #define arguments_tos_address_offset \
148 _z_entry_frame_locals_neg(arguments_tos_address)
149
150 {
151 //
152 // STACK on entry to call_stub:
153 //
154 // F1 [C_FRAME]
155 // ...
156 //
157
158 Register r_argument_addr = Z_tmp_3;
159 Register r_argumentcopy_addr = Z_tmp_4;
160 Register r_argument_size_in_bytes = Z_ARG5;
161 Register r_frame_size = Z_R1;
162
163 Label arguments_copied;
164
165 // Save non-volatile registers to ABI of caller frame.
166 BLOCK_COMMENT("save registers, push frame {");
167 __ z_stmg(Z_R6, Z_R14, 16, Z_SP);
168 __ z_std(Z_F8, 96, Z_SP);
169 __ z_std(Z_F9, 104, Z_SP);
170 __ z_std(Z_F10, 112, Z_SP);
171 __ z_std(Z_F11, 120, Z_SP);
172 __ z_std(Z_F12, 128, Z_SP);
173 __ z_std(Z_F13, 136, Z_SP);
174 __ z_std(Z_F14, 144, Z_SP);
175 __ z_std(Z_F15, 152, Z_SP);
176
177 //
178 // Push ENTRY_FRAME including arguments:
179 //
180 // F0 [TOP_IJAVA_FRAME_ABI]
181 // [outgoing Java arguments]
182 // [ENTRY_FRAME_LOCALS]
183 // F1 [C_FRAME]
184 // ...
185 //
186
187 // Calculate new frame size and push frame.
188 #define abi_plus_locals_size \
189 (frame::z_top_ijava_frame_abi_size + frame::z_entry_frame_locals_size)
190 if (abi_plus_locals_size % BytesPerWord == 0) {
191 // Preload constant part of frame size.
192 __ load_const_optimized(r_frame_size, -abi_plus_locals_size/BytesPerWord);
193 // Keep copy of our frame pointer (caller's SP).
194 __ z_lgr(r_entryframe_fp, Z_SP);
195 // Add space required by arguments to frame size.
196 __ z_slgf(r_frame_size, d_arg_argument_count, Z_R0, Z_SP);
197 // Move Z_ARG5 early, it will be used as a local.
198 __ z_lgr(r_new_arg_entry, r_arg_entry);
199 // Convert frame size from words to bytes.
200 __ z_sllg(r_frame_size, r_frame_size, LogBytesPerWord);
201 __ push_frame(r_frame_size, r_entryframe_fp,
202 false/*don't copy SP*/, true /*frame size sign inverted*/);
203 } else {
204 guarantee(false, "frame sizes should be multiples of word size (BytesPerWord)");
205 }
206 BLOCK_COMMENT("} save, push");
207
208 // Load argument registers for call.
209 BLOCK_COMMENT("prepare/copy arguments {");
210 __ z_lgr(Z_method, r_arg_method);
211 __ z_lg(Z_thread, d_arg_thread, r_entryframe_fp);
212
213 // Calculate top_of_arguments_addr which will be tos (not prepushed) later.
214 // Wimply use SP + frame::top_ijava_frame_size.
215 __ add2reg(r_top_of_arguments_addr,
216 frame::z_top_ijava_frame_abi_size - BytesPerWord, Z_SP);
217
218 // Initialize call_stub locals (step 1).
219 if ((call_wrapper_address_offset + BytesPerWord == result_address_offset) &&
220 (result_address_offset + BytesPerWord == result_type_offset) &&
221 (result_type_offset + BytesPerWord == arguments_tos_address_offset)) {
222
223 __ z_stmg(r_arg_call_wrapper_addr, r_top_of_arguments_addr,
224 call_wrapper_address_offset, r_entryframe_fp);
225 } else {
226 __ z_stg(r_arg_call_wrapper_addr,
227 call_wrapper_address_offset, r_entryframe_fp);
228 __ z_stg(r_arg_result_addr,
229 result_address_offset, r_entryframe_fp);
230 __ z_stg(r_arg_result_type,
231 result_type_offset, r_entryframe_fp);
232 __ z_stg(r_top_of_arguments_addr,
233 arguments_tos_address_offset, r_entryframe_fp);
234 }
235
236 // Copy Java arguments.
237
238 // Any arguments to copy?
239 __ load_and_test_int2long(Z_R1, Address(r_entryframe_fp, d_arg_argument_count));
240 __ z_bre(arguments_copied);
241
242 // Prepare loop and copy arguments in reverse order.
243 {
244 // Calculate argument size in bytes.
245 __ z_sllg(r_argument_size_in_bytes, Z_R1, LogBytesPerWord);
246
247 // Get addr of first incoming Java argument.
248 __ z_lg(r_argument_addr, d_arg_argument_addr, r_entryframe_fp);
249
250 // Let r_argumentcopy_addr point to last outgoing Java argument.
251 __ add2reg(r_argumentcopy_addr, BytesPerWord, r_top_of_arguments_addr); // = Z_SP+160 effectively.
252
253 // Let r_argument_addr point to last incoming Java argument.
254 __ add2reg_with_index(r_argument_addr, -BytesPerWord,
255 r_argument_size_in_bytes, r_argument_addr);
256
257 // Now loop while Z_R1 > 0 and copy arguments.
258 {
259 Label next_argument;
260 __ bind(next_argument);
261 // Mem-mem move.
262 __ z_mvc(0, BytesPerWord-1, r_argumentcopy_addr, 0, r_argument_addr);
263 __ add2reg(r_argument_addr, -BytesPerWord);
264 __ add2reg(r_argumentcopy_addr, BytesPerWord);
265 __ z_brct(Z_R1, next_argument);
266 }
267 } // End of argument copy loop.
268
269 __ bind(arguments_copied);
270 }
271 BLOCK_COMMENT("} arguments");
272
273 BLOCK_COMMENT("call {");
274 {
275 // Call frame manager or native entry.
276
277 //
278 // Register state on entry to frame manager / native entry:
279 //
280 // Z_ARG1 = r_top_of_arguments_addr - intptr_t *sender tos (prepushed)
281 // Lesp = (SP) + copied_arguments_offset - 8
282 // Z_method - method
283 // Z_thread - JavaThread*
284 //
285
286 // Here, the usual SP is the initial_caller_sp.
287 __ z_lgr(Z_R10, Z_SP);
288
289 // Z_esp points to the slot below the last argument.
290 __ z_lgr(Z_esp, r_top_of_arguments_addr);
291
292 //
293 // Stack on entry to frame manager / native entry:
294 //
295 // F0 [TOP_IJAVA_FRAME_ABI]
296 // [outgoing Java arguments]
297 // [ENTRY_FRAME_LOCALS]
298 // F1 [C_FRAME]
299 // ...
300 //
301
302 // Do a light-weight C-call here, r_new_arg_entry holds the address
303 // of the interpreter entry point (frame manager or native entry)
304 // and save runtime-value of return_pc in return_address
305 // (call by reference argument).
306 return_address = __ call_stub(r_new_arg_entry);
307 }
308 BLOCK_COMMENT("} call");
309
310 {
311 BLOCK_COMMENT("restore registers {");
312 // Returned from frame manager or native entry.
313 // Now pop frame, process result, and return to caller.
314
315 //
316 // Stack on exit from frame manager / native entry:
317 //
318 // F0 [ABI]
319 // ...
320 // [ENTRY_FRAME_LOCALS]
321 // F1 [C_FRAME]
322 // ...
323 //
324 // Just pop the topmost frame ...
325 //
326
327 // Restore frame pointer.
328 __ z_lg(r_entryframe_fp, _z_abi(callers_sp), Z_SP);
329 // Pop frame. Done here to minimize stalls.
330 __ pop_frame();
331
332 // Reload some volatile registers which we've spilled before the call
333 // to frame manager / native entry.
334 // Access all locals via frame pointer, because we know nothing about
335 // the topmost frame's size.
336 __ z_lg(r_arg_result_addr, result_address_offset, r_entryframe_fp);
337 __ z_lg(r_arg_result_type, result_type_offset, r_entryframe_fp);
338
339 // Restore non-volatiles.
340 __ z_lmg(Z_R6, Z_R14, 16, Z_SP);
341 __ z_ld(Z_F8, 96, Z_SP);
342 __ z_ld(Z_F9, 104, Z_SP);
343 __ z_ld(Z_F10, 112, Z_SP);
344 __ z_ld(Z_F11, 120, Z_SP);
345 __ z_ld(Z_F12, 128, Z_SP);
346 __ z_ld(Z_F13, 136, Z_SP);
347 __ z_ld(Z_F14, 144, Z_SP);
348 __ z_ld(Z_F15, 152, Z_SP);
349 BLOCK_COMMENT("} restore");
350
351 //
352 // Stack on exit from call_stub:
353 //
354 // 0 [C_FRAME]
355 // ...
356 //
357 // No call_stub frames left.
358 //
359
360 // All non-volatiles have been restored at this point!!
361
362 //------------------------------------------------------------------------
363 // The following code makes some assumptions on the T_<type> enum values.
364 // The enum is defined in globalDefinitions.hpp.
365 // The validity of the assumptions is tested as far as possible.
366 // The assigned values should not be shuffled
367 // T_BOOLEAN==4 - lowest used enum value
368 // T_NARROWOOP==16 - largest used enum value
369 //------------------------------------------------------------------------
370 BLOCK_COMMENT("process result {");
371 Label firstHandler;
372 int handlerLen= 8;
373 #ifdef ASSERT
374 char assertMsg[] = "check BasicType definition in globalDefinitions.hpp";
375 __ z_chi(r_arg_result_type, T_BOOLEAN);
376 __ asm_assert(Assembler::bcondNotLow, assertMsg, 0x0234);
377 __ z_chi(r_arg_result_type, T_NARROWOOP);
378 __ asm_assert(Assembler::bcondNotHigh, assertMsg, 0x0235);
379 #endif
380 __ add2reg(r_arg_result_type, -T_BOOLEAN); // Remove offset.
381 __ z_larl(Z_R1, firstHandler); // location of first handler
382 __ z_sllg(r_arg_result_type, r_arg_result_type, 3); // Each handler is 8 bytes long.
383 __ z_bc(MacroAssembler::bcondAlways, 0, r_arg_result_type, Z_R1);
384
385 __ align(handlerLen);
386 __ bind(firstHandler);
387 // T_BOOLEAN:
388 guarantee(T_BOOLEAN == 4, "check BasicType definition in globalDefinitions.hpp");
389 __ z_st(Z_RET, 0, r_arg_result_addr);
390 __ z_br(Z_R14); // Return to caller.
391 __ align(handlerLen);
392 // T_CHAR:
393 guarantee(T_CHAR == T_BOOLEAN+1, "check BasicType definition in globalDefinitions.hpp");
394 __ z_st(Z_RET, 0, r_arg_result_addr);
395 __ z_br(Z_R14); // Return to caller.
396 __ align(handlerLen);
397 // T_FLOAT:
398 guarantee(T_FLOAT == T_CHAR+1, "check BasicType definition in globalDefinitions.hpp");
399 __ z_ste(Z_FRET, 0, r_arg_result_addr);
400 __ z_br(Z_R14); // Return to caller.
401 __ align(handlerLen);
402 // T_DOUBLE:
403 guarantee(T_DOUBLE == T_FLOAT+1, "check BasicType definition in globalDefinitions.hpp");
404 __ z_std(Z_FRET, 0, r_arg_result_addr);
405 __ z_br(Z_R14); // Return to caller.
406 __ align(handlerLen);
407 // T_BYTE:
408 guarantee(T_BYTE == T_DOUBLE+1, "check BasicType definition in globalDefinitions.hpp");
409 __ z_st(Z_RET, 0, r_arg_result_addr);
410 __ z_br(Z_R14); // Return to caller.
411 __ align(handlerLen);
412 // T_SHORT:
413 guarantee(T_SHORT == T_BYTE+1, "check BasicType definition in globalDefinitions.hpp");
414 __ z_st(Z_RET, 0, r_arg_result_addr);
415 __ z_br(Z_R14); // Return to caller.
416 __ align(handlerLen);
417 // T_INT:
418 guarantee(T_INT == T_SHORT+1, "check BasicType definition in globalDefinitions.hpp");
419 __ z_st(Z_RET, 0, r_arg_result_addr);
420 __ z_br(Z_R14); // Return to caller.
421 __ align(handlerLen);
422 // T_LONG:
423 guarantee(T_LONG == T_INT+1, "check BasicType definition in globalDefinitions.hpp");
424 __ z_stg(Z_RET, 0, r_arg_result_addr);
425 __ z_br(Z_R14); // Return to caller.
426 __ align(handlerLen);
427 // T_OBJECT:
428 guarantee(T_OBJECT == T_LONG+1, "check BasicType definition in globalDefinitions.hpp");
429 __ z_stg(Z_RET, 0, r_arg_result_addr);
430 __ z_br(Z_R14); // Return to caller.
431 __ align(handlerLen);
432 // T_ARRAY:
433 guarantee(T_ARRAY == T_OBJECT+1, "check BasicType definition in globalDefinitions.hpp");
434 __ z_stg(Z_RET, 0, r_arg_result_addr);
435 __ z_br(Z_R14); // Return to caller.
436 __ align(handlerLen);
437 // T_VOID:
438 guarantee(T_VOID == T_ARRAY+1, "check BasicType definition in globalDefinitions.hpp");
439 __ z_stg(Z_RET, 0, r_arg_result_addr);
440 __ z_br(Z_R14); // Return to caller.
441 __ align(handlerLen);
442 // T_ADDRESS:
443 guarantee(T_ADDRESS == T_VOID+1, "check BasicType definition in globalDefinitions.hpp");
444 __ z_stg(Z_RET, 0, r_arg_result_addr);
445 __ z_br(Z_R14); // Return to caller.
446 __ align(handlerLen);
447 // T_NARROWOOP:
448 guarantee(T_NARROWOOP == T_ADDRESS+1, "check BasicType definition in globalDefinitions.hpp");
449 __ z_st(Z_RET, 0, r_arg_result_addr);
450 __ z_br(Z_R14); // Return to caller.
451 __ align(handlerLen);
452 BLOCK_COMMENT("} process result");
453 }
454 return start;
455 }
456
457 // Return point for a Java call if there's an exception thrown in
458 // Java code. The exception is caught and transformed into a
459 // pending exception stored in JavaThread that can be tested from
460 // within the VM.
461 address generate_catch_exception() {
462 StubCodeMark mark(this, "StubRoutines", "catch_exception");
463
464 address start = __ pc();
465
466 //
467 // Registers alive
468 //
469 // Z_thread
470 // Z_ARG1 - address of pending exception
471 // Z_ARG2 - return address in call stub
472 //
473
474 const Register exception_file = Z_R0;
475 const Register exception_line = Z_R1;
476
477 __ load_const_optimized(exception_file, (void*)__FILE__);
478 __ load_const_optimized(exception_line, (void*)__LINE__);
479
480 __ z_stg(Z_ARG1, thread_(pending_exception));
481 // Store into `char *'.
482 __ z_stg(exception_file, thread_(exception_file));
483 // Store into `int'.
484 __ z_st(exception_line, thread_(exception_line));
485
486 // Complete return to VM.
487 assert(StubRoutines::_call_stub_return_address != nullptr, "must have been generated before");
488
489 // Continue in call stub.
490 __ z_br(Z_ARG2);
491
492 return start;
493 }
494
495 // Continuation point for runtime calls returning with a pending
496 // exception. The pending exception check happened in the runtime
497 // or native call stub. The pending exception in Thread is
498 // converted into a Java-level exception.
499 //
500 // Read:
501 // Z_R14: pc the runtime library callee wants to return to.
502 // Since the exception occurred in the callee, the return pc
503 // from the point of view of Java is the exception pc.
504 //
505 // Invalidate:
506 // Volatile registers (except below).
507 //
508 // Update:
509 // Z_ARG1: exception
510 // (Z_R14 is unchanged and is live out).
511 //
512 address generate_forward_exception() {
513 StubCodeMark mark(this, "StubRoutines", "forward_exception");
514 address start = __ pc();
515
516 #define pending_exception_offset in_bytes(Thread::pending_exception_offset())
517 #ifdef ASSERT
518 // Get pending exception oop.
519 __ z_lg(Z_ARG1, pending_exception_offset, Z_thread);
520
521 // Make sure that this code is only executed if there is a pending exception.
522 {
523 Label L;
524 __ z_ltgr(Z_ARG1, Z_ARG1);
525 __ z_brne(L);
526 __ stop("StubRoutines::forward exception: no pending exception (1)");
527 __ bind(L);
528 }
529
530 __ verify_oop(Z_ARG1, "StubRoutines::forward exception: not an oop");
531 #endif
532
533 __ z_lgr(Z_ARG2, Z_R14); // Copy exception pc into Z_ARG2.
534 __ save_return_pc();
535 __ push_frame_abi160(0);
536 // Find exception handler.
537 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
538 Z_thread,
539 Z_ARG2);
540 // Copy handler's address.
541 __ z_lgr(Z_R1, Z_RET);
542 __ pop_frame();
543 __ restore_return_pc();
544
545 // Set up the arguments for the exception handler:
546 // - Z_ARG1: exception oop
547 // - Z_ARG2: exception pc
548
549 // Load pending exception oop.
550 __ z_lg(Z_ARG1, pending_exception_offset, Z_thread);
551
552 // The exception pc is the return address in the caller,
553 // must load it into Z_ARG2
554 __ z_lgr(Z_ARG2, Z_R14);
555
556 #ifdef ASSERT
557 // Make sure exception is set.
558 { Label L;
559 __ z_ltgr(Z_ARG1, Z_ARG1);
560 __ z_brne(L);
561 __ stop("StubRoutines::forward exception: no pending exception (2)");
562 __ bind(L);
563 }
564 #endif
565 // Clear the pending exception.
566 __ clear_mem(Address(Z_thread, pending_exception_offset), sizeof(void *));
567 // Jump to exception handler
568 __ z_br(Z_R1 /*handler address*/);
569
570 return start;
571
572 #undef pending_exception_offset
573 }
574
575 #undef __
576 #ifdef PRODUCT
577 #define __ _masm->
578 #else
579 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
580 #endif
581
582 // Support for uint StubRoutine::zarch::partial_subtype_check(Klass
583 // sub, Klass super);
584 //
585 // Arguments:
586 // ret : Z_RET, returned
587 // sub : Z_ARG2, argument, not changed
588 // super: Z_ARG3, argument, not changed
589 //
590 // raddr: Z_R14, blown by call
591 //
592 address generate_partial_subtype_check() {
593 StubCodeMark mark(this, "StubRoutines", "partial_subtype_check");
594 Label miss;
595
596 address start = __ pc();
597
598 const Register Rsubklass = Z_ARG2; // subklass
599 const Register Rsuperklass = Z_ARG3; // superklass
600
601 // No args, but tmp registers that are killed.
602 const Register Rlength = Z_ARG4; // cache array length
603 const Register Rarray_ptr = Z_ARG5; // Current value from cache array.
604
605 if (UseCompressedOops) {
606 assert(Universe::heap() != nullptr, "java heap must be initialized to generate partial_subtype_check stub");
607 }
608
609 // Always take the slow path.
610 __ check_klass_subtype_slow_path(Rsubklass, Rsuperklass,
611 Rarray_ptr, Rlength, nullptr, &miss);
612
613 // Match falls through here.
614 __ clear_reg(Z_RET); // Zero indicates a match. Set EQ flag in CC.
615 __ z_br(Z_R14);
616
617 __ BIND(miss);
618 __ load_const_optimized(Z_RET, 1); // One indicates a miss.
619 __ z_ltgr(Z_RET, Z_RET); // Set NE flag in CR.
620 __ z_br(Z_R14);
621
622 return start;
623 }
624
625 address generate_lookup_secondary_supers_table_stub(u1 super_klass_index) {
626 StubCodeMark mark(this, "StubRoutines", "lookup_secondary_supers_table");
627
628 const Register
629 r_super_klass = Z_ARG1,
630 r_sub_klass = Z_ARG2,
631 r_array_index = Z_ARG3,
632 r_array_length = Z_ARG4,
633 r_array_base = Z_ARG5,
634 r_bitmap = Z_R10,
635 r_result = Z_R11;
636 address start = __ pc();
637
638 __ lookup_secondary_supers_table(r_sub_klass, r_super_klass,
639 r_array_base, r_array_length, r_array_index,
640 r_bitmap, r_result, super_klass_index);
641
642 __ z_br(Z_R14);
643
644 return start;
645 }
646
647 // Slow path implementation for UseSecondarySupersTable.
648 address generate_lookup_secondary_supers_table_slow_path_stub() {
649 StubCodeMark mark(this, "StubRoutines", "lookup_secondary_supers_table_slow_path");
650
651 address start = __ pc();
652
653 const Register
654 r_super_klass = Z_ARG1,
655 r_array_base = Z_ARG5,
656 r_temp1 = Z_ARG4,
657 r_array_index = Z_ARG3,
658 r_bitmap = Z_R10,
659 r_result = Z_R11;
660
661 __ lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base,
662 r_array_index, r_bitmap, r_result, r_temp1);
663
664 __ z_br(Z_R14);
665
666 return start;
667 }
668
669 #if !defined(PRODUCT)
670 // Wrapper which calls oopDesc::is_oop_or_null()
671 // Only called by MacroAssembler::verify_oop
672 static void verify_oop_helper(const char* message, oopDesc* o) {
673 if (!oopDesc::is_oop_or_null(o)) {
674 fatal("%s. oop: " PTR_FORMAT, message, p2i(o));
675 }
676 ++ StubRoutines::_verify_oop_count;
677 }
678 #endif
679
680 // Return address of code to be called from code generated by
681 // MacroAssembler::verify_oop.
682 //
683 // Don't generate, rather use C++ code.
684 address generate_verify_oop_subroutine() {
685 // Don't generate a StubCodeMark, because no code is generated!
686 // Generating the mark triggers notifying the oprofile jvmti agent
687 // about the dynamic code generation, but the stub without
688 // code (code_size == 0) confuses opjitconv
689 // StubCodeMark mark(this, "StubRoutines", "verify_oop_stub");
690
691 address start = 0;
692
693 #if !defined(PRODUCT)
694 start = CAST_FROM_FN_PTR(address, verify_oop_helper);
695 #endif
696
697 return start;
698 }
699
700 // This is to test that the count register contains a positive int value.
701 // Required because C2 does not respect int to long conversion for stub calls.
702 void assert_positive_int(Register count) {
703 #ifdef ASSERT
704 __ z_srag(Z_R0, count, 31); // Just leave the sign (must be zero) in Z_R0.
705 __ asm_assert(Assembler::bcondZero, "missing zero extend", 0xAFFE);
706 #endif
707 }
708
709 // Generate overlap test for array copy stubs.
710 // If no actual overlap is detected, control is transferred to the
711 // "normal" copy stub (entry address passed in disjoint_copy_target).
712 // Otherwise, execution continues with the code generated by the
713 // caller of array_overlap_test.
714 //
715 // Input:
716 // Z_ARG1 - from
717 // Z_ARG2 - to
718 // Z_ARG3 - element count
719 void array_overlap_test(address disjoint_copy_target, int log2_elem_size) {
720 __ MacroAssembler::compare_and_branch_optimized(Z_ARG2, Z_ARG1, Assembler::bcondNotHigh,
721 disjoint_copy_target, /*len64=*/true, /*has_sign=*/false);
722
723 Register index = Z_ARG3;
724 if (log2_elem_size > 0) {
725 __ z_sllg(Z_R1, Z_ARG3, log2_elem_size); // byte count
726 index = Z_R1;
727 }
728 __ add2reg_with_index(Z_R1, 0, index, Z_ARG1); // First byte after "from" range.
729
730 __ MacroAssembler::compare_and_branch_optimized(Z_R1, Z_ARG2, Assembler::bcondNotHigh,
731 disjoint_copy_target, /*len64=*/true, /*has_sign=*/false);
732
733 // Destructive overlap: let caller generate code for that.
734 }
735
736 // Generate stub for disjoint array copy. If "aligned" is true, the
737 // "from" and "to" addresses are assumed to be heapword aligned.
738 //
739 // Arguments for generated stub:
740 // from: Z_ARG1
741 // to: Z_ARG2
742 // count: Z_ARG3 treated as signed
743 void generate_disjoint_copy(bool aligned, int element_size,
744 bool branchToEnd,
745 bool restoreArgs) {
746 // This is the zarch specific stub generator for general array copy tasks.
747 // It has the following prereqs and features:
748 //
749 // - No destructive overlap allowed (else unpredictable results).
750 // - Destructive overlap does not exist if the leftmost byte of the target
751 // does not coincide with any of the source bytes (except the leftmost).
752 //
753 // Register usage upon entry:
754 // Z_ARG1 == Z_R2 : address of source array
755 // Z_ARG2 == Z_R3 : address of target array
756 // Z_ARG3 == Z_R4 : length of operands (# of elements on entry)
757 //
758 // Register usage within the generator:
759 // - Z_R0 and Z_R1 are KILLed by the stub routine (target addr/len).
760 // Used as pair register operand in complex moves, scratch registers anyway.
761 // - Z_R5 is KILLed by the stub routine (source register pair addr/len) (even/odd reg).
762 // Same as R0/R1, but no scratch register.
763 // - Z_ARG1, Z_ARG2, Z_ARG3 are USEd but preserved by the stub routine,
764 // but they might get temporarily overwritten.
765
766 Register save_reg = Z_ARG4; // (= Z_R5), holds original target operand address for restore.
767
768 {
769 Register llen_reg = Z_R1; // Holds left operand len (odd reg).
770 Register laddr_reg = Z_R0; // Holds left operand addr (even reg), overlaps with data_reg.
771 Register rlen_reg = Z_R5; // Holds right operand len (odd reg), overlaps with save_reg.
772 Register raddr_reg = Z_R4; // Holds right operand addr (even reg), overlaps with len_reg.
773
774 Register data_reg = Z_R0; // Holds copied data chunk in alignment process and copy loop.
775 Register len_reg = Z_ARG3; // Holds operand len (#elements at entry, #bytes shortly after).
776 Register dst_reg = Z_ARG2; // Holds left (target) operand addr.
777 Register src_reg = Z_ARG1; // Holds right (source) operand addr.
778
779 Label doMVCLOOP, doMVCLOOPcount, doMVCLOOPiterate;
780 Label doMVCUnrolled;
781 NearLabel doMVC, doMVCgeneral, done;
782 Label MVC_template;
783 address pcMVCblock_b, pcMVCblock_e;
784
785 bool usedMVCLE = true;
786 bool usedMVCLOOP = true;
787 bool usedMVCUnrolled = false;
788 bool usedMVC = false;
789 bool usedMVCgeneral = false;
790
791 int stride;
792 Register stride_reg;
793 Register ix_reg;
794
795 assert((element_size<=256) && (256%element_size == 0), "element size must be <= 256, power of 2");
796 unsigned int log2_size = exact_log2(element_size);
797
798 switch (element_size) {
799 case 1: BLOCK_COMMENT("ARRAYCOPY DISJOINT byte {"); break;
800 case 2: BLOCK_COMMENT("ARRAYCOPY DISJOINT short {"); break;
801 case 4: BLOCK_COMMENT("ARRAYCOPY DISJOINT int {"); break;
802 case 8: BLOCK_COMMENT("ARRAYCOPY DISJOINT long {"); break;
803 default: BLOCK_COMMENT("ARRAYCOPY DISJOINT {"); break;
804 }
805
806 assert_positive_int(len_reg);
807
808 BLOCK_COMMENT("preparation {");
809
810 // No copying if len <= 0.
811 if (branchToEnd) {
812 __ compare64_and_branch(len_reg, (intptr_t) 0, Assembler::bcondNotHigh, done);
813 } else {
814 if (VM_Version::has_CompareBranch()) {
815 __ z_cgib(len_reg, 0, Assembler::bcondNotHigh, 0, Z_R14);
816 } else {
817 __ z_ltgr(len_reg, len_reg);
818 __ z_bcr(Assembler::bcondNotPositive, Z_R14);
819 }
820 }
821
822 // Prefetch just one cache line. Speculative opt for short arrays.
823 // Do not use Z_R1 in prefetch. Is undefined here.
824 if (VM_Version::has_Prefetch()) {
825 __ z_pfd(0x01, 0, Z_R0, src_reg); // Fetch access.
826 __ z_pfd(0x02, 0, Z_R0, dst_reg); // Store access.
827 }
828
829 BLOCK_COMMENT("} preparation");
830
831 // Save args only if really needed.
832 // Keep len test local to branch. Is generated only once.
833
834 BLOCK_COMMENT("mode selection {");
835
836 // Special handling for arrays with only a few elements.
837 // Nothing fancy: just an executed MVC.
838 if (log2_size > 0) {
839 __ z_sllg(Z_R1, len_reg, log2_size); // Remember #bytes in Z_R1.
840 }
841 if (element_size != 8) {
842 __ z_cghi(len_reg, 256/element_size);
843 __ z_brnh(doMVC);
844 usedMVC = true;
845 }
846 if (element_size == 8) { // Long and oop arrays are always aligned.
847 __ z_cghi(len_reg, 256/element_size);
848 __ z_brnh(doMVCUnrolled);
849 usedMVCUnrolled = true;
850 }
851
852 // Prefetch another cache line. We, for sure, have more than one line to copy.
853 if (VM_Version::has_Prefetch()) {
854 __ z_pfd(0x01, 256, Z_R0, src_reg); // Fetch access.
855 __ z_pfd(0x02, 256, Z_R0, dst_reg); // Store access.
856 }
857
858 if (restoreArgs) {
859 // Remember entry value of ARG2 to restore all arguments later from that knowledge.
860 __ z_lgr(save_reg, dst_reg);
861 }
862
863 __ z_cghi(len_reg, 4096/element_size);
864 if (log2_size == 0) {
865 __ z_lgr(Z_R1, len_reg); // Init Z_R1 with #bytes
866 }
867 __ z_brnh(doMVCLOOP);
868
869 // Fall through to MVCLE case.
870
871 BLOCK_COMMENT("} mode selection");
872
873 // MVCLE: for long arrays
874 // DW aligned: Best performance for sizes > 4kBytes.
875 // unaligned: Least complex for sizes > 256 bytes.
876 if (usedMVCLE) {
877 BLOCK_COMMENT("mode MVCLE {");
878
879 // Setup registers for mvcle.
880 //__ z_lgr(llen_reg, len_reg);// r1 <- r4 #bytes already in Z_R1, aka llen_reg.
881 __ z_lgr(laddr_reg, dst_reg); // r0 <- r3
882 __ z_lgr(raddr_reg, src_reg); // r4 <- r2
883 __ z_lgr(rlen_reg, llen_reg); // r5 <- r1
884
885 __ MacroAssembler::move_long_ext(laddr_reg, raddr_reg, 0xb0); // special: bypass cache
886 // __ MacroAssembler::move_long_ext(laddr_reg, raddr_reg, 0xb8); // special: Hold data in cache.
887 // __ MacroAssembler::move_long_ext(laddr_reg, raddr_reg, 0);
888
889 if (restoreArgs) {
890 // MVCLE updates the source (Z_R4,Z_R5) and target (Z_R0,Z_R1) register pairs.
891 // Dst_reg (Z_ARG2) and src_reg (Z_ARG1) are left untouched. No restore required.
892 // Len_reg (Z_ARG3) is destroyed and must be restored.
893 __ z_slgr(laddr_reg, dst_reg); // copied #bytes
894 if (log2_size > 0) {
895 __ z_srag(Z_ARG3, laddr_reg, log2_size); // Convert back to #elements.
896 } else {
897 __ z_lgr(Z_ARG3, laddr_reg);
898 }
899 }
900 if (branchToEnd) {
901 __ z_bru(done);
902 } else {
903 __ z_br(Z_R14);
904 }
905 BLOCK_COMMENT("} mode MVCLE");
906 }
907 // No fallthru possible here.
908
909 // MVCUnrolled: for short, aligned arrays.
910
911 if (usedMVCUnrolled) {
912 BLOCK_COMMENT("mode MVC unrolled {");
913 stride = 8;
914
915 // Generate unrolled MVC instructions.
916 for (int ii = 32; ii > 1; ii--) {
917 __ z_mvc(0, ii * stride-1, dst_reg, 0, src_reg); // ii*8 byte copy
918 if (branchToEnd) {
919 __ z_bru(done);
920 } else {
921 __ z_br(Z_R14);
922 }
923 }
924
925 pcMVCblock_b = __ pc();
926 __ z_mvc(0, 1 * stride-1, dst_reg, 0, src_reg); // 8 byte copy
927 if (branchToEnd) {
928 __ z_bru(done);
929 } else {
930 __ z_br(Z_R14);
931 }
932
933 pcMVCblock_e = __ pc();
934 Label MVC_ListEnd;
935 __ bind(MVC_ListEnd);
936
937 // This is an absolute fast path:
938 // - Array len in bytes must be not greater than 256.
939 // - Array len in bytes must be an integer mult of DW
940 // to save expensive handling of trailing bytes.
941 // - Argument restore is not done,
942 // i.e. previous code must not alter arguments (this code doesn't either).
943
944 __ bind(doMVCUnrolled);
945
946 // Avoid mul, prefer shift where possible.
947 // Combine shift right (for #DW) with shift left (for block size).
948 // Set CC for zero test below (asm_assert).
949 // Note: #bytes comes in Z_R1, #DW in len_reg.
950 unsigned int MVCblocksize = pcMVCblock_e - pcMVCblock_b;
951 unsigned int logMVCblocksize = 0xffffffffU; // Pacify compiler ("used uninitialized" warning).
952
953 if (log2_size > 0) { // Len was scaled into Z_R1.
954 switch (MVCblocksize) {
955
956 case 8: logMVCblocksize = 3;
957 __ z_ltgr(Z_R0, Z_R1); // #bytes is index
958 break; // reasonable size, use shift
959
960 case 16: logMVCblocksize = 4;
961 __ z_slag(Z_R0, Z_R1, logMVCblocksize-log2_size);
962 break; // reasonable size, use shift
963
964 default: logMVCblocksize = 0;
965 __ z_ltgr(Z_R0, len_reg); // #DW for mul
966 break; // all other sizes: use mul
967 }
968 } else {
969 guarantee(log2_size, "doMVCUnrolled: only for DW entities");
970 }
971
972 // This test (and branch) is redundant. Previous code makes sure that
973 // - element count > 0
974 // - element size == 8.
975 // Thus, len reg should never be zero here. We insert an asm_assert() here,
976 // just to double-check and to be on the safe side.
977 __ asm_assert(false, "zero len cannot occur", 99);
978
979 __ z_larl(Z_R1, MVC_ListEnd); // Get addr of last instr block.
980 // Avoid mul, prefer shift where possible.
981 if (logMVCblocksize == 0) {
982 __ z_mghi(Z_R0, MVCblocksize);
983 }
984 __ z_slgr(Z_R1, Z_R0);
985 __ z_br(Z_R1);
986 BLOCK_COMMENT("} mode MVC unrolled");
987 }
988 // No fallthru possible here.
989
990 // MVC execute template
991 // Must always generate. Usage may be switched on below.
992 // There is no suitable place after here to put the template.
993 __ bind(MVC_template);
994 __ z_mvc(0,0,dst_reg,0,src_reg); // Instr template, never exec directly!
995
996
997 // MVC Loop: for medium-sized arrays
998
999 // Only for DW aligned arrays (src and dst).
1000 // #bytes to copy must be at least 256!!!
1001 // Non-aligned cases handled separately.
1002 stride = 256;
1003 stride_reg = Z_R1; // Holds #bytes when control arrives here.
1004 ix_reg = Z_ARG3; // Alias for len_reg.
1005
1006
1007 if (usedMVCLOOP) {
1008 BLOCK_COMMENT("mode MVC loop {");
1009 __ bind(doMVCLOOP);
1010
1011 __ z_lcgr(ix_reg, Z_R1); // Ix runs from -(n-2)*stride to 1*stride (inclusive).
1012 __ z_llill(stride_reg, stride);
1013 __ add2reg(ix_reg, 2*stride); // Thus: increment ix by 2*stride.
1014
1015 __ bind(doMVCLOOPiterate);
1016 __ z_mvc(0, stride-1, dst_reg, 0, src_reg);
1017 __ add2reg(dst_reg, stride);
1018 __ add2reg(src_reg, stride);
1019 __ bind(doMVCLOOPcount);
1020 __ z_brxlg(ix_reg, stride_reg, doMVCLOOPiterate);
1021
1022 // Don 't use add2reg() here, since we must set the condition code!
1023 __ z_aghi(ix_reg, -2*stride); // Compensate incr from above: zero diff means "all copied".
1024
1025 if (restoreArgs) {
1026 __ z_lcgr(Z_R1, ix_reg); // Prepare ix_reg for copy loop, #bytes expected in Z_R1.
1027 __ z_brnz(doMVCgeneral); // We're not done yet, ix_reg is not zero.
1028
1029 // ARG1, ARG2, and ARG3 were altered by the code above, so restore them building on save_reg.
1030 __ z_slgr(dst_reg, save_reg); // copied #bytes
1031 __ z_slgr(src_reg, dst_reg); // = ARG1 (now restored)
1032 if (log2_size) {
1033 __ z_srag(Z_ARG3, dst_reg, log2_size); // Convert back to #elements to restore ARG3.
1034 } else {
1035 __ z_lgr(Z_ARG3, dst_reg);
1036 }
1037 __ z_lgr(Z_ARG2, save_reg); // ARG2 now restored.
1038
1039 if (branchToEnd) {
1040 __ z_bru(done);
1041 } else {
1042 __ z_br(Z_R14);
1043 }
1044
1045 } else {
1046 if (branchToEnd) {
1047 __ z_brz(done); // CC set by aghi instr.
1048 } else {
1049 __ z_bcr(Assembler::bcondZero, Z_R14); // We're all done if zero.
1050 }
1051
1052 __ z_lcgr(Z_R1, ix_reg); // Prepare ix_reg for copy loop, #bytes expected in Z_R1.
1053 // __ z_bru(doMVCgeneral); // fallthru
1054 }
1055 usedMVCgeneral = true;
1056 BLOCK_COMMENT("} mode MVC loop");
1057 }
1058 // Fallthru to doMVCgeneral
1059
1060 // MVCgeneral: for short, unaligned arrays, after other copy operations
1061
1062 // Somewhat expensive due to use of EX instruction, but simple.
1063 if (usedMVCgeneral) {
1064 BLOCK_COMMENT("mode MVC general {");
1065 __ bind(doMVCgeneral);
1066
1067 __ add2reg(len_reg, -1, Z_R1); // Get #bytes-1 for EXECUTE.
1068 if (VM_Version::has_ExecuteExtensions()) {
1069 __ z_exrl(len_reg, MVC_template); // Execute MVC with variable length.
1070 } else {
1071 __ z_larl(Z_R1, MVC_template); // Get addr of instr template.
1072 __ z_ex(len_reg, 0, Z_R0, Z_R1); // Execute MVC with variable length.
1073 } // penalty: 9 ticks
1074
1075 if (restoreArgs) {
1076 // ARG1, ARG2, and ARG3 were altered by code executed before, so restore them building on save_reg
1077 __ z_slgr(dst_reg, save_reg); // Copied #bytes without the "doMVCgeneral" chunk
1078 __ z_slgr(src_reg, dst_reg); // = ARG1 (now restored), was not advanced for "doMVCgeneral" chunk
1079 __ add2reg_with_index(dst_reg, 1, len_reg, dst_reg); // Len of executed MVC was not accounted for, yet.
1080 if (log2_size) {
1081 __ z_srag(Z_ARG3, dst_reg, log2_size); // Convert back to #elements to restore ARG3
1082 } else {
1083 __ z_lgr(Z_ARG3, dst_reg);
1084 }
1085 __ z_lgr(Z_ARG2, save_reg); // ARG2 now restored.
1086 }
1087
1088 if (usedMVC) {
1089 if (branchToEnd) {
1090 __ z_bru(done);
1091 } else {
1092 __ z_br(Z_R14);
1093 }
1094 } else {
1095 if (!branchToEnd) __ z_br(Z_R14);
1096 }
1097 BLOCK_COMMENT("} mode MVC general");
1098 }
1099 // Fallthru possible if following block not generated.
1100
1101 // MVC: for short, unaligned arrays
1102
1103 // Somewhat expensive due to use of EX instruction, but simple. penalty: 9 ticks.
1104 // Differs from doMVCgeneral in reconstruction of ARG2, ARG3, and ARG4.
1105 if (usedMVC) {
1106 BLOCK_COMMENT("mode MVC {");
1107 __ bind(doMVC);
1108
1109 // get #bytes-1 for EXECUTE
1110 if (log2_size) {
1111 __ add2reg(Z_R1, -1); // Length was scaled into Z_R1.
1112 } else {
1113 __ add2reg(Z_R1, -1, len_reg); // Length was not scaled.
1114 }
1115
1116 if (VM_Version::has_ExecuteExtensions()) {
1117 __ z_exrl(Z_R1, MVC_template); // Execute MVC with variable length.
1118 } else {
1119 __ z_lgr(Z_R0, Z_R5); // Save ARG4, may be unnecessary.
1120 __ z_larl(Z_R5, MVC_template); // Get addr of instr template.
1121 __ z_ex(Z_R1, 0, Z_R0, Z_R5); // Execute MVC with variable length.
1122 __ z_lgr(Z_R5, Z_R0); // Restore ARG4, may be unnecessary.
1123 }
1124
1125 if (!branchToEnd) {
1126 __ z_br(Z_R14);
1127 }
1128 BLOCK_COMMENT("} mode MVC");
1129 }
1130
1131 __ bind(done);
1132
1133 switch (element_size) {
1134 case 1: BLOCK_COMMENT("} ARRAYCOPY DISJOINT byte "); break;
1135 case 2: BLOCK_COMMENT("} ARRAYCOPY DISJOINT short"); break;
1136 case 4: BLOCK_COMMENT("} ARRAYCOPY DISJOINT int "); break;
1137 case 8: BLOCK_COMMENT("} ARRAYCOPY DISJOINT long "); break;
1138 default: BLOCK_COMMENT("} ARRAYCOPY DISJOINT "); break;
1139 }
1140 }
1141 }
1142
1143 // Generate stub for conjoint array copy. If "aligned" is true, the
1144 // "from" and "to" addresses are assumed to be heapword aligned.
1145 //
1146 // Arguments for generated stub:
1147 // from: Z_ARG1
1148 // to: Z_ARG2
1149 // count: Z_ARG3 treated as signed
1150 void generate_conjoint_copy(bool aligned, int element_size, bool branchToEnd) {
1151
1152 // This is the zarch specific stub generator for general array copy tasks.
1153 // It has the following prereqs and features:
1154 //
1155 // - Destructive overlap exists and is handled by reverse copy.
1156 // - Destructive overlap exists if the leftmost byte of the target
1157 // does coincide with any of the source bytes (except the leftmost).
1158 // - Z_R0 and Z_R1 are KILLed by the stub routine (data and stride)
1159 // - Z_ARG1 and Z_ARG2 are USEd but preserved by the stub routine.
1160 // - Z_ARG3 is USED but preserved by the stub routine.
1161 // - Z_ARG4 is used as index register and is thus KILLed.
1162 //
1163 {
1164 Register stride_reg = Z_R1; // Stride & compare value in loop (negative element_size).
1165 Register data_reg = Z_R0; // Holds value of currently processed element.
1166 Register ix_reg = Z_ARG4; // Holds byte index of currently processed element.
1167 Register len_reg = Z_ARG3; // Holds length (in #elements) of arrays.
1168 Register dst_reg = Z_ARG2; // Holds left operand addr.
1169 Register src_reg = Z_ARG1; // Holds right operand addr.
1170
1171 assert(256%element_size == 0, "Element size must be power of 2.");
1172 assert(element_size <= 8, "Can't handle more than DW units.");
1173
1174 switch (element_size) {
1175 case 1: BLOCK_COMMENT("ARRAYCOPY CONJOINT byte {"); break;
1176 case 2: BLOCK_COMMENT("ARRAYCOPY CONJOINT short {"); break;
1177 case 4: BLOCK_COMMENT("ARRAYCOPY CONJOINT int {"); break;
1178 case 8: BLOCK_COMMENT("ARRAYCOPY CONJOINT long {"); break;
1179 default: BLOCK_COMMENT("ARRAYCOPY CONJOINT {"); break;
1180 }
1181
1182 assert_positive_int(len_reg);
1183
1184 if (VM_Version::has_Prefetch()) {
1185 __ z_pfd(0x01, 0, Z_R0, src_reg); // Fetch access.
1186 __ z_pfd(0x02, 0, Z_R0, dst_reg); // Store access.
1187 }
1188
1189 unsigned int log2_size = exact_log2(element_size);
1190 if (log2_size) {
1191 __ z_sllg(ix_reg, len_reg, log2_size);
1192 } else {
1193 __ z_lgr(ix_reg, len_reg);
1194 }
1195
1196 // Optimize reverse copy loop.
1197 // Main loop copies DW units which may be unaligned. Unaligned access adds some penalty ticks.
1198 // Unaligned DW access (neither fetch nor store) is DW-atomic, but should be alignment-atomic.
1199 // Preceding the main loop, some bytes are copied to obtain a DW-multiple remaining length.
1200
1201 Label countLoop1;
1202 Label copyLoop1;
1203 Label skipBY;
1204 Label skipHW;
1205 int stride = -8;
1206
1207 __ load_const_optimized(stride_reg, stride); // Prepare for DW copy loop.
1208
1209 if (element_size == 8) // Nothing to do here.
1210 __ z_bru(countLoop1);
1211 else { // Do not generate dead code.
1212 __ z_tmll(ix_reg, 7); // Check the "odd" bits.
1213 __ z_bre(countLoop1); // There are none, very good!
1214 }
1215
1216 if (log2_size == 0) { // Handle leftover Byte.
1217 __ z_tmll(ix_reg, 1);
1218 __ z_bre(skipBY);
1219 __ z_lb(data_reg, -1, ix_reg, src_reg);
1220 __ z_stcy(data_reg, -1, ix_reg, dst_reg);
1221 __ add2reg(ix_reg, -1); // Decrement delayed to avoid AGI.
1222 __ bind(skipBY);
1223 // fallthru
1224 }
1225 if (log2_size <= 1) { // Handle leftover HW.
1226 __ z_tmll(ix_reg, 2);
1227 __ z_bre(skipHW);
1228 __ z_lhy(data_reg, -2, ix_reg, src_reg);
1229 __ z_sthy(data_reg, -2, ix_reg, dst_reg);
1230 __ add2reg(ix_reg, -2); // Decrement delayed to avoid AGI.
1231 __ bind(skipHW);
1232 __ z_tmll(ix_reg, 4);
1233 __ z_bre(countLoop1);
1234 // fallthru
1235 }
1236 if (log2_size <= 2) { // There are just 4 bytes (left) that need to be copied.
1237 __ z_ly(data_reg, -4, ix_reg, src_reg);
1238 __ z_sty(data_reg, -4, ix_reg, dst_reg);
1239 __ add2reg(ix_reg, -4); // Decrement delayed to avoid AGI.
1240 __ z_bru(countLoop1);
1241 }
1242
1243 // Control can never get to here. Never! Never ever!
1244 __ z_illtrap(0x99);
1245 __ bind(copyLoop1);
1246 __ z_lg(data_reg, 0, ix_reg, src_reg);
1247 __ z_stg(data_reg, 0, ix_reg, dst_reg);
1248 __ bind(countLoop1);
1249 __ z_brxhg(ix_reg, stride_reg, copyLoop1);
1250
1251 if (!branchToEnd)
1252 __ z_br(Z_R14);
1253
1254 switch (element_size) {
1255 case 1: BLOCK_COMMENT("} ARRAYCOPY CONJOINT byte "); break;
1256 case 2: BLOCK_COMMENT("} ARRAYCOPY CONJOINT short"); break;
1257 case 4: BLOCK_COMMENT("} ARRAYCOPY CONJOINT int "); break;
1258 case 8: BLOCK_COMMENT("} ARRAYCOPY CONJOINT long "); break;
1259 default: BLOCK_COMMENT("} ARRAYCOPY CONJOINT "); break;
1260 }
1261 }
1262 }
1263
1264 // Generate stub for disjoint byte copy. If "aligned" is true, the
1265 // "from" and "to" addresses are assumed to be heapword aligned.
1266 address generate_disjoint_byte_copy(bool aligned, const char * name) {
1267 StubCodeMark mark(this, "StubRoutines", name);
1268
1269 // This is the zarch specific stub generator for byte array copy.
1270 // Refer to generate_disjoint_copy for a list of prereqs and features:
1271 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1272 generate_disjoint_copy(aligned, 1, false, false);
1273 return __ addr_at(start_off);
1274 }
1275
1276
1277 address generate_disjoint_short_copy(bool aligned, const char * name) {
1278 StubCodeMark mark(this, "StubRoutines", name);
1279 // This is the zarch specific stub generator for short array copy.
1280 // Refer to generate_disjoint_copy for a list of prereqs and features:
1281 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1282 generate_disjoint_copy(aligned, 2, false, false);
1283 return __ addr_at(start_off);
1284 }
1285
1286
1287 address generate_disjoint_int_copy(bool aligned, const char * name) {
1288 StubCodeMark mark(this, "StubRoutines", name);
1289 // This is the zarch specific stub generator for int array copy.
1290 // Refer to generate_disjoint_copy for a list of prereqs and features:
1291 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1292 generate_disjoint_copy(aligned, 4, false, false);
1293 return __ addr_at(start_off);
1294 }
1295
1296
1297 address generate_disjoint_long_copy(bool aligned, const char * name) {
1298 StubCodeMark mark(this, "StubRoutines", name);
1299 // This is the zarch specific stub generator for long array copy.
1300 // Refer to generate_disjoint_copy for a list of prereqs and features:
1301 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1302 generate_disjoint_copy(aligned, 8, false, false);
1303 return __ addr_at(start_off);
1304 }
1305
1306
1307 address generate_disjoint_oop_copy(bool aligned, const char * name, bool dest_uninitialized) {
1308 StubCodeMark mark(this, "StubRoutines", name);
1309 // This is the zarch specific stub generator for oop array copy.
1310 // Refer to generate_disjoint_copy for a list of prereqs and features.
1311 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1312 unsigned int size = UseCompressedOops ? 4 : 8;
1313
1314 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1315 if (dest_uninitialized) {
1316 decorators |= IS_DEST_UNINITIALIZED;
1317 }
1318 if (aligned) {
1319 decorators |= ARRAYCOPY_ALIGNED;
1320 }
1321
1322 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1323 bs->arraycopy_prologue(_masm, decorators, T_OBJECT, Z_ARG1, Z_ARG2, Z_ARG3);
1324
1325 generate_disjoint_copy(aligned, size, true, true);
1326
1327 bs->arraycopy_epilogue(_masm, decorators, T_OBJECT, Z_ARG2, Z_ARG3, true);
1328
1329 return __ addr_at(start_off);
1330 }
1331
1332
1333 address generate_conjoint_byte_copy(bool aligned, const char * name) {
1334 StubCodeMark mark(this, "StubRoutines", name);
1335 // This is the zarch specific stub generator for overlapping byte array copy.
1336 // Refer to generate_conjoint_copy for a list of prereqs and features:
1337 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1338 address nooverlap_target = aligned ? StubRoutines::arrayof_jbyte_disjoint_arraycopy()
1339 : StubRoutines::jbyte_disjoint_arraycopy();
1340
1341 array_overlap_test(nooverlap_target, 0); // Branch away to nooverlap_target if disjoint.
1342 generate_conjoint_copy(aligned, 1, false);
1343
1344 return __ addr_at(start_off);
1345 }
1346
1347
1348 address generate_conjoint_short_copy(bool aligned, const char * name) {
1349 StubCodeMark mark(this, "StubRoutines", name);
1350 // This is the zarch specific stub generator for overlapping short array copy.
1351 // Refer to generate_conjoint_copy for a list of prereqs and features:
1352 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1353 address nooverlap_target = aligned ? StubRoutines::arrayof_jshort_disjoint_arraycopy()
1354 : StubRoutines::jshort_disjoint_arraycopy();
1355
1356 array_overlap_test(nooverlap_target, 1); // Branch away to nooverlap_target if disjoint.
1357 generate_conjoint_copy(aligned, 2, false);
1358
1359 return __ addr_at(start_off);
1360 }
1361
1362 address generate_conjoint_int_copy(bool aligned, const char * name) {
1363 StubCodeMark mark(this, "StubRoutines", name);
1364 // This is the zarch specific stub generator for overlapping int array copy.
1365 // Refer to generate_conjoint_copy for a list of prereqs and features:
1366
1367 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1368 address nooverlap_target = aligned ? StubRoutines::arrayof_jint_disjoint_arraycopy()
1369 : StubRoutines::jint_disjoint_arraycopy();
1370
1371 array_overlap_test(nooverlap_target, 2); // Branch away to nooverlap_target if disjoint.
1372 generate_conjoint_copy(aligned, 4, false);
1373
1374 return __ addr_at(start_off);
1375 }
1376
1377 address generate_conjoint_long_copy(bool aligned, const char * name) {
1378 StubCodeMark mark(this, "StubRoutines", name);
1379 // This is the zarch specific stub generator for overlapping long array copy.
1380 // Refer to generate_conjoint_copy for a list of prereqs and features:
1381
1382 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1383 address nooverlap_target = aligned ? StubRoutines::arrayof_jlong_disjoint_arraycopy()
1384 : StubRoutines::jlong_disjoint_arraycopy();
1385
1386 array_overlap_test(nooverlap_target, 3); // Branch away to nooverlap_target if disjoint.
1387 generate_conjoint_copy(aligned, 8, false);
1388
1389 return __ addr_at(start_off);
1390 }
1391
1392 address generate_conjoint_oop_copy(bool aligned, const char * name, bool dest_uninitialized) {
1393 StubCodeMark mark(this, "StubRoutines", name);
1394 // This is the zarch specific stub generator for overlapping oop array copy.
1395 // Refer to generate_conjoint_copy for a list of prereqs and features.
1396 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1397 unsigned int size = UseCompressedOops ? 4 : 8;
1398 unsigned int shift = UseCompressedOops ? 2 : 3;
1399
1400 address nooverlap_target = aligned ? StubRoutines::arrayof_oop_disjoint_arraycopy(dest_uninitialized)
1401 : StubRoutines::oop_disjoint_arraycopy(dest_uninitialized);
1402
1403 // Branch to disjoint_copy (if applicable) before pre_barrier to avoid double pre_barrier.
1404 array_overlap_test(nooverlap_target, shift); // Branch away to nooverlap_target if disjoint.
1405
1406 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1407 if (dest_uninitialized) {
1408 decorators |= IS_DEST_UNINITIALIZED;
1409 }
1410 if (aligned) {
1411 decorators |= ARRAYCOPY_ALIGNED;
1412 }
1413
1414 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1415 bs->arraycopy_prologue(_masm, decorators, T_OBJECT, Z_ARG1, Z_ARG2, Z_ARG3);
1416
1417 generate_conjoint_copy(aligned, size, true); // Must preserve ARG2, ARG3.
1418
1419 bs->arraycopy_epilogue(_masm, decorators, T_OBJECT, Z_ARG2, Z_ARG3, true);
1420
1421 return __ addr_at(start_off);
1422 }
1423
1424
1425 void generate_arraycopy_stubs() {
1426
1427 // Note: the disjoint stubs must be generated first, some of
1428 // the conjoint stubs use them.
1429 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy (false, "jbyte_disjoint_arraycopy");
1430 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
1431 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_copy (false, "jint_disjoint_arraycopy");
1432 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_copy (false, "jlong_disjoint_arraycopy");
1433 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_oop_copy (false, "oop_disjoint_arraycopy", false);
1434 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_oop_copy (false, "oop_disjoint_arraycopy_uninit", true);
1435
1436 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy (true, "arrayof_jbyte_disjoint_arraycopy");
1437 StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, "arrayof_jshort_disjoint_arraycopy");
1438 StubRoutines::_arrayof_jint_disjoint_arraycopy = generate_disjoint_int_copy (true, "arrayof_jint_disjoint_arraycopy");
1439 StubRoutines::_arrayof_jlong_disjoint_arraycopy = generate_disjoint_long_copy (true, "arrayof_jlong_disjoint_arraycopy");
1440 StubRoutines::_arrayof_oop_disjoint_arraycopy = generate_disjoint_oop_copy (true, "arrayof_oop_disjoint_arraycopy", false);
1441 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = generate_disjoint_oop_copy (true, "arrayof_oop_disjoint_arraycopy_uninit", true);
1442
1443 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy (false, "jbyte_arraycopy");
1444 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
1445 StubRoutines::_jint_arraycopy = generate_conjoint_int_copy (false, "jint_arraycopy");
1446 StubRoutines::_jlong_arraycopy = generate_conjoint_long_copy (false, "jlong_arraycopy");
1447 StubRoutines::_oop_arraycopy = generate_conjoint_oop_copy (false, "oop_arraycopy", false);
1448 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_oop_copy (false, "oop_arraycopy_uninit", true);
1449
1450 StubRoutines::_arrayof_jbyte_arraycopy = generate_conjoint_byte_copy (true, "arrayof_jbyte_arraycopy");
1451 StubRoutines::_arrayof_jshort_arraycopy = generate_conjoint_short_copy(true, "arrayof_jshort_arraycopy");
1452 StubRoutines::_arrayof_jint_arraycopy = generate_conjoint_int_copy (true, "arrayof_jint_arraycopy");
1453 StubRoutines::_arrayof_jlong_arraycopy = generate_conjoint_long_copy (true, "arrayof_jlong_arraycopy");
1454 StubRoutines::_arrayof_oop_arraycopy = generate_conjoint_oop_copy (true, "arrayof_oop_arraycopy", false);
1455 StubRoutines::_arrayof_oop_arraycopy_uninit = generate_conjoint_oop_copy (true, "arrayof_oop_arraycopy_uninit", true);
1456 }
1457
1458 // Call interface for AES_encryptBlock, AES_decryptBlock stubs.
1459 //
1460 // Z_ARG1 - source data block. Ptr to leftmost byte to be processed.
1461 // Z_ARG2 - destination data block. Ptr to leftmost byte to be stored.
1462 // For in-place encryption/decryption, ARG1 and ARG2 can point
1463 // to the same piece of storage.
1464 // Z_ARG3 - Crypto key address (expanded key). The first n bits of
1465 // the expanded key constitute the original AES-<n> key (see below).
1466 //
1467 // Z_RET - return value. First unprocessed byte offset in src buffer.
1468 //
1469 // Some remarks:
1470 // The crypto key, as passed from the caller to these encryption stubs,
1471 // is a so-called expanded key. It is derived from the original key
1472 // by the Rijndael key schedule, see http://en.wikipedia.org/wiki/Rijndael_key_schedule
1473 // With the expanded key, the cipher/decipher task is decomposed in
1474 // multiple, less complex steps, called rounds. Sun SPARC and Intel
1475 // processors obviously implement support for those less complex steps.
1476 // z/Architecture provides instructions for full cipher/decipher complexity.
1477 // Therefore, we need the original, not the expanded key here.
1478 // Luckily, the first n bits of an AES-<n> expanded key are formed
1479 // by the original key itself. That takes us out of trouble. :-)
1480 // The key length (in bytes) relation is as follows:
1481 // original expanded rounds key bit keylen
1482 // key bytes key bytes length in words
1483 // 16 176 11 128 44
1484 // 24 208 13 192 52
1485 // 32 240 15 256 60
1486 //
1487 // The crypto instructions used in the AES* stubs have some specific register requirements.
1488 // Z_R0 holds the crypto function code. Please refer to the KM/KMC instruction
1489 // description in the "z/Architecture Principles of Operation" manual for details.
1490 // Z_R1 holds the parameter block address. The parameter block contains the cryptographic key
1491 // (KM instruction) and the chaining value (KMC instruction).
1492 // dst must designate an even-numbered register, holding the address of the output message.
1493 // src must designate an even/odd register pair, holding the address/length of the original message
1494
1495 // Helper function which generates code to
1496 // - load the function code in register fCode (== Z_R0).
1497 // - load the data block length (depends on cipher function) into register srclen if requested.
1498 // - is_decipher switches between cipher/decipher function codes
1499 // - set_len requests (if true) loading the data block length in register srclen
1500 void generate_load_AES_fCode(Register keylen, Register fCode, Register srclen, bool is_decipher) {
1501
1502 BLOCK_COMMENT("Set fCode {"); {
1503 Label fCode_set;
1504 int mode = is_decipher ? VM_Version::CipherMode::decipher : VM_Version::CipherMode::cipher;
1505 bool identical_dataBlk_len = (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES192_dataBlk)
1506 && (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES256_dataBlk);
1507 // Expanded key length is 44/52/60 * 4 bytes for AES-128/AES-192/AES-256.
1508 __ z_cghi(keylen, 52); // Check only once at the beginning. keylen and fCode may share the same register.
1509
1510 __ z_lghi(fCode, VM_Version::Cipher::_AES128 + mode);
1511 if (!identical_dataBlk_len) {
1512 __ z_lghi(srclen, VM_Version::Cipher::_AES128_dataBlk);
1513 }
1514 __ z_brl(fCode_set); // keyLen < 52: AES128
1515
1516 __ z_lghi(fCode, VM_Version::Cipher::_AES192 + mode);
1517 if (!identical_dataBlk_len) {
1518 __ z_lghi(srclen, VM_Version::Cipher::_AES192_dataBlk);
1519 }
1520 __ z_bre(fCode_set); // keyLen == 52: AES192
1521
1522 __ z_lghi(fCode, VM_Version::Cipher::_AES256 + mode);
1523 if (!identical_dataBlk_len) {
1524 __ z_lghi(srclen, VM_Version::Cipher::_AES256_dataBlk);
1525 }
1526 // __ z_brh(fCode_set); // keyLen < 52: AES128 // fallthru
1527
1528 __ bind(fCode_set);
1529 if (identical_dataBlk_len) {
1530 __ z_lghi(srclen, VM_Version::Cipher::_AES128_dataBlk);
1531 }
1532 }
1533 BLOCK_COMMENT("} Set fCode");
1534 }
1535
1536 // Push a parameter block for the cipher/decipher instruction on the stack.
1537 // Layout of the additional stack space allocated for AES_cipherBlockChaining:
1538 //
1539 // | |
1540 // +--------+ <-- SP before expansion
1541 // | |
1542 // : : alignment loss (part 2), 0..(AES_parmBlk_align-1) bytes
1543 // | |
1544 // +--------+
1545 // | |
1546 // : : space for parameter block, size VM_Version::Cipher::_AES*_parmBlk_C
1547 // | |
1548 // +--------+ <-- parmBlk, octoword-aligned, start of parameter block
1549 // | |
1550 // : : additional stack space for spills etc., size AES_parmBlk_addspace, DW @ Z_SP not usable!!!
1551 // | |
1552 // +--------+ <-- Z_SP + alignment loss, octoword-aligned
1553 // | |
1554 // : : alignment loss (part 1), 0..(AES_parmBlk_align-1) bytes. DW @ Z_SP not usable!!!
1555 // | |
1556 // +--------+ <-- Z_SP after expansion
1557
1558 void generate_push_Block(int dataBlk_len, int parmBlk_len, int crypto_fCode,
1559 Register parmBlk, Register keylen, Register fCode, Register cv, Register key) {
1560
1561 AES_parmBlk_addspace = AES_parmBlk_align; // Must be multiple of AES_parmblk_align.
1562 // spill space for regs etc., don't use DW @SP!
1563 const int cv_len = dataBlk_len;
1564 const int key_len = parmBlk_len - cv_len;
1565 // This len must be known at JIT compile time. Only then are we able to recalc the SP before resize.
1566 // We buy this knowledge by wasting some (up to AES_parmBlk_align) bytes of stack space.
1567 const int resize_len = cv_len + key_len + AES_parmBlk_align + AES_parmBlk_addspace;
1568
1569 // Use parmBlk as temp reg here to hold the frame pointer.
1570 __ resize_frame(-resize_len, parmBlk, true);
1571
1572 // calculate parmBlk address from updated (resized) SP.
1573 __ add2reg(parmBlk, resize_len - (cv_len + key_len), Z_SP);
1574 __ z_nill(parmBlk, (~(AES_parmBlk_align-1)) & 0xffff); // Align parameter block.
1575
1576 // There is room for stuff in the range [parmBlk-AES_parmBlk_addspace+8, parmBlk).
1577 __ z_stg(keylen, -8, parmBlk); // Spill keylen for later use.
1578
1579 // calculate (SP before resize) from updated SP.
1580 __ add2reg(keylen, resize_len, Z_SP); // keylen holds prev SP for now.
1581 __ z_stg(keylen, -16, parmBlk); // Spill prev SP for easy revert.
1582
1583 __ z_mvc(0, cv_len-1, parmBlk, 0, cv); // Copy cv.
1584 __ z_mvc(cv_len, key_len-1, parmBlk, 0, key); // Copy key.
1585 __ z_lghi(fCode, crypto_fCode);
1586 }
1587
1588 // NOTE:
1589 // Before returning, the stub has to copy the chaining value from
1590 // the parmBlk, where it was updated by the crypto instruction, back
1591 // to the chaining value array the address of which was passed in the cv argument.
1592 // As all the available registers are used and modified by KMC, we need to save
1593 // the key length across the KMC instruction. We do so by spilling it to the stack,
1594 // just preceding the parmBlk (at (parmBlk - 8)).
1595 void generate_push_parmBlk(Register keylen, Register fCode, Register parmBlk, Register key, Register cv, bool is_decipher) {
1596 int mode = is_decipher ? VM_Version::CipherMode::decipher : VM_Version::CipherMode::cipher;
1597 Label parmBlk_128, parmBlk_192, parmBlk_256, parmBlk_set;
1598
1599 BLOCK_COMMENT("push parmBlk {");
1600 // We have just three cipher strengths which translates into three
1601 // possible extended key lengths: 44, 52, and 60 bytes.
1602 // We therefore can compare the actual length against the "middle" length
1603 // and get: lt -> len=44, eq -> len=52, gt -> len=60.
1604 __ z_cghi(keylen, 52);
1605 if (VM_Version::has_Crypto_AES128()) { __ z_brl(parmBlk_128); } // keyLen < 52: AES128
1606 if (VM_Version::has_Crypto_AES192()) { __ z_bre(parmBlk_192); } // keyLen == 52: AES192
1607 if (VM_Version::has_Crypto_AES256()) { __ z_brh(parmBlk_256); } // keyLen > 52: AES256
1608
1609 // Security net: requested AES function not available on this CPU.
1610 // NOTE:
1611 // As of now (March 2015), this safety net is not required. JCE policy files limit the
1612 // cryptographic strength of the keys used to 128 bit. If we have AES hardware support
1613 // at all, we have at least AES-128.
1614 __ stop_static("AES key strength not supported by CPU. Use -XX:-UseAES as remedy.", 0);
1615
1616 if (VM_Version::has_Crypto_AES256()) {
1617 __ bind(parmBlk_256);
1618 generate_push_Block(VM_Version::Cipher::_AES256_dataBlk,
1619 VM_Version::Cipher::_AES256_parmBlk_C,
1620 VM_Version::Cipher::_AES256 + mode,
1621 parmBlk, keylen, fCode, cv, key);
1622 if (VM_Version::has_Crypto_AES128() || VM_Version::has_Crypto_AES192()) {
1623 __ z_bru(parmBlk_set); // Fallthru otherwise.
1624 }
1625 }
1626
1627 if (VM_Version::has_Crypto_AES192()) {
1628 __ bind(parmBlk_192);
1629 generate_push_Block(VM_Version::Cipher::_AES192_dataBlk,
1630 VM_Version::Cipher::_AES192_parmBlk_C,
1631 VM_Version::Cipher::_AES192 + mode,
1632 parmBlk, keylen, fCode, cv, key);
1633 if (VM_Version::has_Crypto_AES128()) {
1634 __ z_bru(parmBlk_set); // Fallthru otherwise.
1635 }
1636 }
1637
1638 if (VM_Version::has_Crypto_AES128()) {
1639 __ bind(parmBlk_128);
1640 generate_push_Block(VM_Version::Cipher::_AES128_dataBlk,
1641 VM_Version::Cipher::_AES128_parmBlk_C,
1642 VM_Version::Cipher::_AES128 + mode,
1643 parmBlk, keylen, fCode, cv, key);
1644 // Fallthru
1645 }
1646
1647 __ bind(parmBlk_set);
1648 BLOCK_COMMENT("} push parmBlk");
1649 }
1650
1651 // Pop a parameter block from the stack. The chaining value portion of the parameter block
1652 // is copied back to the cv array as it is needed for subsequent cipher steps.
1653 // The keylen value as well as the original SP (before resizing) was pushed to the stack
1654 // when pushing the parameter block.
1655 void generate_pop_parmBlk(Register keylen, Register parmBlk, Register key, Register cv) {
1656
1657 BLOCK_COMMENT("pop parmBlk {");
1658 bool identical_dataBlk_len = (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES192_dataBlk) &&
1659 (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES256_dataBlk);
1660 if (identical_dataBlk_len) {
1661 int cv_len = VM_Version::Cipher::_AES128_dataBlk;
1662 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1663 } else {
1664 int cv_len;
1665 Label parmBlk_128, parmBlk_192, parmBlk_256, parmBlk_set;
1666 __ z_lg(keylen, -8, parmBlk); // restore keylen
1667 __ z_cghi(keylen, 52);
1668 if (VM_Version::has_Crypto_AES256()) __ z_brh(parmBlk_256); // keyLen > 52: AES256
1669 if (VM_Version::has_Crypto_AES192()) __ z_bre(parmBlk_192); // keyLen == 52: AES192
1670 // if (VM_Version::has_Crypto_AES128()) __ z_brl(parmBlk_128); // keyLen < 52: AES128 // fallthru
1671
1672 // Security net: there is no one here. If we would need it, we should have
1673 // fallen into it already when pushing the parameter block.
1674 if (VM_Version::has_Crypto_AES128()) {
1675 __ bind(parmBlk_128);
1676 cv_len = VM_Version::Cipher::_AES128_dataBlk;
1677 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1678 if (VM_Version::has_Crypto_AES192() || VM_Version::has_Crypto_AES256()) {
1679 __ z_bru(parmBlk_set);
1680 }
1681 }
1682
1683 if (VM_Version::has_Crypto_AES192()) {
1684 __ bind(parmBlk_192);
1685 cv_len = VM_Version::Cipher::_AES192_dataBlk;
1686 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1687 if (VM_Version::has_Crypto_AES256()) {
1688 __ z_bru(parmBlk_set);
1689 }
1690 }
1691
1692 if (VM_Version::has_Crypto_AES256()) {
1693 __ bind(parmBlk_256);
1694 cv_len = VM_Version::Cipher::_AES256_dataBlk;
1695 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1696 // __ z_bru(parmBlk_set); // fallthru
1697 }
1698 __ bind(parmBlk_set);
1699 }
1700 __ z_lg(Z_SP, -16, parmBlk); // Revert resize_frame_absolute. Z_SP saved by push_parmBlk.
1701 BLOCK_COMMENT("} pop parmBlk");
1702 }
1703
1704 // Compute AES encrypt/decrypt function.
1705 void generate_AES_cipherBlock(bool is_decipher) {
1706 // Incoming arguments.
1707 Register from = Z_ARG1; // source byte array
1708 Register to = Z_ARG2; // destination byte array
1709 Register key = Z_ARG3; // expanded key array
1710
1711 const Register keylen = Z_R0; // Temporarily (until fCode is set) holds the expanded key array length.
1712
1713 // Register definitions as required by KM instruction.
1714 const Register fCode = Z_R0; // crypto function code
1715 const Register parmBlk = Z_R1; // parameter block address (points to crypto key)
1716 const Register src = Z_ARG1; // Must be even reg (KM requirement).
1717 const Register srclen = Z_ARG2; // Must be odd reg and pair with src. Overwrites destination address.
1718 const Register dst = Z_ARG3; // Must be even reg (KM requirement). Overwrites expanded key address.
1719
1720 // Read key len of expanded key (in 4-byte words).
1721 __ z_lgf(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1722
1723 // Copy arguments to registers as required by crypto instruction.
1724 __ z_lgr(parmBlk, key); // crypto key (in T_INT array).
1725 __ lgr_if_needed(src, from); // Copy src address. Will not emit, src/from are identical.
1726 __ z_lgr(dst, to); // Copy dst address, even register required.
1727
1728 // Construct function code into fCode(Z_R0), data block length into srclen(Z_ARG2).
1729 generate_load_AES_fCode(keylen, fCode, srclen, is_decipher);
1730
1731 __ km(dst, src); // Cipher the message.
1732
1733 __ z_br(Z_R14);
1734 }
1735
1736 // Compute AES encrypt function.
1737 address generate_AES_encryptBlock(const char* name) {
1738 __ align(CodeEntryAlignment);
1739 StubCodeMark mark(this, "StubRoutines", name);
1740 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1741
1742 generate_AES_cipherBlock(false);
1743
1744 return __ addr_at(start_off);
1745 }
1746
1747 // Compute AES decrypt function.
1748 address generate_AES_decryptBlock(const char* name) {
1749 __ align(CodeEntryAlignment);
1750 StubCodeMark mark(this, "StubRoutines", name);
1751 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1752
1753 generate_AES_cipherBlock(true);
1754
1755 return __ addr_at(start_off);
1756 }
1757
1758 // These stubs receive the addresses of the cryptographic key and of the chaining value as two separate
1759 // arguments (registers "key" and "cv", respectively). The KMC instruction, on the other hand, requires
1760 // chaining value and key to be, in this sequence, adjacent in storage. Thus, we need to allocate some
1761 // thread-local working storage. Using heap memory incurs all the hassles of allocating/freeing.
1762 // Stack space, on the contrary, is deallocated automatically when we return from the stub to the caller.
1763 // *** WARNING ***
1764 // Please note that we do not formally allocate stack space, nor do we
1765 // update the stack pointer. Therefore, no function calls are allowed
1766 // and nobody else must use the stack range where the parameter block
1767 // is located.
1768 // We align the parameter block to the next available octoword.
1769 //
1770 // Compute chained AES encrypt function.
1771 void generate_AES_cipherBlockChaining(bool is_decipher) {
1772
1773 Register from = Z_ARG1; // source byte array (clear text)
1774 Register to = Z_ARG2; // destination byte array (ciphered)
1775 Register key = Z_ARG3; // expanded key array.
1776 Register cv = Z_ARG4; // chaining value
1777 const Register msglen = Z_ARG5; // Total length of the msg to be encrypted. Value must be returned
1778 // in Z_RET upon completion of this stub. Is 32-bit integer.
1779
1780 const Register keylen = Z_R0; // Expanded key length, as read from key array. Temp only.
1781 const Register fCode = Z_R0; // crypto function code
1782 const Register parmBlk = Z_R1; // parameter block address (points to crypto key)
1783 const Register src = Z_ARG1; // is Z_R2
1784 const Register srclen = Z_ARG2; // Overwrites destination address.
1785 const Register dst = Z_ARG3; // Overwrites key address.
1786
1787 // Read key len of expanded key (in 4-byte words).
1788 __ z_lgf(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1789
1790 // Construct parm block address in parmBlk (== Z_R1), copy cv and key to parm block.
1791 // Construct function code in fCode (Z_R0).
1792 generate_push_parmBlk(keylen, fCode, parmBlk, key, cv, is_decipher);
1793
1794 // Prepare other registers for instruction.
1795 __ lgr_if_needed(src, from); // Copy src address. Will not emit, src/from are identical.
1796 __ z_lgr(dst, to);
1797 __ z_llgfr(srclen, msglen); // We pass the offsets as ints, not as longs as required.
1798
1799 __ kmc(dst, src); // Cipher the message.
1800
1801 generate_pop_parmBlk(keylen, parmBlk, key, cv);
1802
1803 __ z_llgfr(Z_RET, msglen); // We pass the offsets as ints, not as longs as required.
1804 __ z_br(Z_R14);
1805 }
1806
1807 // Compute chained AES encrypt function.
1808 address generate_cipherBlockChaining_AES_encrypt(const char* name) {
1809 __ align(CodeEntryAlignment);
1810 StubCodeMark mark(this, "StubRoutines", name);
1811 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1812
1813 generate_AES_cipherBlockChaining(false);
1814
1815 return __ addr_at(start_off);
1816 }
1817
1818 // Compute chained AES decrypt function.
1819 address generate_cipherBlockChaining_AES_decrypt(const char* name) {
1820 __ align(CodeEntryAlignment);
1821 StubCodeMark mark(this, "StubRoutines", name);
1822 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1823
1824 generate_AES_cipherBlockChaining(true);
1825
1826 return __ addr_at(start_off);
1827 }
1828
1829
1830 // *****************************************************************************
1831
1832 // AES CounterMode
1833 // Push a parameter block for the cipher/decipher instruction on the stack.
1834 // Layout of the additional stack space allocated for counterMode_AES_cipherBlock
1835 //
1836 // | |
1837 // +--------+ <-- SP before expansion
1838 // | |
1839 // : : alignment loss (part 2), 0..(AES_parmBlk_align-1) bytes.
1840 // | |
1841 // +--------+ <-- gap = parmBlk + parmBlk_len + ctrArea_len
1842 // | |
1843 // : : byte[] ctr - kmctr expects a counter vector the size of the input vector.
1844 // : : The interface only provides byte[16] iv, the init vector.
1845 // : : The size of this area is a tradeoff between stack space, init effort, and speed.
1846 // | | Each counter is a 128bit int. Vector element [0] is a copy of iv.
1847 // | | Vector element [i] is formed by incrementing element [i-1].
1848 // +--------+ <-- ctr = parmBlk + parmBlk_len
1849 // | |
1850 // : : space for parameter block, size VM_Version::Cipher::_AES*_parmBlk_G
1851 // | |
1852 // +--------+ <-- parmBlk = Z_SP + (alignment loss (part 1+2)) + AES_dataBlk_space + AES_parmBlk_addSpace, octoword-aligned, start of parameter block
1853 // | |
1854 // : : additional stack space for spills etc., min. size AES_parmBlk_addspace, all bytes usable.
1855 // | |
1856 // +--------+ <-- Z_SP + alignment loss (part 1+2) + AES_dataBlk_space, octoword-aligned
1857 // | |
1858 // : : space for one source data block and one dest data block.
1859 // | |
1860 // +--------+ <-- Z_SP + alignment loss (part 1+2), octoword-aligned
1861 // | |
1862 // : : additional alignment loss. Blocks above can't tolerate unusable DW @SP.
1863 // | |
1864 // +--------+ <-- Z_SP + alignment loss (part 1), octoword-aligned
1865 // | |
1866 // : : alignment loss (part 1), 0..(AES_parmBlk_align-1) bytes. DW @ Z_SP holds frame ptr.
1867 // | |
1868 // +--------+ <-- Z_SP after expansion
1869 //
1870 // additional space allocation (per DW):
1871 // spillSpace = parmBlk - AES_parmBlk_addspace
1872 // dataBlocks = spillSpace - AES_dataBlk_space
1873 //
1874 // parmBlk-8 various fields of various lengths
1875 // parmBlk-1: key_len (only one byte is stored at parmBlk-1)
1876 // parmBlk-2: fCode (only one byte is stored at parmBlk-2)
1877 // parmBlk-4: ctrVal_len (as retrieved from iv array), in bytes, as HW
1878 // parmBlk-8: msglen length (in bytes) of crypto msg, as passed in by caller
1879 // return value is calculated from this: rv = msglen - processed.
1880 // parmBlk-16 old_SP (SP before resize)
1881 // parmBlk-24 temp values
1882 // up to and including main loop in generate_counterMode_AES
1883 // - parmBlk-20: remmsg_len remaining msg len (aka unprocessed msg bytes)
1884 // after main loop in generate_counterMode_AES
1885 // - parmBlk-24: spill slot for various address values
1886 //
1887 // parmBlk-40 free spill slot, used for local spills.
1888 // parmBlk-64 ARG2(dst) ptr spill slot
1889 // parmBlk-56 ARG3(crypto key) ptr spill slot
1890 // parmBlk-48 ARG4(icv value) ptr spill slot
1891 //
1892 // parmBlk-72
1893 // parmBlk-80
1894 // parmBlk-88 counter vector current position
1895 // parmBlk-96 reduced msg len (after preLoop processing)
1896 //
1897 // parmBlk-104 Z_R13 spill slot (preLoop only)
1898 // parmBlk-112 Z_R12 spill slot (preLoop only)
1899 // parmBlk-120 Z_R11 spill slot (preLoop only)
1900 // parmBlk-128 Z_R10 spill slot (preLoop only)
1901 //
1902 //
1903 // Layout of the parameter block (instruction KMCTR, function KMCTR-AES*
1904 //
1905 // +--------+ key_len: +16 (AES-128), +24 (AES-192), +32 (AES-256)
1906 // | |
1907 // | | cryptographic key
1908 // | |
1909 // +--------+ <-- parmBlk
1910 //
1911 // On exit:
1912 // Z_SP points to resized frame
1913 // Z_SP before resize available from -16(parmBlk)
1914 // parmBlk points to crypto instruction parameter block
1915 // parameter block is filled with crypto key.
1916 // msglen unchanged, saved for later at -24(parmBlk)
1917 // fCode contains function code for instruction
1918 // key unchanged
1919 //
1920 void generate_counterMode_prepare_Stack(Register parmBlk, Register ctr, Register counter, Register scratch) {
1921
1922 BLOCK_COMMENT("prepare stack counterMode_AESCrypt {");
1923
1924 // save argument registers.
1925 // ARG1(from) is Z_RET as well. Not saved or restored.
1926 // ARG5(msglen) is restored by other means.
1927 __ z_stmg(Z_ARG2, Z_ARG4, argsave_offset, parmBlk);
1928
1929 assert(AES_ctrVec_len > 0, "sanity. We need a counter vector");
1930 __ add2reg(counter, AES_parmBlk_align, parmBlk); // counter array is located behind crypto key. Available range is disp12 only.
1931 __ z_mvc(0, AES_ctrVal_len-1, counter, 0, ctr); // move first copy of iv
1932 for (int j = 1; j < AES_ctrVec_len; j+=j) { // j (and amount of moved data) doubles with every iteration
1933 int offset = j * AES_ctrVal_len;
1934 if (offset <= 256) {
1935 __ z_mvc(offset, offset-1, counter, 0, counter); // move iv
1936 } else {
1937 for (int k = 0; k < offset; k += 256) {
1938 __ z_mvc(offset+k, 255, counter, 0, counter);
1939 }
1940 }
1941 }
1942
1943 Label noCarry, done;
1944 __ z_lg(scratch, Address(ctr, 8)); // get low-order DW of initial counter.
1945 __ z_algfi(scratch, AES_ctrVec_len); // check if we will overflow during init.
1946 __ z_brc(Assembler::bcondLogNoCarry, noCarry); // No, 64-bit increment is sufficient.
1947
1948 for (int j = 1; j < AES_ctrVec_len; j++) { // start with j = 1; no need to add 0 to the first counter value.
1949 int offset = j * AES_ctrVal_len;
1950 generate_increment128(counter, offset, j, scratch); // increment iv by index value
1951 }
1952 __ z_bru(done);
1953
1954 __ bind(noCarry);
1955 for (int j = 1; j < AES_ctrVec_len; j++) { // start with j = 1; no need to add 0 to the first counter value.
1956 int offset = j * AES_ctrVal_len;
1957 generate_increment64(counter, offset, j); // increment iv by index value
1958 }
1959
1960 __ bind(done);
1961
1962 BLOCK_COMMENT("} prepare stack counterMode_AESCrypt");
1963 }
1964
1965
1966 void generate_counterMode_increment_ctrVector(Register parmBlk, Register counter, Register scratch, bool v0_only) {
1967
1968 BLOCK_COMMENT("increment ctrVector counterMode_AESCrypt {");
1969
1970 __ add2reg(counter, AES_parmBlk_align, parmBlk); // ptr to counter array needs to be restored
1971
1972 if (v0_only) {
1973 int offset = 0;
1974 generate_increment128(counter, offset, AES_ctrVec_len, scratch); // increment iv by # vector elements
1975 } else {
1976 int j = 0;
1977 if (VM_Version::has_VectorFacility()) {
1978 bool first_call = true;
1979 for (; j < (AES_ctrVec_len - 3); j+=4) { // increment blocks of 4 iv elements
1980 int offset = j * AES_ctrVal_len;
1981 generate_increment128x4(counter, offset, AES_ctrVec_len, first_call);
1982 first_call = false;
1983 }
1984 }
1985 for (; j < AES_ctrVec_len; j++) {
1986 int offset = j * AES_ctrVal_len;
1987 generate_increment128(counter, offset, AES_ctrVec_len, scratch); // increment iv by # vector elements
1988 }
1989 }
1990
1991 BLOCK_COMMENT("} increment ctrVector counterMode_AESCrypt");
1992 }
1993
1994 // IBM s390 (IBM z/Architecture, to be more exact) uses Big-Endian number representation.
1995 // Therefore, the bits are ordered from most significant to least significant. The address
1996 // of a number in memory points to its lowest location where the most significant bit is stored.
1997 void generate_increment64(Register counter, int offset, int increment) {
1998 __ z_algsi(offset + 8, counter, increment); // increment, no overflow check
1999 }
2000
2001 void generate_increment128(Register counter, int offset, int increment, Register scratch) {
2002 __ clear_reg(scratch); // prepare to add carry to high-order DW
2003 __ z_algsi(offset + 8, counter, increment); // increment low order DW
2004 __ z_alcg(scratch, Address(counter, offset)); // add carry to high-order DW
2005 __ z_stg(scratch, Address(counter, offset)); // store back
2006 }
2007
2008 void generate_increment128(Register counter, int offset, Register increment, Register scratch) {
2009 __ clear_reg(scratch); // prepare to add carry to high-order DW
2010 __ z_alg(increment, Address(counter, offset + 8)); // increment low order DW
2011 __ z_stg(increment, Address(counter, offset + 8)); // store back
2012 __ z_alcg(scratch, Address(counter, offset)); // add carry to high-order DW
2013 __ z_stg(scratch, Address(counter, offset)); // store back
2014 }
2015
2016 // This is the vector variant of increment128, incrementing 4 ctr vector elements per call.
2017 void generate_increment128x4(Register counter, int offset, int increment, bool init) {
2018 VectorRegister Vincr = Z_V16;
2019 VectorRegister Vctr0 = Z_V20;
2020 VectorRegister Vctr1 = Z_V21;
2021 VectorRegister Vctr2 = Z_V22;
2022 VectorRegister Vctr3 = Z_V23;
2023
2024 // Initialize the increment value only once for a series of increments.
2025 // It must be assured that the non-initializing generator calls are
2026 // immediately subsequent. Otherwise, there is no guarantee for Vincr to be unchanged.
2027 if (init) {
2028 __ z_vzero(Vincr); // preset VReg with constant increment
2029 __ z_vleih(Vincr, increment, 7); // rightmost HW has ix = 7
2030 }
2031
2032 __ z_vlm(Vctr0, Vctr3, offset, counter); // get the counter values
2033 __ z_vaq(Vctr0, Vctr0, Vincr); // increment them
2034 __ z_vaq(Vctr1, Vctr1, Vincr);
2035 __ z_vaq(Vctr2, Vctr2, Vincr);
2036 __ z_vaq(Vctr3, Vctr3, Vincr);
2037 __ z_vstm(Vctr0, Vctr3, offset, counter); // store the counter values
2038 }
2039
2040 unsigned int generate_counterMode_push_Block(int dataBlk_len, int parmBlk_len, int crypto_fCode,
2041 Register parmBlk, Register msglen, Register fCode, Register key) {
2042
2043 // space for data blocks (src and dst, one each) for partial block processing)
2044 AES_parmBlk_addspace = AES_stackSpace_incr // spill space (temp data)
2045 + AES_stackSpace_incr // for argument save/restore
2046 + AES_stackSpace_incr*2 // for work reg save/restore
2047 ;
2048 AES_dataBlk_space = roundup(2*dataBlk_len, AES_parmBlk_align);
2049 AES_dataBlk_offset = -(AES_parmBlk_addspace+AES_dataBlk_space);
2050 const int key_len = parmBlk_len; // The length of the unextended key (16, 24, 32)
2051
2052 assert((AES_ctrVal_len == 0) || (AES_ctrVal_len == dataBlk_len), "varying dataBlk_len is not supported.");
2053 AES_ctrVal_len = dataBlk_len; // ctr init value len (in bytes)
2054 AES_ctrArea_len = AES_ctrVec_len * AES_ctrVal_len; // space required on stack for ctr vector
2055
2056 // This len must be known at JIT compile time. Only then are we able to recalc the SP before resize.
2057 // We buy this knowledge by wasting some (up to AES_parmBlk_align) bytes of stack space.
2058 const int resize_len = AES_parmBlk_align // room for alignment of parmBlk
2059 + AES_parmBlk_align // extra room for alignment
2060 + AES_dataBlk_space // one src and one dst data blk
2061 + AES_parmBlk_addspace // spill space for local data
2062 + roundup(parmBlk_len, AES_parmBlk_align) // aligned length of parmBlk
2063 + AES_ctrArea_len // stack space for ctr vector
2064 ;
2065 Register scratch = fCode; // We can use fCode as a scratch register. It's contents on entry
2066 // is irrelevant and it is set at the very end of this code block.
2067
2068 assert(key_len < 256, "excessive crypto key len: %d, limit: 256", key_len);
2069
2070 BLOCK_COMMENT(err_msg("push_Block (%d bytes) counterMode_AESCrypt%d {", resize_len, parmBlk_len*8));
2071
2072 // After the frame is resized, the parmBlk is positioned such
2073 // that it is octoword-aligned. This potentially creates some
2074 // alignment waste in addspace and/or in the gap area.
2075 // After resize_frame, scratch contains the frame pointer.
2076 __ resize_frame(-resize_len, scratch, true);
2077 #ifdef ASSERT
2078 __ clear_mem(Address(Z_SP, (intptr_t)8), resize_len - 8);
2079 #endif
2080
2081 // calculate aligned parmBlk address from updated (resized) SP.
2082 __ add2reg(parmBlk, AES_parmBlk_addspace + AES_dataBlk_space + (2*AES_parmBlk_align-1), Z_SP);
2083 __ z_nill(parmBlk, (~(AES_parmBlk_align-1)) & 0xffff); // Align parameter block.
2084
2085 // There is room to spill stuff in the range [parmBlk-AES_parmBlk_addspace+8, parmBlk).
2086 __ z_mviy(keylen_offset, parmBlk, key_len - 1); // Spill crypto key length for later use. Decrement by one for direct use with xc template.
2087 __ z_mviy(fCode_offset, parmBlk, crypto_fCode); // Crypto function code, will be loaded into Z_R0 later.
2088 __ z_sty(msglen, msglen_offset, parmBlk); // full plaintext/ciphertext len.
2089 __ z_sty(msglen, msglen_red_offset, parmBlk); // save for main loop, may get updated in preLoop.
2090 __ z_sra(msglen, exact_log2(dataBlk_len)); // # full cipher blocks that can be formed from input text.
2091 __ z_sty(msglen, rem_msgblk_offset, parmBlk);
2092
2093 __ add2reg(scratch, resize_len, Z_SP); // calculate (SP before resize) from resized SP.
2094 __ z_stg(scratch, unextSP_offset, parmBlk); // Spill unextended SP for easy revert.
2095 __ z_stmg(Z_R10, Z_R13, regsave_offset, parmBlk); // make some regs available as work registers
2096
2097 // Fill parmBlk with all required data
2098 __ z_mvc(0, key_len-1, parmBlk, 0, key); // Copy key. Need to do it here - key_len is only known here.
2099 BLOCK_COMMENT(err_msg("} push_Block (%d bytes) counterMode_AESCrypt%d", resize_len, parmBlk_len*8));
2100 return resize_len;
2101 }
2102
2103
2104 void generate_counterMode_pop_Block(Register parmBlk, Register msglen, Label& eraser) {
2105 // For added safety, clear the stack area where the crypto key was stored.
2106 Register scratch = msglen;
2107 assert_different_registers(scratch, Z_R0); // can't use Z_R0 for exrl.
2108
2109 // wipe out key on stack
2110 __ z_llgc(scratch, keylen_offset, parmBlk); // get saved (key_len-1) value (we saved just one byte!)
2111 __ z_exrl(scratch, eraser); // template relies on parmBlk still pointing to key on stack
2112
2113 // restore argument registers.
2114 // ARG1(from) is Z_RET as well. Not restored - will hold return value anyway.
2115 // ARG5(msglen) is restored further down.
2116 __ z_lmg(Z_ARG2, Z_ARG4, argsave_offset, parmBlk);
2117
2118 // restore work registers
2119 __ z_lmg(Z_R10, Z_R13, regsave_offset, parmBlk); // make some regs available as work registers
2120
2121 __ z_lgf(msglen, msglen_offset, parmBlk); // Restore msglen, only low order FW is valid
2122 #ifdef ASSERT
2123 {
2124 Label skip2last, skip2done;
2125 // Z_RET (aka Z_R2) can be used as scratch as well. It will be set from msglen before return.
2126 __ z_lgr(Z_RET, Z_SP); // save extended SP
2127 __ z_lg(Z_SP, unextSP_offset, parmBlk); // trim stack back to unextended size
2128 __ z_sgrk(Z_R1, Z_SP, Z_RET);
2129
2130 __ z_cghi(Z_R1, 256);
2131 __ z_brl(skip2last);
2132 __ z_xc(0, 255, Z_RET, 0, Z_RET);
2133 __ z_aghi(Z_RET, 256);
2134 __ z_aghi(Z_R1, -256);
2135
2136 __ z_cghi(Z_R1, 256);
2137 __ z_brl(skip2last);
2138 __ z_xc(0, 255, Z_RET, 0, Z_RET);
2139 __ z_aghi(Z_RET, 256);
2140 __ z_aghi(Z_R1, -256);
2141
2142 __ z_cghi(Z_R1, 256);
2143 __ z_brl(skip2last);
2144 __ z_xc(0, 255, Z_RET, 0, Z_RET);
2145 __ z_aghi(Z_RET, 256);
2146 __ z_aghi(Z_R1, -256);
2147
2148 __ bind(skip2last);
2149 __ z_lgr(Z_R0, Z_RET);
2150 __ z_aghik(Z_RET, Z_R1, -1); // decrement for exrl
2151 __ z_brl(skip2done);
2152 __ z_lgr(parmBlk, Z_R0); // parmBlk == Z_R1, used in eraser template
2153 __ z_exrl(Z_RET, eraser);
2154
2155 __ bind(skip2done);
2156 }
2157 #else
2158 __ z_lg(Z_SP, unextSP_offset, parmBlk); // trim stack back to unextended size
2159 #endif
2160 }
2161
2162
2163 int generate_counterMode_push_parmBlk(Register parmBlk, Register msglen, Register fCode, Register key, bool is_decipher) {
2164 int resize_len = 0;
2165 int mode = is_decipher ? VM_Version::CipherMode::decipher : VM_Version::CipherMode::cipher;
2166 Label parmBlk_128, parmBlk_192, parmBlk_256, parmBlk_set;
2167 Register keylen = fCode; // Expanded key length, as read from key array, Temp only.
2168 // use fCode as scratch; fCode receives its final value later.
2169
2170 // Read key len of expanded key (in 4-byte words).
2171 __ z_lgf(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2172 __ z_cghi(keylen, 52);
2173 if (VM_Version::has_Crypto_AES_CTR256()) { __ z_brh(parmBlk_256); } // keyLen > 52: AES256. Assume: most frequent
2174 if (VM_Version::has_Crypto_AES_CTR128()) { __ z_brl(parmBlk_128); } // keyLen < 52: AES128.
2175 if (VM_Version::has_Crypto_AES_CTR192()) { __ z_bre(parmBlk_192); } // keyLen == 52: AES192. Assume: least frequent
2176
2177 // Safety net: requested AES_CTR function for requested keylen not available on this CPU.
2178 __ stop_static("AES key strength not supported by CPU. Use -XX:-UseAESCTRIntrinsics as remedy.", 0);
2179
2180 if (VM_Version::has_Crypto_AES_CTR128()) {
2181 __ bind(parmBlk_128);
2182 resize_len = generate_counterMode_push_Block(VM_Version::Cipher::_AES128_dataBlk,
2183 VM_Version::Cipher::_AES128_parmBlk_G,
2184 VM_Version::Cipher::_AES128 + mode,
2185 parmBlk, msglen, fCode, key);
2186 if (VM_Version::has_Crypto_AES_CTR256() || VM_Version::has_Crypto_AES_CTR192()) {
2187 __ z_bru(parmBlk_set); // Fallthru otherwise.
2188 }
2189 }
2190
2191 if (VM_Version::has_Crypto_AES_CTR192()) {
2192 __ bind(parmBlk_192);
2193 resize_len = generate_counterMode_push_Block(VM_Version::Cipher::_AES192_dataBlk,
2194 VM_Version::Cipher::_AES192_parmBlk_G,
2195 VM_Version::Cipher::_AES192 + mode,
2196 parmBlk, msglen, fCode, key);
2197 if (VM_Version::has_Crypto_AES_CTR256()) {
2198 __ z_bru(parmBlk_set); // Fallthru otherwise.
2199 }
2200 }
2201
2202 if (VM_Version::has_Crypto_AES_CTR256()) {
2203 __ bind(parmBlk_256);
2204 resize_len = generate_counterMode_push_Block(VM_Version::Cipher::_AES256_dataBlk,
2205 VM_Version::Cipher::_AES256_parmBlk_G,
2206 VM_Version::Cipher::_AES256 + mode,
2207 parmBlk, msglen, fCode, key);
2208 // Fallthru
2209 }
2210
2211 __ bind(parmBlk_set);
2212 return resize_len;
2213 }
2214
2215
2216 void generate_counterMode_pop_parmBlk(Register parmBlk, Register msglen, Label& eraser) {
2217
2218 BLOCK_COMMENT("pop parmBlk counterMode_AESCrypt {");
2219
2220 generate_counterMode_pop_Block(parmBlk, msglen, eraser);
2221
2222 BLOCK_COMMENT("} pop parmBlk counterMode_AESCrypt");
2223 }
2224
2225 // Implementation of counter-mode AES encrypt/decrypt function.
2226 //
2227 void generate_counterMode_AES_impl(bool is_decipher) {
2228
2229 // On entry:
2230 // if there was a previous call to update(), and this previous call did not fully use
2231 // the current encrypted counter, that counter is available at arg6_Offset(Z_SP).
2232 // The index of the first unused bayte in the encrypted counter is available at arg7_Offset(Z_SP).
2233 // The index is in the range [1..AES_ctrVal_len] ([1..16]), where index == 16 indicates a fully
2234 // used previous encrypted counter.
2235 // The unencrypted counter has already been incremented and is ready to be used for the next
2236 // data block, after the unused bytes from the previous call have been consumed.
2237 // The unencrypted counter follows the "increment-after use" principle.
2238
2239 // On exit:
2240 // The index of the first unused byte of the encrypted counter is written back to arg7_Offset(Z_SP).
2241 // A value of AES_ctrVal_len (16) indicates there is no leftover byte.
2242 // If there is at least one leftover byte (1 <= index < AES_ctrVal_len), the encrypted counter value
2243 // is written back to arg6_Offset(Z_SP). If there is no leftover, nothing is written back.
2244 // The unencrypted counter value is written back after having been incremented.
2245
2246 Register from = Z_ARG1; // byte[], source byte array (clear text)
2247 Register to = Z_ARG2; // byte[], destination byte array (ciphered)
2248 Register key = Z_ARG3; // byte[], expanded key array.
2249 Register ctr = Z_ARG4; // byte[], counter byte array.
2250 const Register msglen = Z_ARG5; // int, Total length of the msg to be encrypted. Value must be
2251 // returned in Z_RET upon completion of this stub.
2252 // This is a jint. Negative values are illegal, but technically possible.
2253 // Do not rely on high word. Contents is undefined.
2254 // encCtr = Z_ARG6 - encrypted counter (byte array),
2255 // address passed on stack at _z_abi(remaining_cargs) + 0 * WordSize
2256 // cvIndex = Z_ARG7 - # used (consumed) bytes of encrypted counter,
2257 // passed on stack at _z_abi(remaining_cargs) + 1 * WordSize
2258 // Caution:4-byte value, right-justified in 8-byte stack word
2259
2260 const Register fCode = Z_R0; // crypto function code
2261 const Register parmBlk = Z_R1; // parameter block address (points to crypto key)
2262 const Register src = Z_ARG1; // is Z_R2, forms even/odd pair with srclen
2263 const Register srclen = Z_ARG2; // Overwrites destination address.
2264 const Register dst = Z_ARG3; // Overwrites key address.
2265 const Register counter = Z_ARG5; // Overwrites msglen. Must have counter array in an even register.
2266
2267 Label srcMover, dstMover, fromMover, ctrXOR, dataEraser; // EXRL (execution) templates.
2268 Label CryptoLoop, CryptoLoop_doit, CryptoLoop_end, CryptoLoop_setupAndDoLast, CryptoLoop_ctrVal_inc;
2269 Label allDone, allDone_noInc, popAndExit, Exit;
2270
2271 int arg6_Offset = _z_abi(remaining_cargs) + 0 * HeapWordSize;
2272 int arg7_Offset = _z_abi(remaining_cargs) + 1 * HeapWordSize; // stack slot holds ptr to int value
2273 int oldSP_Offset = 0;
2274
2275 // Is there anything to do at all? Protect against negative len as well.
2276 __ z_ltr(msglen, msglen);
2277 __ z_brnh(Exit);
2278
2279 // Expand stack, load parm block address into parmBlk (== Z_R1), copy crypto key to parm block.
2280 oldSP_Offset = generate_counterMode_push_parmBlk(parmBlk, msglen, fCode, key, is_decipher);
2281 arg6_Offset += oldSP_Offset;
2282 arg7_Offset += oldSP_Offset;
2283
2284 // Check if there is a leftover, partially used encrypted counter from last invocation.
2285 // If so, use those leftover counter bytes first before starting the "normal" encryption.
2286
2287 // We do not have access to the encrypted counter value. It is generated and used only
2288 // internally within the previous kmctr instruction. But, at the end of call to this stub,
2289 // the last encrypted couner is extracted by ciphering a 0x00 byte stream. The result is
2290 // stored at the arg6 location for use with the subsequent call.
2291 //
2292 // The #used bytes of the encrypted counter (from a previous call) is provided via arg7.
2293 // It is used as index into the encrypted counter to access the first byte availabla for ciphering.
2294 // To cipher the input text, we move the number of remaining bytes in the encrypted counter from
2295 // input to output. Then we simply XOR the output bytes with the associated encrypted counter bytes.
2296
2297 Register cvIxAddr = Z_R10; // Address of index into encCtr. Preserved for use @CryptoLoop_end.
2298 __ z_lg(cvIxAddr, arg7_Offset, Z_SP); // arg7: addr of field encCTR_index.
2299
2300 {
2301 Register cvUnused = Z_R11; // # unused bytes of encrypted counter value (= 16 - cvIndex)
2302 Register encCtr = Z_R12; // encrypted counter value, points to first ununsed byte.
2303 Register cvIndex = Z_R13; // # index of first unused byte of encrypted counter value
2304 Label preLoop_end;
2305
2306 // preLoop is necessary only if there is a partially used encrypted counter (encCtr).
2307 // Partially used means cvIndex is in [1, dataBlk_len-1].
2308 // cvIndex == 0: encCtr is set up but not used at all. Should not occur.
2309 // cvIndex == dataBlk_len: encCtr is exhausted, all bytes used.
2310 // Using unsigned compare protects against cases where (cvIndex < 0).
2311 __ z_clfhsi(0, cvIxAddr, AES_ctrVal_len); // check #used bytes in encCtr against ctr len.
2312 __ z_brnl(preLoop_end); // if encCtr is fully used, skip to normal processing.
2313 __ z_ltgf(cvIndex, 0, Z_R0, cvIxAddr); // # used bytes in encCTR.
2314 __ z_brz(preLoop_end); // if encCtr has no used bytes, skip to normal processing.
2315
2316 __ z_lg(encCtr, arg6_Offset, Z_SP); // encrypted counter from last call to update()
2317 __ z_agr(encCtr, cvIndex); // now points to first unused byte
2318
2319 __ add2reg(cvUnused, -AES_ctrVal_len, cvIndex); // calculate #unused bytes in encCtr.
2320 __ z_lcgr(cvUnused, cvUnused); // previous checks ensure cvUnused in range [1, dataBlk_len-1]
2321
2322 __ z_lgf(msglen, msglen_offset, parmBlk); // Restore msglen (jint value)
2323 __ z_cr(cvUnused, msglen); // check if msg can consume all unused encCtr bytes
2324 __ z_locr(cvUnused, msglen, Assembler::bcondHigh); // take the shorter length
2325 __ z_aghi(cvUnused, -1); // decrement # unused bytes by 1 for exrl instruction
2326 // preceding checks ensure cvUnused in range [1, dataBlk_len-1]
2327 __ z_exrl(cvUnused, fromMover);
2328 __ z_exrl(cvUnused, ctrXOR);
2329
2330 __ z_aghi(cvUnused, 1); // revert decrement from above
2331 __ z_agr(cvIndex, cvUnused); // update index into encCtr (first unused byte)
2332 __ z_st(cvIndex, 0, cvIxAddr); // write back arg7, cvIxAddr is still valid
2333
2334 // update pointers and counters to prepare for main loop
2335 __ z_agr(from, cvUnused);
2336 __ z_agr(to, cvUnused);
2337 __ z_sr(msglen, cvUnused); // #bytes not yet processed
2338 __ z_sty(msglen, msglen_red_offset, parmBlk); // save for calculations in main loop
2339 __ z_srak(Z_R0, msglen, exact_log2(AES_ctrVal_len));// # full cipher blocks that can be formed from input text.
2340 __ z_sty(Z_R0, rem_msgblk_offset, parmBlk);
2341
2342 // check remaining msglen. If zero, all msg bytes were processed in preLoop.
2343 __ z_ltr(msglen, msglen);
2344 __ z_brnh(popAndExit);
2345
2346 __ bind(preLoop_end);
2347 }
2348
2349 // Create count vector on stack to accommodate up to AES_ctrVec_len blocks.
2350 generate_counterMode_prepare_Stack(parmBlk, ctr, counter, fCode);
2351
2352 // Prepare other registers for instruction.
2353 __ lgr_if_needed(src, from); // Copy src address. Will not emit, src/from are identical.
2354 __ z_lgr(dst, to);
2355 __ z_llgc(fCode, fCode_offset, Z_R0, parmBlk);
2356
2357 __ bind(CryptoLoop);
2358 __ z_lghi(srclen, AES_ctrArea_len); // preset len (#bytes) for next iteration: max possible.
2359 __ z_asi(rem_msgblk_offset, parmBlk, -AES_ctrVec_len); // decrement #remaining blocks (16 bytes each). Range: [+127..-128]
2360 __ z_brl(CryptoLoop_setupAndDoLast); // Handling the last iteration (using less than max #blocks) out-of-line
2361
2362 __ bind(CryptoLoop_doit);
2363 __ kmctr(dst, counter, src); // Cipher the message.
2364
2365 __ z_lt(srclen, rem_msgblk_offset, Z_R0, parmBlk); // check if this was the last iteration
2366 __ z_brz(CryptoLoop_ctrVal_inc); // == 0: ctrVector fully used. Need to increment the first
2367 // vector element to encrypt remaining unprocessed bytes.
2368 // __ z_brl(CryptoLoop_end); // < 0: this was detected before and handled at CryptoLoop_setupAndDoLast
2369 // > 0: this is the fallthru case, need another iteration
2370
2371 generate_counterMode_increment_ctrVector(parmBlk, counter, srclen, false); // srclen unused here (serves as scratch)
2372 __ z_bru(CryptoLoop);
2373
2374 __ bind(CryptoLoop_end);
2375
2376 // OK, when we arrive here, we have encrypted all of the "from" byte stream
2377 // except for the last few [0..dataBlk_len) bytes. In addition, we know that
2378 // there are no more unused bytes in the previously generated encrypted counter.
2379 // The (unencrypted) counter, however, is ready to use (it was incremented before).
2380
2381 // To encrypt the few remaining bytes, we need to form an extra src and dst
2382 // data block of dataBlk_len each. This is because we can only process full
2383 // blocks but we must not read or write beyond the boundaries of the argument
2384 // arrays. Here is what we do:
2385 // - The ctrVector has at least one unused element. This is ensured by CryptoLoop code.
2386 // - The (first) unused element is pointed at by the counter register.
2387 // - The src data block is filled with the remaining "from" bytes, remainder of block undefined.
2388 // - The single src data block is encrypted into the dst data block.
2389 // - The dst data block is copied into the "to" array, but only the leftmost few bytes
2390 // (as many as were left in the source byte stream).
2391 // - The counter value to be used is pointed at by the counter register.
2392 // - Fortunately, the crypto instruction (kmctr) has updated all related addresses such that
2393 // we know where to continue with "from" and "to" and which counter value to use next.
2394
2395 Register encCtr = Z_R12; // encrypted counter value, points to stub argument.
2396 Register tmpDst = Z_R12; // addr of temp destination (for last partial block encryption)
2397
2398 __ z_lgf(srclen, msglen_red_offset, parmBlk); // plaintext/ciphertext len after potential preLoop processing.
2399 __ z_nilf(srclen, AES_ctrVal_len - 1); // those rightmost bits indicate the unprocessed #bytes
2400 __ z_stg(srclen, localSpill_offset, parmBlk); // save for later reuse
2401 __ z_mvhi(0, cvIxAddr, 16); // write back arg7 (default 16 in case of allDone).
2402 __ z_braz(allDone_noInc); // no unprocessed bytes? Then we are done.
2403 // This also means the last block of data processed was
2404 // a full-sized block (AES_ctrVal_len bytes) which results
2405 // in no leftover encrypted counter bytes.
2406 __ z_st(srclen, 0, cvIxAddr); // This will be the index of the first unused byte in the encrypted counter.
2407 __ z_stg(counter, counter_offset, parmBlk); // save counter location for easy later restore
2408
2409 // calculate address (on stack) for final dst and src blocks.
2410 __ add2reg(tmpDst, AES_dataBlk_offset, parmBlk); // tmp dst (on stack) is right before tmp src
2411
2412 // We have a residue of [1..15] unprocessed bytes, srclen holds the exact number.
2413 // Residue == 0 was checked just above, residue == AES_ctrVal_len would be another
2414 // full-sized block and would have been handled by CryptoLoop.
2415
2416 __ add2reg(srclen, -1); // decrement for exrl
2417 __ z_exrl(srclen, srcMover); // copy remaining bytes of src byte stream
2418 __ load_const_optimized(srclen, AES_ctrVal_len); // kmctr processes only complete blocks
2419 __ add2reg(src, AES_ctrVal_len, tmpDst); // tmp dst is right before tmp src
2420
2421 __ kmctr(tmpDst, counter, src); // Cipher the remaining bytes.
2422
2423 __ add2reg(tmpDst, -AES_ctrVal_len, tmpDst); // restore tmp dst address
2424 __ z_lg(srclen, localSpill_offset, parmBlk); // residual len, saved above
2425 __ add2reg(srclen, -1); // decrement for exrl
2426 __ z_exrl(srclen, dstMover);
2427
2428 // Write back new encrypted counter
2429 __ add2reg(src, AES_dataBlk_offset, parmBlk);
2430 __ clear_mem(Address(src, RegisterOrConstant((intptr_t)0)), AES_ctrVal_len);
2431 __ load_const_optimized(srclen, AES_ctrVal_len); // kmctr processes only complete blocks
2432 __ z_lg(encCtr, arg6_Offset, Z_SP); // write encrypted counter to arg6
2433 __ z_lg(counter, counter_offset, parmBlk); // restore counter
2434 __ kmctr(encCtr, counter, src);
2435
2436 // The last used element of the counter vector contains the latest counter value that was used.
2437 // As described above, the counter value on exit must be the one to be used next.
2438 __ bind(allDone);
2439 __ z_lg(counter, counter_offset, parmBlk); // restore counter
2440 generate_increment128(counter, 0, 1, Z_R0);
2441
2442 __ bind(allDone_noInc);
2443 __ z_mvc(0, AES_ctrVal_len, ctr, 0, counter);
2444
2445 __ bind(popAndExit);
2446 generate_counterMode_pop_parmBlk(parmBlk, msglen, dataEraser);
2447
2448 __ bind(Exit);
2449 __ z_lgfr(Z_RET, msglen);
2450
2451 __ z_br(Z_R14);
2452
2453 //----------------------------
2454 //---< out-of-line code >---
2455 //----------------------------
2456 __ bind(CryptoLoop_setupAndDoLast);
2457 __ z_lgf(srclen, rem_msgblk_offset, parmBlk); // remaining #blocks in memory is < 0
2458 __ z_aghi(srclen, AES_ctrVec_len); // recalculate the actually remaining #blocks
2459 __ z_sllg(srclen, srclen, exact_log2(AES_ctrVal_len)); // convert to #bytes. Counter value is same length as data block
2460 __ kmctr(dst, counter, src); // Cipher the last integral blocks of the message.
2461 __ z_bru(CryptoLoop_end); // There is at least one unused counter vector element.
2462 // no need to increment.
2463
2464 __ bind(CryptoLoop_ctrVal_inc);
2465 generate_counterMode_increment_ctrVector(parmBlk, counter, srclen, true); // srclen unused here (serves as scratch)
2466 __ z_bru(CryptoLoop_end);
2467
2468 //-------------------------------------------
2469 //---< execution templates for preLoop >---
2470 //-------------------------------------------
2471 __ bind(fromMover);
2472 __ z_mvc(0, 0, to, 0, from); // Template instruction to move input data to dst.
2473 __ bind(ctrXOR);
2474 __ z_xc(0, 0, to, 0, encCtr); // Template instruction to XOR input data (now in to) with encrypted counter.
2475
2476 //-------------------------------
2477 //---< execution templates >---
2478 //-------------------------------
2479 __ bind(dataEraser);
2480 __ z_xc(0, 0, parmBlk, 0, parmBlk); // Template instruction to erase crypto key on stack.
2481 __ bind(dstMover);
2482 __ z_mvc(0, 0, dst, 0, tmpDst); // Template instruction to move encrypted reminder from stack to dst.
2483 __ bind(srcMover);
2484 __ z_mvc(AES_ctrVal_len, 0, tmpDst, 0, src); // Template instruction to move reminder of source byte stream to stack.
2485 }
2486
2487
2488 // Create two intrinsic variants, optimized for short and long plaintexts.
2489 void generate_counterMode_AES(bool is_decipher) {
2490
2491 const Register msglen = Z_ARG5; // int, Total length of the msg to be encrypted. Value must be
2492 // returned in Z_RET upon completion of this stub.
2493 const int threshold = 256; // above this length (in bytes), text is considered long.
2494 const int vec_short = threshold>>6; // that many blocks (16 bytes each) per iteration, max 4 loop iterations
2495 const int vec_long = threshold>>2; // that many blocks (16 bytes each) per iteration.
2496
2497 Label AESCTR_short, AESCTR_long;
2498
2499 __ z_chi(msglen, threshold);
2500 __ z_brh(AESCTR_long);
2501
2502 __ bind(AESCTR_short);
2503
2504 BLOCK_COMMENT(err_msg("counterMode_AESCrypt (text len <= %d, block size = %d) {", threshold, vec_short*16));
2505
2506 AES_ctrVec_len = vec_short;
2507 generate_counterMode_AES_impl(false); // control of generated code will not return
2508
2509 BLOCK_COMMENT(err_msg("} counterMode_AESCrypt (text len <= %d, block size = %d)", threshold, vec_short*16));
2510
2511 __ align(32); // Octoword alignment benefits branch targets.
2512
2513 BLOCK_COMMENT(err_msg("counterMode_AESCrypt (text len > %d, block size = %d) {", threshold, vec_long*16));
2514
2515 __ bind(AESCTR_long);
2516 AES_ctrVec_len = vec_long;
2517 generate_counterMode_AES_impl(false); // control of generated code will not return
2518
2519 BLOCK_COMMENT(err_msg("} counterMode_AESCrypt (text len > %d, block size = %d)", threshold, vec_long*16));
2520 }
2521
2522
2523 // Compute AES-CTR crypto function.
2524 // Encrypt or decrypt is selected via parameters. Only one stub is necessary.
2525 address generate_counterMode_AESCrypt(const char* name) {
2526 __ align(CodeEntryAlignment);
2527 StubCodeMark mark(this, "StubRoutines", name);
2528 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2529
2530 generate_counterMode_AES(false);
2531
2532 return __ addr_at(start_off);
2533 }
2534
2535 // *****************************************************************************
2536
2537 // Compute GHASH function.
2538 address generate_ghash_processBlocks() {
2539 __ align(CodeEntryAlignment);
2540 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
2541 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2542
2543 const Register state = Z_ARG1;
2544 const Register subkeyH = Z_ARG2;
2545 const Register data = Z_ARG3; // 1st of even-odd register pair.
2546 const Register blocks = Z_ARG4;
2547 const Register len = blocks; // 2nd of even-odd register pair.
2548
2549 const int param_block_size = 4 * 8;
2550 const int frame_resize = param_block_size + 8; // Extra space for copy of fp.
2551
2552 // Reserve stack space for parameter block (R1).
2553 __ z_lgr(Z_R1, Z_SP);
2554 __ resize_frame(-frame_resize, Z_R0, true);
2555 __ z_aghi(Z_R1, -param_block_size);
2556
2557 // Fill parameter block.
2558 __ z_mvc(Address(Z_R1) , Address(state) , 16);
2559 __ z_mvc(Address(Z_R1, 16), Address(subkeyH), 16);
2560
2561 // R4+5: data pointer + length
2562 __ z_llgfr(len, blocks); // Cast to 64-bit.
2563
2564 // R0: function code
2565 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_GHASH);
2566
2567 // Compute.
2568 __ z_sllg(len, len, 4); // In bytes.
2569 __ kimd(data);
2570
2571 // Copy back result and free parameter block.
2572 __ z_mvc(Address(state), Address(Z_R1), 16);
2573 __ z_xc(Address(Z_R1), param_block_size, Address(Z_R1));
2574 __ z_aghi(Z_SP, frame_resize);
2575
2576 __ z_br(Z_R14);
2577
2578 return __ addr_at(start_off);
2579 }
2580
2581
2582 // Call interface for all SHA* stubs.
2583 //
2584 // Z_ARG1 - source data block. Ptr to leftmost byte to be processed.
2585 // Z_ARG2 - current SHA state. Ptr to state area. This area serves as
2586 // parameter block as required by the crypto instruction.
2587 // Z_ARG3 - current byte offset in source data block.
2588 // Z_ARG4 - last byte offset in source data block.
2589 // (Z_ARG4 - Z_ARG3) gives the #bytes remaining to be processed.
2590 //
2591 // Z_RET - return value. First unprocessed byte offset in src buffer.
2592 //
2593 // A few notes on the call interface:
2594 // - All stubs, whether they are single-block or multi-block, are assumed to
2595 // digest an integer multiple of the data block length of data. All data
2596 // blocks are digested using the intermediate message digest (KIMD) instruction.
2597 // Special end processing, as done by the KLMD instruction, seems to be
2598 // emulated by the calling code.
2599 //
2600 // - Z_ARG1 addresses the first byte of source data. The offset (Z_ARG3) is
2601 // already accounted for.
2602 //
2603 // - The current SHA state (the intermediate message digest value) is contained
2604 // in an area addressed by Z_ARG2. The area size depends on the SHA variant
2605 // and is accessible via the enum VM_Version::MsgDigest::_SHA<n>_parmBlk_I
2606 //
2607 // - The single-block stub is expected to digest exactly one data block, starting
2608 // at the address passed in Z_ARG1.
2609 //
2610 // - The multi-block stub is expected to digest all data blocks which start in
2611 // the offset interval [srcOff(Z_ARG3), srcLimit(Z_ARG4)). The exact difference
2612 // (srcLimit-srcOff), rounded up to the next multiple of the data block length,
2613 // gives the number of blocks to digest. It must be assumed that the calling code
2614 // provides for a large enough source data buffer.
2615 //
2616 // Compute SHA-1 function.
2617 address generate_SHA1_stub(bool multiBlock, const char* name) {
2618 __ align(CodeEntryAlignment);
2619 StubCodeMark mark(this, "StubRoutines", name);
2620 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2621
2622 const Register srcBuff = Z_ARG1; // Points to first block to process (offset already added).
2623 const Register SHAState = Z_ARG2; // Only on entry. Reused soon thereafter for kimd register pairs.
2624 const Register srcOff = Z_ARG3; // int
2625 const Register srcLimit = Z_ARG4; // Only passed in multiBlock case. int
2626
2627 const Register SHAState_local = Z_R1;
2628 const Register SHAState_save = Z_ARG3;
2629 const Register srcBufLen = Z_ARG2; // Destroys state address, must be copied before.
2630 Label useKLMD, rtn;
2631
2632 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_SHA1); // function code
2633 __ z_lgr(SHAState_local, SHAState); // SHAState == parameter block
2634
2635 if (multiBlock) { // Process everything from offset to limit.
2636
2637 // The following description is valid if we get a raw (unpimped) source data buffer,
2638 // spanning the range between [srcOff(Z_ARG3), srcLimit(Z_ARG4)). As detailed above,
2639 // the calling convention for these stubs is different. We leave the description in
2640 // to inform the reader what must be happening hidden in the calling code.
2641 //
2642 // The data block to be processed can have arbitrary length, i.e. its length does not
2643 // need to be an integer multiple of SHA<n>_datablk. Therefore, we need to implement
2644 // two different paths. If the length is an integer multiple, we use KIMD, saving us
2645 // to copy the SHA state back and forth. If the length is odd, we copy the SHA state
2646 // to the stack, execute a KLMD instruction on it and copy the result back to the
2647 // caller's SHA state location.
2648
2649 // Total #srcBuff blocks to process.
2650 if (VM_Version::has_DistinctOpnds()) {
2651 __ z_srk(srcBufLen, srcLimit, srcOff); // exact difference
2652 __ z_ahi(srcBufLen, VM_Version::MsgDigest::_SHA1_dataBlk-1); // round up
2653 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA1_dataBlk-1)) & 0xffff);
2654 __ z_ark(srcLimit, srcOff, srcBufLen); // Srclimit temporarily holds return value.
2655 __ z_llgfr(srcBufLen, srcBufLen); // Cast to 64-bit.
2656 } else {
2657 __ z_lgfr(srcBufLen, srcLimit); // Exact difference. srcLimit passed as int.
2658 __ z_sgfr(srcBufLen, srcOff); // SrcOff passed as int, now properly casted to long.
2659 __ z_aghi(srcBufLen, VM_Version::MsgDigest::_SHA1_dataBlk-1); // round up
2660 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA1_dataBlk-1)) & 0xffff);
2661 __ z_lgr(srcLimit, srcOff); // SrcLimit temporarily holds return value.
2662 __ z_agr(srcLimit, srcBufLen);
2663 }
2664
2665 // Integral #blocks to digest?
2666 // As a result of the calculations above, srcBufLen MUST be an integer
2667 // multiple of _SHA1_dataBlk, or else we are in big trouble.
2668 // We insert an asm_assert into the KLMD case to guard against that.
2669 __ z_tmll(srcBufLen, VM_Version::MsgDigest::_SHA1_dataBlk-1);
2670 __ z_brc(Assembler::bcondNotAllZero, useKLMD);
2671
2672 // Process all full blocks.
2673 __ kimd(srcBuff);
2674
2675 __ z_lgr(Z_RET, srcLimit); // Offset of first unprocessed byte in buffer.
2676 } else { // Process one data block only.
2677 __ load_const_optimized(srcBufLen, (int)VM_Version::MsgDigest::_SHA1_dataBlk); // #srcBuff bytes to process
2678 __ kimd(srcBuff);
2679 __ add2reg(Z_RET, (int)VM_Version::MsgDigest::_SHA1_dataBlk, srcOff); // Offset of first unprocessed byte in buffer. No 32 to 64 bit extension needed.
2680 }
2681
2682 __ bind(rtn);
2683 __ z_br(Z_R14);
2684
2685 if (multiBlock) {
2686 __ bind(useKLMD);
2687
2688 #if 1
2689 // Security net: this stub is believed to be called for full-sized data blocks only
2690 // NOTE: The following code is believed to be correct, but is is not tested.
2691 __ stop_static("SHA128 stub can digest full data blocks only. Use -XX:-UseSHA as remedy.", 0);
2692 #endif
2693 }
2694
2695 return __ addr_at(start_off);
2696 }
2697
2698 // Compute SHA-256 function.
2699 address generate_SHA256_stub(bool multiBlock, const char* name) {
2700 __ align(CodeEntryAlignment);
2701 StubCodeMark mark(this, "StubRoutines", name);
2702 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2703
2704 const Register srcBuff = Z_ARG1;
2705 const Register SHAState = Z_ARG2; // Only on entry. Reused soon thereafter.
2706 const Register SHAState_local = Z_R1;
2707 const Register SHAState_save = Z_ARG3;
2708 const Register srcOff = Z_ARG3;
2709 const Register srcLimit = Z_ARG4;
2710 const Register srcBufLen = Z_ARG2; // Destroys state address, must be copied before.
2711 Label useKLMD, rtn;
2712
2713 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_SHA256); // function code
2714 __ z_lgr(SHAState_local, SHAState); // SHAState == parameter block
2715
2716 if (multiBlock) { // Process everything from offset to limit.
2717 // The following description is valid if we get a raw (unpimped) source data buffer,
2718 // spanning the range between [srcOff(Z_ARG3), srcLimit(Z_ARG4)). As detailed above,
2719 // the calling convention for these stubs is different. We leave the description in
2720 // to inform the reader what must be happening hidden in the calling code.
2721 //
2722 // The data block to be processed can have arbitrary length, i.e. its length does not
2723 // need to be an integer multiple of SHA<n>_datablk. Therefore, we need to implement
2724 // two different paths. If the length is an integer multiple, we use KIMD, saving us
2725 // to copy the SHA state back and forth. If the length is odd, we copy the SHA state
2726 // to the stack, execute a KLMD instruction on it and copy the result back to the
2727 // caller's SHA state location.
2728
2729 // total #srcBuff blocks to process
2730 if (VM_Version::has_DistinctOpnds()) {
2731 __ z_srk(srcBufLen, srcLimit, srcOff); // exact difference
2732 __ z_ahi(srcBufLen, VM_Version::MsgDigest::_SHA256_dataBlk-1); // round up
2733 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA256_dataBlk-1)) & 0xffff);
2734 __ z_ark(srcLimit, srcOff, srcBufLen); // Srclimit temporarily holds return value.
2735 __ z_llgfr(srcBufLen, srcBufLen); // Cast to 64-bit.
2736 } else {
2737 __ z_lgfr(srcBufLen, srcLimit); // exact difference
2738 __ z_sgfr(srcBufLen, srcOff);
2739 __ z_aghi(srcBufLen, VM_Version::MsgDigest::_SHA256_dataBlk-1); // round up
2740 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA256_dataBlk-1)) & 0xffff);
2741 __ z_lgr(srcLimit, srcOff); // Srclimit temporarily holds return value.
2742 __ z_agr(srcLimit, srcBufLen);
2743 }
2744
2745 // Integral #blocks to digest?
2746 // As a result of the calculations above, srcBufLen MUST be an integer
2747 // multiple of _SHA1_dataBlk, or else we are in big trouble.
2748 // We insert an asm_assert into the KLMD case to guard against that.
2749 __ z_tmll(srcBufLen, VM_Version::MsgDigest::_SHA256_dataBlk-1);
2750 __ z_brc(Assembler::bcondNotAllZero, useKLMD);
2751
2752 // Process all full blocks.
2753 __ kimd(srcBuff);
2754
2755 __ z_lgr(Z_RET, srcLimit); // Offset of first unprocessed byte in buffer.
2756 } else { // Process one data block only.
2757 __ load_const_optimized(srcBufLen, (int)VM_Version::MsgDigest::_SHA256_dataBlk); // #srcBuff bytes to process
2758 __ kimd(srcBuff);
2759 __ add2reg(Z_RET, (int)VM_Version::MsgDigest::_SHA256_dataBlk, srcOff); // Offset of first unprocessed byte in buffer.
2760 }
2761
2762 __ bind(rtn);
2763 __ z_br(Z_R14);
2764
2765 if (multiBlock) {
2766 __ bind(useKLMD);
2767 #if 1
2768 // Security net: this stub is believed to be called for full-sized data blocks only.
2769 // NOTE:
2770 // The following code is believed to be correct, but is is not tested.
2771 __ stop_static("SHA256 stub can digest full data blocks only. Use -XX:-UseSHA as remedy.", 0);
2772 #endif
2773 }
2774
2775 return __ addr_at(start_off);
2776 }
2777
2778 // Compute SHA-512 function.
2779 address generate_SHA512_stub(bool multiBlock, const char* name) {
2780 __ align(CodeEntryAlignment);
2781 StubCodeMark mark(this, "StubRoutines", name);
2782 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2783
2784 const Register srcBuff = Z_ARG1;
2785 const Register SHAState = Z_ARG2; // Only on entry. Reused soon thereafter.
2786 const Register SHAState_local = Z_R1;
2787 const Register SHAState_save = Z_ARG3;
2788 const Register srcOff = Z_ARG3;
2789 const Register srcLimit = Z_ARG4;
2790 const Register srcBufLen = Z_ARG2; // Destroys state address, must be copied before.
2791 Label useKLMD, rtn;
2792
2793 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_SHA512); // function code
2794 __ z_lgr(SHAState_local, SHAState); // SHAState == parameter block
2795
2796 if (multiBlock) { // Process everything from offset to limit.
2797 // The following description is valid if we get a raw (unpimped) source data buffer,
2798 // spanning the range between [srcOff(Z_ARG3), srcLimit(Z_ARG4)). As detailed above,
2799 // the calling convention for these stubs is different. We leave the description in
2800 // to inform the reader what must be happening hidden in the calling code.
2801 //
2802 // The data block to be processed can have arbitrary length, i.e. its length does not
2803 // need to be an integer multiple of SHA<n>_datablk. Therefore, we need to implement
2804 // two different paths. If the length is an integer multiple, we use KIMD, saving us
2805 // to copy the SHA state back and forth. If the length is odd, we copy the SHA state
2806 // to the stack, execute a KLMD instruction on it and copy the result back to the
2807 // caller's SHA state location.
2808
2809 // total #srcBuff blocks to process
2810 if (VM_Version::has_DistinctOpnds()) {
2811 __ z_srk(srcBufLen, srcLimit, srcOff); // exact difference
2812 __ z_ahi(srcBufLen, VM_Version::MsgDigest::_SHA512_dataBlk-1); // round up
2813 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA512_dataBlk-1)) & 0xffff);
2814 __ z_ark(srcLimit, srcOff, srcBufLen); // Srclimit temporarily holds return value.
2815 __ z_llgfr(srcBufLen, srcBufLen); // Cast to 64-bit.
2816 } else {
2817 __ z_lgfr(srcBufLen, srcLimit); // exact difference
2818 __ z_sgfr(srcBufLen, srcOff);
2819 __ z_aghi(srcBufLen, VM_Version::MsgDigest::_SHA512_dataBlk-1); // round up
2820 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA512_dataBlk-1)) & 0xffff);
2821 __ z_lgr(srcLimit, srcOff); // Srclimit temporarily holds return value.
2822 __ z_agr(srcLimit, srcBufLen);
2823 }
2824
2825 // integral #blocks to digest?
2826 // As a result of the calculations above, srcBufLen MUST be an integer
2827 // multiple of _SHA1_dataBlk, or else we are in big trouble.
2828 // We insert an asm_assert into the KLMD case to guard against that.
2829 __ z_tmll(srcBufLen, VM_Version::MsgDigest::_SHA512_dataBlk-1);
2830 __ z_brc(Assembler::bcondNotAllZero, useKLMD);
2831
2832 // Process all full blocks.
2833 __ kimd(srcBuff);
2834
2835 __ z_lgr(Z_RET, srcLimit); // Offset of first unprocessed byte in buffer.
2836 } else { // Process one data block only.
2837 __ load_const_optimized(srcBufLen, (int)VM_Version::MsgDigest::_SHA512_dataBlk); // #srcBuff bytes to process
2838 __ kimd(srcBuff);
2839 __ add2reg(Z_RET, (int)VM_Version::MsgDigest::_SHA512_dataBlk, srcOff); // Offset of first unprocessed byte in buffer.
2840 }
2841
2842 __ bind(rtn);
2843 __ z_br(Z_R14);
2844
2845 if (multiBlock) {
2846 __ bind(useKLMD);
2847 #if 1
2848 // Security net: this stub is believed to be called for full-sized data blocks only
2849 // NOTE:
2850 // The following code is believed to be correct, but is is not tested.
2851 __ stop_static("SHA512 stub can digest full data blocks only. Use -XX:-UseSHA as remedy.", 0);
2852 #endif
2853 }
2854
2855 return __ addr_at(start_off);
2856 }
2857
2858
2859 /**
2860 * Arguments:
2861 *
2862 * Inputs:
2863 * Z_ARG1 - int crc
2864 * Z_ARG2 - byte* buf
2865 * Z_ARG3 - int length (of buffer)
2866 *
2867 * Result:
2868 * Z_RET - int crc result
2869 **/
2870 // Compute CRC function (generic, for all polynomials).
2871 void generate_CRC_updateBytes(const char* name, Register table, bool invertCRC) {
2872
2873 // arguments to kernel_crc32:
2874 Register crc = Z_ARG1; // Current checksum, preset by caller or result from previous call, int.
2875 Register data = Z_ARG2; // source byte array
2876 Register dataLen = Z_ARG3; // #bytes to process, int
2877 // Register table = Z_ARG4; // crc table address. Preloaded and passed in by caller.
2878 const Register t0 = Z_R10; // work reg for kernel* emitters
2879 const Register t1 = Z_R11; // work reg for kernel* emitters
2880 const Register t2 = Z_R12; // work reg for kernel* emitters
2881 const Register t3 = Z_R13; // work reg for kernel* emitters
2882
2883
2884 assert_different_registers(crc, data, dataLen, table);
2885
2886 // We pass these values as ints, not as longs as required by C calling convention.
2887 // Crc used as int.
2888 __ z_llgfr(dataLen, dataLen);
2889
2890 __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2891 __ z_stmg(Z_R10, Z_R13, 1*8, Z_SP); // Spill regs 10..11 to make them available as work registers.
2892 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, invertCRC);
2893 __ z_lmg(Z_R10, Z_R13, 1*8, Z_SP); // Spill regs 10..11 back from stack.
2894 __ resize_frame(+(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2895
2896 __ z_llgfr(Z_RET, crc); // Updated crc is function result. No copying required, just zero upper 32 bits.
2897 __ z_br(Z_R14); // Result already in Z_RET == Z_ARG1.
2898 }
2899
2900
2901 // Compute CRC32 function.
2902 address generate_CRC32_updateBytes(const char* name) {
2903 __ align(CodeEntryAlignment);
2904 StubCodeMark mark(this, "StubRoutines", name);
2905 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2906
2907 assert(UseCRC32Intrinsics, "should not generate this stub (%s) with CRC32 intrinsics disabled", name);
2908
2909 BLOCK_COMMENT("CRC32_updateBytes {");
2910 Register table = Z_ARG4; // crc32 table address.
2911 StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
2912
2913 generate_CRC_updateBytes(name, table, true);
2914 BLOCK_COMMENT("} CRC32_updateBytes");
2915
2916 return __ addr_at(start_off);
2917 }
2918
2919
2920 // Compute CRC32C function.
2921 address generate_CRC32C_updateBytes(const char* name) {
2922 __ align(CodeEntryAlignment);
2923 StubCodeMark mark(this, "StubRoutines", name);
2924 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2925
2926 assert(UseCRC32CIntrinsics, "should not generate this stub (%s) with CRC32C intrinsics disabled", name);
2927
2928 BLOCK_COMMENT("CRC32C_updateBytes {");
2929 Register table = Z_ARG4; // crc32c table address.
2930 StubRoutines::zarch::generate_load_crc32c_table_addr(_masm, table);
2931
2932 generate_CRC_updateBytes(name, table, false);
2933 BLOCK_COMMENT("} CRC32C_updateBytes");
2934
2935 return __ addr_at(start_off);
2936 }
2937
2938
2939 // Arguments:
2940 // Z_ARG1 - x address
2941 // Z_ARG2 - x length
2942 // Z_ARG3 - y address
2943 // Z_ARG4 - y length
2944 // Z_ARG5 - z address
2945 address generate_multiplyToLen() {
2946 __ align(CodeEntryAlignment);
2947 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
2948
2949 address start = __ pc();
2950
2951 const Register x = Z_ARG1;
2952 const Register xlen = Z_ARG2;
2953 const Register y = Z_ARG3;
2954 const Register ylen = Z_ARG4;
2955 const Register z = Z_ARG5;
2956
2957 // Next registers will be saved on stack in multiply_to_len().
2958 const Register tmp1 = Z_tmp_1;
2959 const Register tmp2 = Z_tmp_2;
2960 const Register tmp3 = Z_tmp_3;
2961 const Register tmp4 = Z_tmp_4;
2962 const Register tmp5 = Z_R9;
2963
2964 BLOCK_COMMENT("Entry:");
2965
2966 __ z_llgfr(xlen, xlen);
2967 __ z_llgfr(ylen, ylen);
2968
2969 __ multiply_to_len(x, xlen, y, ylen, z, tmp1, tmp2, tmp3, tmp4, tmp5);
2970
2971 __ z_br(Z_R14); // Return to caller.
2972
2973 return start;
2974 }
2975
2976 address generate_method_entry_barrier() {
2977 __ align(CodeEntryAlignment);
2978 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
2979
2980 address start = __ pc();
2981
2982 int nbytes_volatile = (8 + 5) * BytesPerWord;
2983
2984 // VM-Call Prologue
2985 __ save_return_pc();
2986 __ push_frame_abi160(nbytes_volatile);
2987 __ save_volatile_regs(Z_SP, frame::z_abi_160_size, true, false);
2988
2989 // Prep arg for VM call
2990 // Create ptr to stored return_pc in caller frame.
2991 __ z_la(Z_ARG1, _z_abi(return_pc) + frame::z_abi_160_size + nbytes_volatile, Z_R0, Z_SP);
2992
2993 // VM-Call: BarrierSetNMethod::nmethod_stub_entry_barrier(address* return_address_ptr)
2994 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSetNMethod::nmethod_stub_entry_barrier));
2995 __ z_ltr(Z_R0_scratch, Z_RET);
2996
2997 // VM-Call Epilogue
2998 __ restore_volatile_regs(Z_SP, frame::z_abi_160_size, true, false);
2999 __ pop_frame();
3000 __ restore_return_pc();
3001
3002 // Check return val of VM-Call
3003 __ z_bcr(Assembler::bcondZero, Z_R14);
3004
3005 // Pop frame built in prologue.
3006 // Required so wrong_method_stub can deduce caller.
3007 __ pop_frame();
3008 __ restore_return_pc();
3009
3010 // VM-Call indicates deoptimization required
3011 __ load_const_optimized(Z_R1_scratch, SharedRuntime::get_handle_wrong_method_stub());
3012 __ z_br(Z_R1_scratch);
3013
3014 return start;
3015 }
3016
3017 address generate_cont_thaw(bool return_barrier, bool exception) {
3018 if (!Continuations::enabled()) return nullptr;
3019 Unimplemented();
3020 return nullptr;
3021 }
3022
3023 address generate_cont_thaw() {
3024 if (!Continuations::enabled()) return nullptr;
3025 Unimplemented();
3026 return nullptr;
3027 }
3028
3029 address generate_cont_returnBarrier() {
3030 if (!Continuations::enabled()) return nullptr;
3031 Unimplemented();
3032 return nullptr;
3033 }
3034
3035 address generate_cont_returnBarrier_exception() {
3036 if (!Continuations::enabled()) return nullptr;
3037 Unimplemented();
3038 return nullptr;
3039 }
3040
3041 // exception handler for upcall stubs
3042 address generate_upcall_stub_exception_handler() {
3043 StubCodeMark mark(this, "StubRoutines", "upcall stub exception handler");
3044 address start = __ pc();
3045
3046 // Native caller has no idea how to handle exceptions,
3047 // so we just crash here. Up to callee to catch exceptions.
3048 __ verify_oop(Z_ARG1);
3049 __ load_const_optimized(Z_R1_scratch, CAST_FROM_FN_PTR(uint64_t, UpcallLinker::handle_uncaught_exception));
3050 __ call_c(Z_R1_scratch);
3051 __ should_not_reach_here();
3052
3053 return start;
3054 }
3055
3056 // load Method* target of MethodHandle
3057 // Z_ARG1 = jobject receiver
3058 // Z_method = Method* result
3059 address generate_upcall_stub_load_target() {
3060 StubCodeMark mark(this, "StubRoutines", "upcall_stub_load_target");
3061 address start = __ pc();
3062
3063 __ resolve_global_jobject(Z_ARG1, Z_tmp_1, Z_tmp_2);
3064 // Load target method from receiver
3065 __ load_heap_oop(Z_method, Address(Z_ARG1, java_lang_invoke_MethodHandle::form_offset()),
3066 noreg, noreg, IS_NOT_NULL);
3067 __ load_heap_oop(Z_method, Address(Z_method, java_lang_invoke_LambdaForm::vmentry_offset()),
3068 noreg, noreg, IS_NOT_NULL);
3069 __ load_heap_oop(Z_method, Address(Z_method, java_lang_invoke_MemberName::method_offset()),
3070 noreg, noreg, IS_NOT_NULL);
3071 __ z_lg(Z_method, Address(Z_method, java_lang_invoke_ResolvedMethodName::vmtarget_offset()));
3072 __ z_stg(Z_method, Address(Z_thread, JavaThread::callee_target_offset())); // just in case callee is deoptimized
3073
3074 __ z_br(Z_R14);
3075
3076 return start;
3077 }
3078
3079 void generate_initial_stubs() {
3080 // Generates all stubs and initializes the entry points.
3081
3082 // Entry points that exist in all platforms.
3083 // Note: This is code that could be shared among different
3084 // platforms - however the benefit seems to be smaller than the
3085 // disadvantage of having a much more complicated generator
3086 // structure. See also comment in stubRoutines.hpp.
3087 StubRoutines::_forward_exception_entry = generate_forward_exception();
3088
3089 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address);
3090 StubRoutines::_catch_exception_entry = generate_catch_exception();
3091
3092 //----------------------------------------------------------------------
3093 // Entry points that are platform specific.
3094
3095 if (UseCRC32Intrinsics) {
3096 StubRoutines::_crc_table_adr = (address)StubRoutines::zarch::_crc_table;
3097 StubRoutines::_updateBytesCRC32 = generate_CRC32_updateBytes("CRC32_updateBytes");
3098 }
3099
3100 if (UseCRC32CIntrinsics) {
3101 StubRoutines::_crc32c_table_addr = (address)StubRoutines::zarch::_crc32c_table;
3102 StubRoutines::_updateBytesCRC32C = generate_CRC32C_updateBytes("CRC32C_updateBytes");
3103 }
3104
3105 // Comapct string intrinsics: Translate table for string inflate intrinsic. Used by trot instruction.
3106 StubRoutines::zarch::_trot_table_addr = (address)StubRoutines::zarch::_trot_table;
3107 }
3108
3109 void generate_continuation_stubs() {
3110 if (!Continuations::enabled()) return;
3111
3112 // Continuation stubs:
3113 StubRoutines::_cont_thaw = generate_cont_thaw();
3114 StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
3115 StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
3116 }
3117
3118 void generate_final_stubs() {
3119 // Generates all stubs and initializes the entry points.
3120
3121 StubRoutines::zarch::_partial_subtype_check = generate_partial_subtype_check();
3122
3123 // Support for verify_oop (must happen after universe_init).
3124 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop_subroutine();
3125
3126 // Arraycopy stubs used by compilers.
3127 generate_arraycopy_stubs();
3128
3129 // nmethod entry barriers for concurrent class unloading
3130 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
3131 if (bs_nm != nullptr) {
3132 StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
3133 }
3134
3135 StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
3136 StubRoutines::_upcall_stub_load_target = generate_upcall_stub_load_target();
3137 }
3138
3139 void generate_compiler_stubs() {
3140 #if COMPILER2_OR_JVMCI
3141 // Generate AES intrinsics code.
3142 if (UseAESIntrinsics) {
3143 if (VM_Version::has_Crypto_AES()) {
3144 StubRoutines::_aescrypt_encryptBlock = generate_AES_encryptBlock("AES_encryptBlock");
3145 StubRoutines::_aescrypt_decryptBlock = generate_AES_decryptBlock("AES_decryptBlock");
3146 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_AES_encrypt("AES_encryptBlock_chaining");
3147 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_AES_decrypt("AES_decryptBlock_chaining");
3148 } else {
3149 // In PRODUCT builds, the function pointers will keep their initial (null) value.
3150 // LibraryCallKit::try_to_inline() will return false then, preventing the intrinsic to be called.
3151 assert(VM_Version::has_Crypto_AES(), "Inconsistent settings. Check vm_version_s390.cpp");
3152 }
3153 }
3154
3155 if (UseAESCTRIntrinsics) {
3156 if (VM_Version::has_Crypto_AES_CTR()) {
3157 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt("counterMode_AESCrypt");
3158 } else {
3159 // In PRODUCT builds, the function pointers will keep their initial (null) value.
3160 // LibraryCallKit::try_to_inline() will return false then, preventing the intrinsic to be called.
3161 assert(VM_Version::has_Crypto_AES_CTR(), "Inconsistent settings. Check vm_version_s390.cpp");
3162 }
3163 }
3164
3165 // Generate GHASH intrinsics code
3166 if (UseGHASHIntrinsics) {
3167 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
3168 }
3169
3170 // Generate SHA1/SHA256/SHA512 intrinsics code.
3171 if (UseSHA1Intrinsics) {
3172 StubRoutines::_sha1_implCompress = generate_SHA1_stub(false, "SHA1_singleBlock");
3173 StubRoutines::_sha1_implCompressMB = generate_SHA1_stub(true, "SHA1_multiBlock");
3174 }
3175 if (UseSHA256Intrinsics) {
3176 StubRoutines::_sha256_implCompress = generate_SHA256_stub(false, "SHA256_singleBlock");
3177 StubRoutines::_sha256_implCompressMB = generate_SHA256_stub(true, "SHA256_multiBlock");
3178 }
3179 if (UseSHA512Intrinsics) {
3180 StubRoutines::_sha512_implCompress = generate_SHA512_stub(false, "SHA512_singleBlock");
3181 StubRoutines::_sha512_implCompressMB = generate_SHA512_stub(true, "SHA512_multiBlock");
3182 }
3183 #ifdef COMPILER2
3184 if (UseMultiplyToLenIntrinsic) {
3185 StubRoutines::_multiplyToLen = generate_multiplyToLen();
3186 }
3187 if (UseMontgomeryMultiplyIntrinsic) {
3188 StubRoutines::_montgomeryMultiply
3189 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
3190 }
3191 if (UseMontgomerySquareIntrinsic) {
3192 StubRoutines::_montgomerySquare
3193 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
3194 }
3195 if (UseSecondarySupersTable) {
3196 StubRoutines::_lookup_secondary_supers_table_slow_path_stub = generate_lookup_secondary_supers_table_slow_path_stub();
3197 if (!InlineSecondarySupersTest) {
3198 for (int slot = 0; slot < Klass::SECONDARY_SUPERS_TABLE_SIZE; slot++) {
3199 StubRoutines::_lookup_secondary_supers_table_stubs[slot] = generate_lookup_secondary_supers_table_stub(slot);
3200 }
3201 }
3202 }
3203 #endif
3204 #endif // COMPILER2_OR_JVMCI
3205 }
3206
3207 public:
3208 StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
3209 switch(kind) {
3210 case Initial_stubs:
3211 generate_initial_stubs();
3212 break;
3213 case Continuation_stubs:
3214 generate_continuation_stubs();
3215 break;
3216 case Compiler_stubs:
3217 generate_compiler_stubs();
3218 break;
3219 case Final_stubs:
3220 generate_final_stubs();
3221 break;
3222 default:
3223 fatal("unexpected stubs kind: %d", kind);
3224 break;
3225 };
3226 }
3227
3228 private:
3229 int _stub_count;
3230 void stub_prolog(StubCodeDesc* cdesc) {
3231 #ifdef ASSERT
3232 // Put extra information in the stub code, to make it more readable.
3233 // Write the high part of the address.
3234 // [RGV] Check if there is a dependency on the size of this prolog.
3235 __ emit_data((intptr_t)cdesc >> 32);
3236 __ emit_data((intptr_t)cdesc);
3237 __ emit_data(++_stub_count);
3238 #endif
3239 align(true);
3240 }
3241
3242 void align(bool at_header = false) {
3243 // z/Architecture cache line size is 256 bytes.
3244 // There is no obvious benefit in aligning stub
3245 // code to cache lines. Use CodeEntryAlignment instead.
3246 const unsigned int icache_line_size = CodeEntryAlignment;
3247 const unsigned int icache_half_line_size = MIN2<unsigned int>(32, CodeEntryAlignment);
3248
3249 if (at_header) {
3250 while ((intptr_t)(__ pc()) % icache_line_size != 0) {
3251 __ z_illtrap();
3252 }
3253 } else {
3254 while ((intptr_t)(__ pc()) % icache_half_line_size != 0) {
3255 __ z_nop();
3256 }
3257 }
3258 }
3259
3260 };
3261
3262 void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
3263 StubGenerator g(code, kind);
3264 }