1 /*
2 * Copyright (c) 2016, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016, 2023 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 // Continuation point for throwing of implicit exceptions that are
576 // not handled in the current activation. Fabricates an exception
577 // oop and initiates normal exception dispatching in this
578 // frame. Only callee-saved registers are preserved (through the
579 // normal RegisterMap handling). If the compiler
580 // needs all registers to be preserved between the fault point and
581 // the exception handler then it must assume responsibility for that
582 // in AbstractCompiler::continuation_for_implicit_null_exception or
583 // continuation_for_implicit_division_by_zero_exception. All other
584 // implicit exceptions (e.g., NullPointerException or
585 // AbstractMethodError on entry) are either at call sites or
586 // otherwise assume that stack unwinding will be initiated, so
587 // caller saved registers were assumed volatile in the compiler.
588
589 // Note that we generate only this stub into a RuntimeStub, because
590 // it needs to be properly traversed and ignored during GC, so we
591 // change the meaning of the "__" macro within this method.
592
593 // Note: the routine set_pc_not_at_call_for_caller in
594 // SharedRuntime.cpp requires that this code be generated into a
595 // RuntimeStub.
596 #undef __
597 #define __ masm->
598
599 address generate_throw_exception(const char* name, address runtime_entry,
600 bool restore_saved_exception_pc,
601 Register arg1 = noreg, Register arg2 = noreg) {
602 assert_different_registers(arg1, Z_R0_scratch); // would be destroyed by push_frame()
603 assert_different_registers(arg2, Z_R0_scratch); // would be destroyed by push_frame()
604
605 int insts_size = 256;
606 int locs_size = 0;
607 CodeBuffer code(name, insts_size, locs_size);
608 MacroAssembler* masm = new MacroAssembler(&code);
609 int framesize_in_bytes;
610 address start = __ pc();
611
612 __ save_return_pc();
613 framesize_in_bytes = __ push_frame_abi160(0);
614
615 address frame_complete_pc = __ pc();
616 if (restore_saved_exception_pc) {
617 __ unimplemented("StubGenerator::throw_exception", 74);
618 }
619
620 // Note that we always have a runtime stub frame on the top of stack at this point.
621 __ get_PC(Z_R1);
622 __ set_last_Java_frame(/*sp*/Z_SP, /*pc*/Z_R1);
623
624 // Do the call.
625 BLOCK_COMMENT("call runtime_entry");
626 __ call_VM_leaf(runtime_entry, Z_thread, arg1, arg2);
627
628 __ reset_last_Java_frame();
629
630 #ifdef ASSERT
631 // Make sure that this code is only executed if there is a pending exception.
632 { Label L;
633 __ z_lg(Z_R0,
634 in_bytes(Thread::pending_exception_offset()),
635 Z_thread);
636 __ z_ltgr(Z_R0, Z_R0);
637 __ z_brne(L);
638 __ stop("StubRoutines::throw_exception: no pending exception");
639 __ bind(L);
640 }
641 #endif
642
643 __ pop_frame();
644 __ restore_return_pc();
645
646 __ load_const_optimized(Z_R1, StubRoutines::forward_exception_entry());
647 __ z_br(Z_R1);
648
649 RuntimeStub* stub =
650 RuntimeStub::new_runtime_stub(name, &code,
651 frame_complete_pc - start,
652 framesize_in_bytes/wordSize,
653 nullptr /*oop_maps*/, false);
654
655 return stub->entry_point();
656 }
657
658 #undef __
659 #ifdef PRODUCT
660 #define __ _masm->
661 #else
662 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
663 #endif
664
665 // Support for uint StubRoutine::zarch::partial_subtype_check(Klass
666 // sub, Klass super);
667 //
668 // Arguments:
669 // ret : Z_RET, returned
670 // sub : Z_ARG2, argument, not changed
671 // super: Z_ARG3, argument, not changed
672 //
673 // raddr: Z_R14, blown by call
674 //
675 address generate_partial_subtype_check() {
676 StubCodeMark mark(this, "StubRoutines", "partial_subtype_check");
677 Label miss;
678
679 address start = __ pc();
680
681 const Register Rsubklass = Z_ARG2; // subklass
682 const Register Rsuperklass = Z_ARG3; // superklass
683
684 // No args, but tmp registers that are killed.
685 const Register Rlength = Z_ARG4; // cache array length
686 const Register Rarray_ptr = Z_ARG5; // Current value from cache array.
687
688 if (UseCompressedOops) {
689 assert(Universe::heap() != nullptr, "java heap must be initialized to generate partial_subtype_check stub");
690 }
691
692 // Always take the slow path.
693 __ check_klass_subtype_slow_path(Rsubklass, Rsuperklass,
694 Rarray_ptr, Rlength, nullptr, &miss);
695
696 // Match falls through here.
697 __ clear_reg(Z_RET); // Zero indicates a match. Set EQ flag in CC.
698 __ z_br(Z_R14);
699
700 __ BIND(miss);
701 __ load_const_optimized(Z_RET, 1); // One indicates a miss.
702 __ z_ltgr(Z_RET, Z_RET); // Set NE flag in CR.
703 __ z_br(Z_R14);
704
705 return start;
706 }
707
708 #if !defined(PRODUCT)
709 // Wrapper which calls oopDesc::is_oop_or_null()
710 // Only called by MacroAssembler::verify_oop
711 static void verify_oop_helper(const char* message, oopDesc* o) {
712 if (!oopDesc::is_oop_or_null(o)) {
713 fatal("%s. oop: " PTR_FORMAT, message, p2i(o));
714 }
715 ++ StubRoutines::_verify_oop_count;
716 }
717 #endif
718
719 // Return address of code to be called from code generated by
720 // MacroAssembler::verify_oop.
721 //
722 // Don't generate, rather use C++ code.
723 address generate_verify_oop_subroutine() {
724 // Don't generate a StubCodeMark, because no code is generated!
725 // Generating the mark triggers notifying the oprofile jvmti agent
726 // about the dynamic code generation, but the stub without
727 // code (code_size == 0) confuses opjitconv
728 // StubCodeMark mark(this, "StubRoutines", "verify_oop_stub");
729
730 address start = 0;
731
732 #if !defined(PRODUCT)
733 start = CAST_FROM_FN_PTR(address, verify_oop_helper);
734 #endif
735
736 return start;
737 }
738
739 // This is to test that the count register contains a positive int value.
740 // Required because C2 does not respect int to long conversion for stub calls.
741 void assert_positive_int(Register count) {
742 #ifdef ASSERT
743 __ z_srag(Z_R0, count, 31); // Just leave the sign (must be zero) in Z_R0.
744 __ asm_assert(Assembler::bcondZero, "missing zero extend", 0xAFFE);
745 #endif
746 }
747
748 // Generate overlap test for array copy stubs.
749 // If no actual overlap is detected, control is transferred to the
750 // "normal" copy stub (entry address passed in disjoint_copy_target).
751 // Otherwise, execution continues with the code generated by the
752 // caller of array_overlap_test.
753 //
754 // Input:
755 // Z_ARG1 - from
756 // Z_ARG2 - to
757 // Z_ARG3 - element count
758 void array_overlap_test(address disjoint_copy_target, int log2_elem_size) {
759 __ MacroAssembler::compare_and_branch_optimized(Z_ARG2, Z_ARG1, Assembler::bcondNotHigh,
760 disjoint_copy_target, /*len64=*/true, /*has_sign=*/false);
761
762 Register index = Z_ARG3;
763 if (log2_elem_size > 0) {
764 __ z_sllg(Z_R1, Z_ARG3, log2_elem_size); // byte count
765 index = Z_R1;
766 }
767 __ add2reg_with_index(Z_R1, 0, index, Z_ARG1); // First byte after "from" range.
768
769 __ MacroAssembler::compare_and_branch_optimized(Z_R1, Z_ARG2, Assembler::bcondNotHigh,
770 disjoint_copy_target, /*len64=*/true, /*has_sign=*/false);
771
772 // Destructive overlap: let caller generate code for that.
773 }
774
775 // Generate stub for disjoint array copy. If "aligned" is true, the
776 // "from" and "to" addresses are assumed to be heapword aligned.
777 //
778 // Arguments for generated stub:
779 // from: Z_ARG1
780 // to: Z_ARG2
781 // count: Z_ARG3 treated as signed
782 void generate_disjoint_copy(bool aligned, int element_size,
783 bool branchToEnd,
784 bool restoreArgs) {
785 // This is the zarch specific stub generator for general array copy tasks.
786 // It has the following prereqs and features:
787 //
788 // - No destructive overlap allowed (else unpredictable results).
789 // - Destructive overlap does not exist if the leftmost byte of the target
790 // does not coincide with any of the source bytes (except the leftmost).
791 //
792 // Register usage upon entry:
793 // Z_ARG1 == Z_R2 : address of source array
794 // Z_ARG2 == Z_R3 : address of target array
795 // Z_ARG3 == Z_R4 : length of operands (# of elements on entry)
796 //
797 // Register usage within the generator:
798 // - Z_R0 and Z_R1 are KILLed by the stub routine (target addr/len).
799 // Used as pair register operand in complex moves, scratch registers anyway.
800 // - Z_R5 is KILLed by the stub routine (source register pair addr/len) (even/odd reg).
801 // Same as R0/R1, but no scratch register.
802 // - Z_ARG1, Z_ARG2, Z_ARG3 are USEd but preserved by the stub routine,
803 // but they might get temporarily overwritten.
804
805 Register save_reg = Z_ARG4; // (= Z_R5), holds original target operand address for restore.
806
807 {
808 Register llen_reg = Z_R1; // Holds left operand len (odd reg).
809 Register laddr_reg = Z_R0; // Holds left operand addr (even reg), overlaps with data_reg.
810 Register rlen_reg = Z_R5; // Holds right operand len (odd reg), overlaps with save_reg.
811 Register raddr_reg = Z_R4; // Holds right operand addr (even reg), overlaps with len_reg.
812
813 Register data_reg = Z_R0; // Holds copied data chunk in alignment process and copy loop.
814 Register len_reg = Z_ARG3; // Holds operand len (#elements at entry, #bytes shortly after).
815 Register dst_reg = Z_ARG2; // Holds left (target) operand addr.
816 Register src_reg = Z_ARG1; // Holds right (source) operand addr.
817
818 Label doMVCLOOP, doMVCLOOPcount, doMVCLOOPiterate;
819 Label doMVCUnrolled;
820 NearLabel doMVC, doMVCgeneral, done;
821 Label MVC_template;
822 address pcMVCblock_b, pcMVCblock_e;
823
824 bool usedMVCLE = true;
825 bool usedMVCLOOP = true;
826 bool usedMVCUnrolled = false;
827 bool usedMVC = false;
828 bool usedMVCgeneral = false;
829
830 int stride;
831 Register stride_reg;
832 Register ix_reg;
833
834 assert((element_size<=256) && (256%element_size == 0), "element size must be <= 256, power of 2");
835 unsigned int log2_size = exact_log2(element_size);
836
837 switch (element_size) {
838 case 1: BLOCK_COMMENT("ARRAYCOPY DISJOINT byte {"); break;
839 case 2: BLOCK_COMMENT("ARRAYCOPY DISJOINT short {"); break;
840 case 4: BLOCK_COMMENT("ARRAYCOPY DISJOINT int {"); break;
841 case 8: BLOCK_COMMENT("ARRAYCOPY DISJOINT long {"); break;
842 default: BLOCK_COMMENT("ARRAYCOPY DISJOINT {"); break;
843 }
844
845 assert_positive_int(len_reg);
846
847 BLOCK_COMMENT("preparation {");
848
849 // No copying if len <= 0.
850 if (branchToEnd) {
851 __ compare64_and_branch(len_reg, (intptr_t) 0, Assembler::bcondNotHigh, done);
852 } else {
853 if (VM_Version::has_CompareBranch()) {
854 __ z_cgib(len_reg, 0, Assembler::bcondNotHigh, 0, Z_R14);
855 } else {
856 __ z_ltgr(len_reg, len_reg);
857 __ z_bcr(Assembler::bcondNotPositive, Z_R14);
858 }
859 }
860
861 // Prefetch just one cache line. Speculative opt for short arrays.
862 // Do not use Z_R1 in prefetch. Is undefined here.
863 if (VM_Version::has_Prefetch()) {
864 __ z_pfd(0x01, 0, Z_R0, src_reg); // Fetch access.
865 __ z_pfd(0x02, 0, Z_R0, dst_reg); // Store access.
866 }
867
868 BLOCK_COMMENT("} preparation");
869
870 // Save args only if really needed.
871 // Keep len test local to branch. Is generated only once.
872
873 BLOCK_COMMENT("mode selection {");
874
875 // Special handling for arrays with only a few elements.
876 // Nothing fancy: just an executed MVC.
877 if (log2_size > 0) {
878 __ z_sllg(Z_R1, len_reg, log2_size); // Remember #bytes in Z_R1.
879 }
880 if (element_size != 8) {
881 __ z_cghi(len_reg, 256/element_size);
882 __ z_brnh(doMVC);
883 usedMVC = true;
884 }
885 if (element_size == 8) { // Long and oop arrays are always aligned.
886 __ z_cghi(len_reg, 256/element_size);
887 __ z_brnh(doMVCUnrolled);
888 usedMVCUnrolled = true;
889 }
890
891 // Prefetch another cache line. We, for sure, have more than one line to copy.
892 if (VM_Version::has_Prefetch()) {
893 __ z_pfd(0x01, 256, Z_R0, src_reg); // Fetch access.
894 __ z_pfd(0x02, 256, Z_R0, dst_reg); // Store access.
895 }
896
897 if (restoreArgs) {
898 // Remember entry value of ARG2 to restore all arguments later from that knowledge.
899 __ z_lgr(save_reg, dst_reg);
900 }
901
902 __ z_cghi(len_reg, 4096/element_size);
903 if (log2_size == 0) {
904 __ z_lgr(Z_R1, len_reg); // Init Z_R1 with #bytes
905 }
906 __ z_brnh(doMVCLOOP);
907
908 // Fall through to MVCLE case.
909
910 BLOCK_COMMENT("} mode selection");
911
912 // MVCLE: for long arrays
913 // DW aligned: Best performance for sizes > 4kBytes.
914 // unaligned: Least complex for sizes > 256 bytes.
915 if (usedMVCLE) {
916 BLOCK_COMMENT("mode MVCLE {");
917
918 // Setup registers for mvcle.
919 //__ z_lgr(llen_reg, len_reg);// r1 <- r4 #bytes already in Z_R1, aka llen_reg.
920 __ z_lgr(laddr_reg, dst_reg); // r0 <- r3
921 __ z_lgr(raddr_reg, src_reg); // r4 <- r2
922 __ z_lgr(rlen_reg, llen_reg); // r5 <- r1
923
924 __ MacroAssembler::move_long_ext(laddr_reg, raddr_reg, 0xb0); // special: bypass cache
925 // __ MacroAssembler::move_long_ext(laddr_reg, raddr_reg, 0xb8); // special: Hold data in cache.
926 // __ MacroAssembler::move_long_ext(laddr_reg, raddr_reg, 0);
927
928 if (restoreArgs) {
929 // MVCLE updates the source (Z_R4,Z_R5) and target (Z_R0,Z_R1) register pairs.
930 // Dst_reg (Z_ARG2) and src_reg (Z_ARG1) are left untouched. No restore required.
931 // Len_reg (Z_ARG3) is destroyed and must be restored.
932 __ z_slgr(laddr_reg, dst_reg); // copied #bytes
933 if (log2_size > 0) {
934 __ z_srag(Z_ARG3, laddr_reg, log2_size); // Convert back to #elements.
935 } else {
936 __ z_lgr(Z_ARG3, laddr_reg);
937 }
938 }
939 if (branchToEnd) {
940 __ z_bru(done);
941 } else {
942 __ z_br(Z_R14);
943 }
944 BLOCK_COMMENT("} mode MVCLE");
945 }
946 // No fallthru possible here.
947
948 // MVCUnrolled: for short, aligned arrays.
949
950 if (usedMVCUnrolled) {
951 BLOCK_COMMENT("mode MVC unrolled {");
952 stride = 8;
953
954 // Generate unrolled MVC instructions.
955 for (int ii = 32; ii > 1; ii--) {
956 __ z_mvc(0, ii * stride-1, dst_reg, 0, src_reg); // ii*8 byte copy
957 if (branchToEnd) {
958 __ z_bru(done);
959 } else {
960 __ z_br(Z_R14);
961 }
962 }
963
964 pcMVCblock_b = __ pc();
965 __ z_mvc(0, 1 * stride-1, dst_reg, 0, src_reg); // 8 byte copy
966 if (branchToEnd) {
967 __ z_bru(done);
968 } else {
969 __ z_br(Z_R14);
970 }
971
972 pcMVCblock_e = __ pc();
973 Label MVC_ListEnd;
974 __ bind(MVC_ListEnd);
975
976 // This is an absolute fast path:
977 // - Array len in bytes must be not greater than 256.
978 // - Array len in bytes must be an integer mult of DW
979 // to save expensive handling of trailing bytes.
980 // - Argument restore is not done,
981 // i.e. previous code must not alter arguments (this code doesn't either).
982
983 __ bind(doMVCUnrolled);
984
985 // Avoid mul, prefer shift where possible.
986 // Combine shift right (for #DW) with shift left (for block size).
987 // Set CC for zero test below (asm_assert).
988 // Note: #bytes comes in Z_R1, #DW in len_reg.
989 unsigned int MVCblocksize = pcMVCblock_e - pcMVCblock_b;
990 unsigned int logMVCblocksize = 0xffffffffU; // Pacify compiler ("used uninitialized" warning).
991
992 if (log2_size > 0) { // Len was scaled into Z_R1.
993 switch (MVCblocksize) {
994
995 case 8: logMVCblocksize = 3;
996 __ z_ltgr(Z_R0, Z_R1); // #bytes is index
997 break; // reasonable size, use shift
998
999 case 16: logMVCblocksize = 4;
1000 __ z_slag(Z_R0, Z_R1, logMVCblocksize-log2_size);
1001 break; // reasonable size, use shift
1002
1003 default: logMVCblocksize = 0;
1004 __ z_ltgr(Z_R0, len_reg); // #DW for mul
1005 break; // all other sizes: use mul
1006 }
1007 } else {
1008 guarantee(log2_size, "doMVCUnrolled: only for DW entities");
1009 }
1010
1011 // This test (and branch) is redundant. Previous code makes sure that
1012 // - element count > 0
1013 // - element size == 8.
1014 // Thus, len reg should never be zero here. We insert an asm_assert() here,
1015 // just to double-check and to be on the safe side.
1016 __ asm_assert(false, "zero len cannot occur", 99);
1017
1018 __ z_larl(Z_R1, MVC_ListEnd); // Get addr of last instr block.
1019 // Avoid mul, prefer shift where possible.
1020 if (logMVCblocksize == 0) {
1021 __ z_mghi(Z_R0, MVCblocksize);
1022 }
1023 __ z_slgr(Z_R1, Z_R0);
1024 __ z_br(Z_R1);
1025 BLOCK_COMMENT("} mode MVC unrolled");
1026 }
1027 // No fallthru possible here.
1028
1029 // MVC execute template
1030 // Must always generate. Usage may be switched on below.
1031 // There is no suitable place after here to put the template.
1032 __ bind(MVC_template);
1033 __ z_mvc(0,0,dst_reg,0,src_reg); // Instr template, never exec directly!
1034
1035
1036 // MVC Loop: for medium-sized arrays
1037
1038 // Only for DW aligned arrays (src and dst).
1039 // #bytes to copy must be at least 256!!!
1040 // Non-aligned cases handled separately.
1041 stride = 256;
1042 stride_reg = Z_R1; // Holds #bytes when control arrives here.
1043 ix_reg = Z_ARG3; // Alias for len_reg.
1044
1045
1046 if (usedMVCLOOP) {
1047 BLOCK_COMMENT("mode MVC loop {");
1048 __ bind(doMVCLOOP);
1049
1050 __ z_lcgr(ix_reg, Z_R1); // Ix runs from -(n-2)*stride to 1*stride (inclusive).
1051 __ z_llill(stride_reg, stride);
1052 __ add2reg(ix_reg, 2*stride); // Thus: increment ix by 2*stride.
1053
1054 __ bind(doMVCLOOPiterate);
1055 __ z_mvc(0, stride-1, dst_reg, 0, src_reg);
1056 __ add2reg(dst_reg, stride);
1057 __ add2reg(src_reg, stride);
1058 __ bind(doMVCLOOPcount);
1059 __ z_brxlg(ix_reg, stride_reg, doMVCLOOPiterate);
1060
1061 // Don 't use add2reg() here, since we must set the condition code!
1062 __ z_aghi(ix_reg, -2*stride); // Compensate incr from above: zero diff means "all copied".
1063
1064 if (restoreArgs) {
1065 __ z_lcgr(Z_R1, ix_reg); // Prepare ix_reg for copy loop, #bytes expected in Z_R1.
1066 __ z_brnz(doMVCgeneral); // We're not done yet, ix_reg is not zero.
1067
1068 // ARG1, ARG2, and ARG3 were altered by the code above, so restore them building on save_reg.
1069 __ z_slgr(dst_reg, save_reg); // copied #bytes
1070 __ z_slgr(src_reg, dst_reg); // = ARG1 (now restored)
1071 if (log2_size) {
1072 __ z_srag(Z_ARG3, dst_reg, log2_size); // Convert back to #elements to restore ARG3.
1073 } else {
1074 __ z_lgr(Z_ARG3, dst_reg);
1075 }
1076 __ z_lgr(Z_ARG2, save_reg); // ARG2 now restored.
1077
1078 if (branchToEnd) {
1079 __ z_bru(done);
1080 } else {
1081 __ z_br(Z_R14);
1082 }
1083
1084 } else {
1085 if (branchToEnd) {
1086 __ z_brz(done); // CC set by aghi instr.
1087 } else {
1088 __ z_bcr(Assembler::bcondZero, Z_R14); // We're all done if zero.
1089 }
1090
1091 __ z_lcgr(Z_R1, ix_reg); // Prepare ix_reg for copy loop, #bytes expected in Z_R1.
1092 // __ z_bru(doMVCgeneral); // fallthru
1093 }
1094 usedMVCgeneral = true;
1095 BLOCK_COMMENT("} mode MVC loop");
1096 }
1097 // Fallthru to doMVCgeneral
1098
1099 // MVCgeneral: for short, unaligned arrays, after other copy operations
1100
1101 // Somewhat expensive due to use of EX instruction, but simple.
1102 if (usedMVCgeneral) {
1103 BLOCK_COMMENT("mode MVC general {");
1104 __ bind(doMVCgeneral);
1105
1106 __ add2reg(len_reg, -1, Z_R1); // Get #bytes-1 for EXECUTE.
1107 if (VM_Version::has_ExecuteExtensions()) {
1108 __ z_exrl(len_reg, MVC_template); // Execute MVC with variable length.
1109 } else {
1110 __ z_larl(Z_R1, MVC_template); // Get addr of instr template.
1111 __ z_ex(len_reg, 0, Z_R0, Z_R1); // Execute MVC with variable length.
1112 } // penalty: 9 ticks
1113
1114 if (restoreArgs) {
1115 // ARG1, ARG2, and ARG3 were altered by code executed before, so restore them building on save_reg
1116 __ z_slgr(dst_reg, save_reg); // Copied #bytes without the "doMVCgeneral" chunk
1117 __ z_slgr(src_reg, dst_reg); // = ARG1 (now restored), was not advanced for "doMVCgeneral" chunk
1118 __ add2reg_with_index(dst_reg, 1, len_reg, dst_reg); // Len of executed MVC was not accounted for, yet.
1119 if (log2_size) {
1120 __ z_srag(Z_ARG3, dst_reg, log2_size); // Convert back to #elements to restore ARG3
1121 } else {
1122 __ z_lgr(Z_ARG3, dst_reg);
1123 }
1124 __ z_lgr(Z_ARG2, save_reg); // ARG2 now restored.
1125 }
1126
1127 if (usedMVC) {
1128 if (branchToEnd) {
1129 __ z_bru(done);
1130 } else {
1131 __ z_br(Z_R14);
1132 }
1133 } else {
1134 if (!branchToEnd) __ z_br(Z_R14);
1135 }
1136 BLOCK_COMMENT("} mode MVC general");
1137 }
1138 // Fallthru possible if following block not generated.
1139
1140 // MVC: for short, unaligned arrays
1141
1142 // Somewhat expensive due to use of EX instruction, but simple. penalty: 9 ticks.
1143 // Differs from doMVCgeneral in reconstruction of ARG2, ARG3, and ARG4.
1144 if (usedMVC) {
1145 BLOCK_COMMENT("mode MVC {");
1146 __ bind(doMVC);
1147
1148 // get #bytes-1 for EXECUTE
1149 if (log2_size) {
1150 __ add2reg(Z_R1, -1); // Length was scaled into Z_R1.
1151 } else {
1152 __ add2reg(Z_R1, -1, len_reg); // Length was not scaled.
1153 }
1154
1155 if (VM_Version::has_ExecuteExtensions()) {
1156 __ z_exrl(Z_R1, MVC_template); // Execute MVC with variable length.
1157 } else {
1158 __ z_lgr(Z_R0, Z_R5); // Save ARG4, may be unnecessary.
1159 __ z_larl(Z_R5, MVC_template); // Get addr of instr template.
1160 __ z_ex(Z_R1, 0, Z_R0, Z_R5); // Execute MVC with variable length.
1161 __ z_lgr(Z_R5, Z_R0); // Restore ARG4, may be unnecessary.
1162 }
1163
1164 if (!branchToEnd) {
1165 __ z_br(Z_R14);
1166 }
1167 BLOCK_COMMENT("} mode MVC");
1168 }
1169
1170 __ bind(done);
1171
1172 switch (element_size) {
1173 case 1: BLOCK_COMMENT("} ARRAYCOPY DISJOINT byte "); break;
1174 case 2: BLOCK_COMMENT("} ARRAYCOPY DISJOINT short"); break;
1175 case 4: BLOCK_COMMENT("} ARRAYCOPY DISJOINT int "); break;
1176 case 8: BLOCK_COMMENT("} ARRAYCOPY DISJOINT long "); break;
1177 default: BLOCK_COMMENT("} ARRAYCOPY DISJOINT "); break;
1178 }
1179 }
1180 }
1181
1182 // Generate stub for conjoint array copy. If "aligned" is true, the
1183 // "from" and "to" addresses are assumed to be heapword aligned.
1184 //
1185 // Arguments for generated stub:
1186 // from: Z_ARG1
1187 // to: Z_ARG2
1188 // count: Z_ARG3 treated as signed
1189 void generate_conjoint_copy(bool aligned, int element_size, bool branchToEnd) {
1190
1191 // This is the zarch specific stub generator for general array copy tasks.
1192 // It has the following prereqs and features:
1193 //
1194 // - Destructive overlap exists and is handled by reverse copy.
1195 // - Destructive overlap exists if the leftmost byte of the target
1196 // does coincide with any of the source bytes (except the leftmost).
1197 // - Z_R0 and Z_R1 are KILLed by the stub routine (data and stride)
1198 // - Z_ARG1 and Z_ARG2 are USEd but preserved by the stub routine.
1199 // - Z_ARG3 is USED but preserved by the stub routine.
1200 // - Z_ARG4 is used as index register and is thus KILLed.
1201 //
1202 {
1203 Register stride_reg = Z_R1; // Stride & compare value in loop (negative element_size).
1204 Register data_reg = Z_R0; // Holds value of currently processed element.
1205 Register ix_reg = Z_ARG4; // Holds byte index of currently processed element.
1206 Register len_reg = Z_ARG3; // Holds length (in #elements) of arrays.
1207 Register dst_reg = Z_ARG2; // Holds left operand addr.
1208 Register src_reg = Z_ARG1; // Holds right operand addr.
1209
1210 assert(256%element_size == 0, "Element size must be power of 2.");
1211 assert(element_size <= 8, "Can't handle more than DW units.");
1212
1213 switch (element_size) {
1214 case 1: BLOCK_COMMENT("ARRAYCOPY CONJOINT byte {"); break;
1215 case 2: BLOCK_COMMENT("ARRAYCOPY CONJOINT short {"); break;
1216 case 4: BLOCK_COMMENT("ARRAYCOPY CONJOINT int {"); break;
1217 case 8: BLOCK_COMMENT("ARRAYCOPY CONJOINT long {"); break;
1218 default: BLOCK_COMMENT("ARRAYCOPY CONJOINT {"); break;
1219 }
1220
1221 assert_positive_int(len_reg);
1222
1223 if (VM_Version::has_Prefetch()) {
1224 __ z_pfd(0x01, 0, Z_R0, src_reg); // Fetch access.
1225 __ z_pfd(0x02, 0, Z_R0, dst_reg); // Store access.
1226 }
1227
1228 unsigned int log2_size = exact_log2(element_size);
1229 if (log2_size) {
1230 __ z_sllg(ix_reg, len_reg, log2_size);
1231 } else {
1232 __ z_lgr(ix_reg, len_reg);
1233 }
1234
1235 // Optimize reverse copy loop.
1236 // Main loop copies DW units which may be unaligned. Unaligned access adds some penalty ticks.
1237 // Unaligned DW access (neither fetch nor store) is DW-atomic, but should be alignment-atomic.
1238 // Preceding the main loop, some bytes are copied to obtain a DW-multiple remaining length.
1239
1240 Label countLoop1;
1241 Label copyLoop1;
1242 Label skipBY;
1243 Label skipHW;
1244 int stride = -8;
1245
1246 __ load_const_optimized(stride_reg, stride); // Prepare for DW copy loop.
1247
1248 if (element_size == 8) // Nothing to do here.
1249 __ z_bru(countLoop1);
1250 else { // Do not generate dead code.
1251 __ z_tmll(ix_reg, 7); // Check the "odd" bits.
1252 __ z_bre(countLoop1); // There are none, very good!
1253 }
1254
1255 if (log2_size == 0) { // Handle leftover Byte.
1256 __ z_tmll(ix_reg, 1);
1257 __ z_bre(skipBY);
1258 __ z_lb(data_reg, -1, ix_reg, src_reg);
1259 __ z_stcy(data_reg, -1, ix_reg, dst_reg);
1260 __ add2reg(ix_reg, -1); // Decrement delayed to avoid AGI.
1261 __ bind(skipBY);
1262 // fallthru
1263 }
1264 if (log2_size <= 1) { // Handle leftover HW.
1265 __ z_tmll(ix_reg, 2);
1266 __ z_bre(skipHW);
1267 __ z_lhy(data_reg, -2, ix_reg, src_reg);
1268 __ z_sthy(data_reg, -2, ix_reg, dst_reg);
1269 __ add2reg(ix_reg, -2); // Decrement delayed to avoid AGI.
1270 __ bind(skipHW);
1271 __ z_tmll(ix_reg, 4);
1272 __ z_bre(countLoop1);
1273 // fallthru
1274 }
1275 if (log2_size <= 2) { // There are just 4 bytes (left) that need to be copied.
1276 __ z_ly(data_reg, -4, ix_reg, src_reg);
1277 __ z_sty(data_reg, -4, ix_reg, dst_reg);
1278 __ add2reg(ix_reg, -4); // Decrement delayed to avoid AGI.
1279 __ z_bru(countLoop1);
1280 }
1281
1282 // Control can never get to here. Never! Never ever!
1283 __ z_illtrap(0x99);
1284 __ bind(copyLoop1);
1285 __ z_lg(data_reg, 0, ix_reg, src_reg);
1286 __ z_stg(data_reg, 0, ix_reg, dst_reg);
1287 __ bind(countLoop1);
1288 __ z_brxhg(ix_reg, stride_reg, copyLoop1);
1289
1290 if (!branchToEnd)
1291 __ z_br(Z_R14);
1292
1293 switch (element_size) {
1294 case 1: BLOCK_COMMENT("} ARRAYCOPY CONJOINT byte "); break;
1295 case 2: BLOCK_COMMENT("} ARRAYCOPY CONJOINT short"); break;
1296 case 4: BLOCK_COMMENT("} ARRAYCOPY CONJOINT int "); break;
1297 case 8: BLOCK_COMMENT("} ARRAYCOPY CONJOINT long "); break;
1298 default: BLOCK_COMMENT("} ARRAYCOPY CONJOINT "); break;
1299 }
1300 }
1301 }
1302
1303 // Generate stub for disjoint byte copy. If "aligned" is true, the
1304 // "from" and "to" addresses are assumed to be heapword aligned.
1305 address generate_disjoint_byte_copy(bool aligned, const char * name) {
1306 StubCodeMark mark(this, "StubRoutines", name);
1307
1308 // This is the zarch specific stub generator for byte array copy.
1309 // Refer to generate_disjoint_copy for a list of prereqs and features:
1310 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1311 generate_disjoint_copy(aligned, 1, false, false);
1312 return __ addr_at(start_off);
1313 }
1314
1315
1316 address generate_disjoint_short_copy(bool aligned, const char * name) {
1317 StubCodeMark mark(this, "StubRoutines", name);
1318 // This is the zarch specific stub generator for short array copy.
1319 // Refer to generate_disjoint_copy for a list of prereqs and features:
1320 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1321 generate_disjoint_copy(aligned, 2, false, false);
1322 return __ addr_at(start_off);
1323 }
1324
1325
1326 address generate_disjoint_int_copy(bool aligned, const char * name) {
1327 StubCodeMark mark(this, "StubRoutines", name);
1328 // This is the zarch specific stub generator for int array copy.
1329 // Refer to generate_disjoint_copy for a list of prereqs and features:
1330 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1331 generate_disjoint_copy(aligned, 4, false, false);
1332 return __ addr_at(start_off);
1333 }
1334
1335
1336 address generate_disjoint_long_copy(bool aligned, const char * name) {
1337 StubCodeMark mark(this, "StubRoutines", name);
1338 // This is the zarch specific stub generator for long array copy.
1339 // Refer to generate_disjoint_copy for a list of prereqs and features:
1340 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1341 generate_disjoint_copy(aligned, 8, false, false);
1342 return __ addr_at(start_off);
1343 }
1344
1345
1346 address generate_disjoint_oop_copy(bool aligned, const char * name, bool dest_uninitialized) {
1347 StubCodeMark mark(this, "StubRoutines", name);
1348 // This is the zarch specific stub generator for oop array copy.
1349 // Refer to generate_disjoint_copy for a list of prereqs and features.
1350 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1351 unsigned int size = UseCompressedOops ? 4 : 8;
1352
1353 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1354 if (dest_uninitialized) {
1355 decorators |= IS_DEST_UNINITIALIZED;
1356 }
1357 if (aligned) {
1358 decorators |= ARRAYCOPY_ALIGNED;
1359 }
1360
1361 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1362 bs->arraycopy_prologue(_masm, decorators, T_OBJECT, Z_ARG1, Z_ARG2, Z_ARG3);
1363
1364 generate_disjoint_copy(aligned, size, true, true);
1365
1366 bs->arraycopy_epilogue(_masm, decorators, T_OBJECT, Z_ARG2, Z_ARG3, true);
1367
1368 return __ addr_at(start_off);
1369 }
1370
1371
1372 address generate_conjoint_byte_copy(bool aligned, const char * name) {
1373 StubCodeMark mark(this, "StubRoutines", name);
1374 // This is the zarch specific stub generator for overlapping byte array copy.
1375 // Refer to generate_conjoint_copy for a list of prereqs and features:
1376 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1377 address nooverlap_target = aligned ? StubRoutines::arrayof_jbyte_disjoint_arraycopy()
1378 : StubRoutines::jbyte_disjoint_arraycopy();
1379
1380 array_overlap_test(nooverlap_target, 0); // Branch away to nooverlap_target if disjoint.
1381 generate_conjoint_copy(aligned, 1, false);
1382
1383 return __ addr_at(start_off);
1384 }
1385
1386
1387 address generate_conjoint_short_copy(bool aligned, const char * name) {
1388 StubCodeMark mark(this, "StubRoutines", name);
1389 // This is the zarch specific stub generator for overlapping short array copy.
1390 // Refer to generate_conjoint_copy for a list of prereqs and features:
1391 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1392 address nooverlap_target = aligned ? StubRoutines::arrayof_jshort_disjoint_arraycopy()
1393 : StubRoutines::jshort_disjoint_arraycopy();
1394
1395 array_overlap_test(nooverlap_target, 1); // Branch away to nooverlap_target if disjoint.
1396 generate_conjoint_copy(aligned, 2, false);
1397
1398 return __ addr_at(start_off);
1399 }
1400
1401 address generate_conjoint_int_copy(bool aligned, const char * name) {
1402 StubCodeMark mark(this, "StubRoutines", name);
1403 // This is the zarch specific stub generator for overlapping int array copy.
1404 // Refer to generate_conjoint_copy for a list of prereqs and features:
1405
1406 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1407 address nooverlap_target = aligned ? StubRoutines::arrayof_jint_disjoint_arraycopy()
1408 : StubRoutines::jint_disjoint_arraycopy();
1409
1410 array_overlap_test(nooverlap_target, 2); // Branch away to nooverlap_target if disjoint.
1411 generate_conjoint_copy(aligned, 4, false);
1412
1413 return __ addr_at(start_off);
1414 }
1415
1416 address generate_conjoint_long_copy(bool aligned, const char * name) {
1417 StubCodeMark mark(this, "StubRoutines", name);
1418 // This is the zarch specific stub generator for overlapping long array copy.
1419 // Refer to generate_conjoint_copy for a list of prereqs and features:
1420
1421 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1422 address nooverlap_target = aligned ? StubRoutines::arrayof_jlong_disjoint_arraycopy()
1423 : StubRoutines::jlong_disjoint_arraycopy();
1424
1425 array_overlap_test(nooverlap_target, 3); // Branch away to nooverlap_target if disjoint.
1426 generate_conjoint_copy(aligned, 8, false);
1427
1428 return __ addr_at(start_off);
1429 }
1430
1431 address generate_conjoint_oop_copy(bool aligned, const char * name, bool dest_uninitialized) {
1432 StubCodeMark mark(this, "StubRoutines", name);
1433 // This is the zarch specific stub generator for overlapping oop array copy.
1434 // Refer to generate_conjoint_copy for a list of prereqs and features.
1435 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1436 unsigned int size = UseCompressedOops ? 4 : 8;
1437 unsigned int shift = UseCompressedOops ? 2 : 3;
1438
1439 address nooverlap_target = aligned ? StubRoutines::arrayof_oop_disjoint_arraycopy(dest_uninitialized)
1440 : StubRoutines::oop_disjoint_arraycopy(dest_uninitialized);
1441
1442 // Branch to disjoint_copy (if applicable) before pre_barrier to avoid double pre_barrier.
1443 array_overlap_test(nooverlap_target, shift); // Branch away to nooverlap_target if disjoint.
1444
1445 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1446 if (dest_uninitialized) {
1447 decorators |= IS_DEST_UNINITIALIZED;
1448 }
1449 if (aligned) {
1450 decorators |= ARRAYCOPY_ALIGNED;
1451 }
1452
1453 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1454 bs->arraycopy_prologue(_masm, decorators, T_OBJECT, Z_ARG1, Z_ARG2, Z_ARG3);
1455
1456 generate_conjoint_copy(aligned, size, true); // Must preserve ARG2, ARG3.
1457
1458 bs->arraycopy_epilogue(_masm, decorators, T_OBJECT, Z_ARG2, Z_ARG3, true);
1459
1460 return __ addr_at(start_off);
1461 }
1462
1463
1464 void generate_arraycopy_stubs() {
1465
1466 // Note: the disjoint stubs must be generated first, some of
1467 // the conjoint stubs use them.
1468 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy (false, "jbyte_disjoint_arraycopy");
1469 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
1470 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_copy (false, "jint_disjoint_arraycopy");
1471 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_copy (false, "jlong_disjoint_arraycopy");
1472 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_oop_copy (false, "oop_disjoint_arraycopy", false);
1473 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_oop_copy (false, "oop_disjoint_arraycopy_uninit", true);
1474
1475 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy (true, "arrayof_jbyte_disjoint_arraycopy");
1476 StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, "arrayof_jshort_disjoint_arraycopy");
1477 StubRoutines::_arrayof_jint_disjoint_arraycopy = generate_disjoint_int_copy (true, "arrayof_jint_disjoint_arraycopy");
1478 StubRoutines::_arrayof_jlong_disjoint_arraycopy = generate_disjoint_long_copy (true, "arrayof_jlong_disjoint_arraycopy");
1479 StubRoutines::_arrayof_oop_disjoint_arraycopy = generate_disjoint_oop_copy (true, "arrayof_oop_disjoint_arraycopy", false);
1480 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = generate_disjoint_oop_copy (true, "arrayof_oop_disjoint_arraycopy_uninit", true);
1481
1482 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy (false, "jbyte_arraycopy");
1483 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
1484 StubRoutines::_jint_arraycopy = generate_conjoint_int_copy (false, "jint_arraycopy");
1485 StubRoutines::_jlong_arraycopy = generate_conjoint_long_copy (false, "jlong_arraycopy");
1486 StubRoutines::_oop_arraycopy = generate_conjoint_oop_copy (false, "oop_arraycopy", false);
1487 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_oop_copy (false, "oop_arraycopy_uninit", true);
1488
1489 StubRoutines::_arrayof_jbyte_arraycopy = generate_conjoint_byte_copy (true, "arrayof_jbyte_arraycopy");
1490 StubRoutines::_arrayof_jshort_arraycopy = generate_conjoint_short_copy(true, "arrayof_jshort_arraycopy");
1491 StubRoutines::_arrayof_jint_arraycopy = generate_conjoint_int_copy (true, "arrayof_jint_arraycopy");
1492 StubRoutines::_arrayof_jlong_arraycopy = generate_conjoint_long_copy (true, "arrayof_jlong_arraycopy");
1493 StubRoutines::_arrayof_oop_arraycopy = generate_conjoint_oop_copy (true, "arrayof_oop_arraycopy", false);
1494 StubRoutines::_arrayof_oop_arraycopy_uninit = generate_conjoint_oop_copy (true, "arrayof_oop_arraycopy_uninit", true);
1495 }
1496
1497 // Call interface for AES_encryptBlock, AES_decryptBlock stubs.
1498 //
1499 // Z_ARG1 - source data block. Ptr to leftmost byte to be processed.
1500 // Z_ARG2 - destination data block. Ptr to leftmost byte to be stored.
1501 // For in-place encryption/decryption, ARG1 and ARG2 can point
1502 // to the same piece of storage.
1503 // Z_ARG3 - Crypto key address (expanded key). The first n bits of
1504 // the expanded key constitute the original AES-<n> key (see below).
1505 //
1506 // Z_RET - return value. First unprocessed byte offset in src buffer.
1507 //
1508 // Some remarks:
1509 // The crypto key, as passed from the caller to these encryption stubs,
1510 // is a so-called expanded key. It is derived from the original key
1511 // by the Rijndael key schedule, see http://en.wikipedia.org/wiki/Rijndael_key_schedule
1512 // With the expanded key, the cipher/decipher task is decomposed in
1513 // multiple, less complex steps, called rounds. Sun SPARC and Intel
1514 // processors obviously implement support for those less complex steps.
1515 // z/Architecture provides instructions for full cipher/decipher complexity.
1516 // Therefore, we need the original, not the expanded key here.
1517 // Luckily, the first n bits of an AES-<n> expanded key are formed
1518 // by the original key itself. That takes us out of trouble. :-)
1519 // The key length (in bytes) relation is as follows:
1520 // original expanded rounds key bit keylen
1521 // key bytes key bytes length in words
1522 // 16 176 11 128 44
1523 // 24 208 13 192 52
1524 // 32 240 15 256 60
1525 //
1526 // The crypto instructions used in the AES* stubs have some specific register requirements.
1527 // Z_R0 holds the crypto function code. Please refer to the KM/KMC instruction
1528 // description in the "z/Architecture Principles of Operation" manual for details.
1529 // Z_R1 holds the parameter block address. The parameter block contains the cryptographic key
1530 // (KM instruction) and the chaining value (KMC instruction).
1531 // dst must designate an even-numbered register, holding the address of the output message.
1532 // src must designate an even/odd register pair, holding the address/length of the original message
1533
1534 // Helper function which generates code to
1535 // - load the function code in register fCode (== Z_R0).
1536 // - load the data block length (depends on cipher function) into register srclen if requested.
1537 // - is_decipher switches between cipher/decipher function codes
1538 // - set_len requests (if true) loading the data block length in register srclen
1539 void generate_load_AES_fCode(Register keylen, Register fCode, Register srclen, bool is_decipher) {
1540
1541 BLOCK_COMMENT("Set fCode {"); {
1542 Label fCode_set;
1543 int mode = is_decipher ? VM_Version::CipherMode::decipher : VM_Version::CipherMode::cipher;
1544 bool identical_dataBlk_len = (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES192_dataBlk)
1545 && (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES256_dataBlk);
1546 // Expanded key length is 44/52/60 * 4 bytes for AES-128/AES-192/AES-256.
1547 __ z_cghi(keylen, 52); // Check only once at the beginning. keylen and fCode may share the same register.
1548
1549 __ z_lghi(fCode, VM_Version::Cipher::_AES128 + mode);
1550 if (!identical_dataBlk_len) {
1551 __ z_lghi(srclen, VM_Version::Cipher::_AES128_dataBlk);
1552 }
1553 __ z_brl(fCode_set); // keyLen < 52: AES128
1554
1555 __ z_lghi(fCode, VM_Version::Cipher::_AES192 + mode);
1556 if (!identical_dataBlk_len) {
1557 __ z_lghi(srclen, VM_Version::Cipher::_AES192_dataBlk);
1558 }
1559 __ z_bre(fCode_set); // keyLen == 52: AES192
1560
1561 __ z_lghi(fCode, VM_Version::Cipher::_AES256 + mode);
1562 if (!identical_dataBlk_len) {
1563 __ z_lghi(srclen, VM_Version::Cipher::_AES256_dataBlk);
1564 }
1565 // __ z_brh(fCode_set); // keyLen < 52: AES128 // fallthru
1566
1567 __ bind(fCode_set);
1568 if (identical_dataBlk_len) {
1569 __ z_lghi(srclen, VM_Version::Cipher::_AES128_dataBlk);
1570 }
1571 }
1572 BLOCK_COMMENT("} Set fCode");
1573 }
1574
1575 // Push a parameter block for the cipher/decipher instruction on the stack.
1576 // Layout of the additional stack space allocated for AES_cipherBlockChaining:
1577 //
1578 // | |
1579 // +--------+ <-- SP before expansion
1580 // | |
1581 // : : alignment loss (part 2), 0..(AES_parmBlk_align-1) bytes
1582 // | |
1583 // +--------+
1584 // | |
1585 // : : space for parameter block, size VM_Version::Cipher::_AES*_parmBlk_C
1586 // | |
1587 // +--------+ <-- parmBlk, octoword-aligned, start of parameter block
1588 // | |
1589 // : : additional stack space for spills etc., size AES_parmBlk_addspace, DW @ Z_SP not usable!!!
1590 // | |
1591 // +--------+ <-- Z_SP + alignment loss, octoword-aligned
1592 // | |
1593 // : : alignment loss (part 1), 0..(AES_parmBlk_align-1) bytes. DW @ Z_SP not usable!!!
1594 // | |
1595 // +--------+ <-- Z_SP after expansion
1596
1597 void generate_push_Block(int dataBlk_len, int parmBlk_len, int crypto_fCode,
1598 Register parmBlk, Register keylen, Register fCode, Register cv, Register key) {
1599
1600 AES_parmBlk_addspace = AES_parmBlk_align; // Must be multiple of AES_parmblk_align.
1601 // spill space for regs etc., don't use DW @SP!
1602 const int cv_len = dataBlk_len;
1603 const int key_len = parmBlk_len - cv_len;
1604 // This len must be known at JIT compile time. Only then are we able to recalc the SP before resize.
1605 // We buy this knowledge by wasting some (up to AES_parmBlk_align) bytes of stack space.
1606 const int resize_len = cv_len + key_len + AES_parmBlk_align + AES_parmBlk_addspace;
1607
1608 // Use parmBlk as temp reg here to hold the frame pointer.
1609 __ resize_frame(-resize_len, parmBlk, true);
1610
1611 // calculate parmBlk address from updated (resized) SP.
1612 __ add2reg(parmBlk, resize_len - (cv_len + key_len), Z_SP);
1613 __ z_nill(parmBlk, (~(AES_parmBlk_align-1)) & 0xffff); // Align parameter block.
1614
1615 // There is room for stuff in the range [parmBlk-AES_parmBlk_addspace+8, parmBlk).
1616 __ z_stg(keylen, -8, parmBlk); // Spill keylen for later use.
1617
1618 // calculate (SP before resize) from updated SP.
1619 __ add2reg(keylen, resize_len, Z_SP); // keylen holds prev SP for now.
1620 __ z_stg(keylen, -16, parmBlk); // Spill prev SP for easy revert.
1621
1622 __ z_mvc(0, cv_len-1, parmBlk, 0, cv); // Copy cv.
1623 __ z_mvc(cv_len, key_len-1, parmBlk, 0, key); // Copy key.
1624 __ z_lghi(fCode, crypto_fCode);
1625 }
1626
1627 // NOTE:
1628 // Before returning, the stub has to copy the chaining value from
1629 // the parmBlk, where it was updated by the crypto instruction, back
1630 // to the chaining value array the address of which was passed in the cv argument.
1631 // As all the available registers are used and modified by KMC, we need to save
1632 // the key length across the KMC instruction. We do so by spilling it to the stack,
1633 // just preceding the parmBlk (at (parmBlk - 8)).
1634 void generate_push_parmBlk(Register keylen, Register fCode, Register parmBlk, Register key, Register cv, bool is_decipher) {
1635 int mode = is_decipher ? VM_Version::CipherMode::decipher : VM_Version::CipherMode::cipher;
1636 Label parmBlk_128, parmBlk_192, parmBlk_256, parmBlk_set;
1637
1638 BLOCK_COMMENT("push parmBlk {");
1639 // We have just three cipher strengths which translates into three
1640 // possible extended key lengths: 44, 52, and 60 bytes.
1641 // We therefore can compare the actual length against the "middle" length
1642 // and get: lt -> len=44, eq -> len=52, gt -> len=60.
1643 __ z_cghi(keylen, 52);
1644 if (VM_Version::has_Crypto_AES128()) { __ z_brl(parmBlk_128); } // keyLen < 52: AES128
1645 if (VM_Version::has_Crypto_AES192()) { __ z_bre(parmBlk_192); } // keyLen == 52: AES192
1646 if (VM_Version::has_Crypto_AES256()) { __ z_brh(parmBlk_256); } // keyLen > 52: AES256
1647
1648 // Security net: requested AES function not available on this CPU.
1649 // NOTE:
1650 // As of now (March 2015), this safety net is not required. JCE policy files limit the
1651 // cryptographic strength of the keys used to 128 bit. If we have AES hardware support
1652 // at all, we have at least AES-128.
1653 __ stop_static("AES key strength not supported by CPU. Use -XX:-UseAES as remedy.", 0);
1654
1655 if (VM_Version::has_Crypto_AES256()) {
1656 __ bind(parmBlk_256);
1657 generate_push_Block(VM_Version::Cipher::_AES256_dataBlk,
1658 VM_Version::Cipher::_AES256_parmBlk_C,
1659 VM_Version::Cipher::_AES256 + mode,
1660 parmBlk, keylen, fCode, cv, key);
1661 if (VM_Version::has_Crypto_AES128() || VM_Version::has_Crypto_AES192()) {
1662 __ z_bru(parmBlk_set); // Fallthru otherwise.
1663 }
1664 }
1665
1666 if (VM_Version::has_Crypto_AES192()) {
1667 __ bind(parmBlk_192);
1668 generate_push_Block(VM_Version::Cipher::_AES192_dataBlk,
1669 VM_Version::Cipher::_AES192_parmBlk_C,
1670 VM_Version::Cipher::_AES192 + mode,
1671 parmBlk, keylen, fCode, cv, key);
1672 if (VM_Version::has_Crypto_AES128()) {
1673 __ z_bru(parmBlk_set); // Fallthru otherwise.
1674 }
1675 }
1676
1677 if (VM_Version::has_Crypto_AES128()) {
1678 __ bind(parmBlk_128);
1679 generate_push_Block(VM_Version::Cipher::_AES128_dataBlk,
1680 VM_Version::Cipher::_AES128_parmBlk_C,
1681 VM_Version::Cipher::_AES128 + mode,
1682 parmBlk, keylen, fCode, cv, key);
1683 // Fallthru
1684 }
1685
1686 __ bind(parmBlk_set);
1687 BLOCK_COMMENT("} push parmBlk");
1688 }
1689
1690 // Pop a parameter block from the stack. The chaining value portion of the parameter block
1691 // is copied back to the cv array as it is needed for subsequent cipher steps.
1692 // The keylen value as well as the original SP (before resizing) was pushed to the stack
1693 // when pushing the parameter block.
1694 void generate_pop_parmBlk(Register keylen, Register parmBlk, Register key, Register cv) {
1695
1696 BLOCK_COMMENT("pop parmBlk {");
1697 bool identical_dataBlk_len = (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES192_dataBlk) &&
1698 (VM_Version::Cipher::_AES128_dataBlk == VM_Version::Cipher::_AES256_dataBlk);
1699 if (identical_dataBlk_len) {
1700 int cv_len = VM_Version::Cipher::_AES128_dataBlk;
1701 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1702 } else {
1703 int cv_len;
1704 Label parmBlk_128, parmBlk_192, parmBlk_256, parmBlk_set;
1705 __ z_lg(keylen, -8, parmBlk); // restore keylen
1706 __ z_cghi(keylen, 52);
1707 if (VM_Version::has_Crypto_AES256()) __ z_brh(parmBlk_256); // keyLen > 52: AES256
1708 if (VM_Version::has_Crypto_AES192()) __ z_bre(parmBlk_192); // keyLen == 52: AES192
1709 // if (VM_Version::has_Crypto_AES128()) __ z_brl(parmBlk_128); // keyLen < 52: AES128 // fallthru
1710
1711 // Security net: there is no one here. If we would need it, we should have
1712 // fallen into it already when pushing the parameter block.
1713 if (VM_Version::has_Crypto_AES128()) {
1714 __ bind(parmBlk_128);
1715 cv_len = VM_Version::Cipher::_AES128_dataBlk;
1716 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1717 if (VM_Version::has_Crypto_AES192() || VM_Version::has_Crypto_AES256()) {
1718 __ z_bru(parmBlk_set);
1719 }
1720 }
1721
1722 if (VM_Version::has_Crypto_AES192()) {
1723 __ bind(parmBlk_192);
1724 cv_len = VM_Version::Cipher::_AES192_dataBlk;
1725 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1726 if (VM_Version::has_Crypto_AES256()) {
1727 __ z_bru(parmBlk_set);
1728 }
1729 }
1730
1731 if (VM_Version::has_Crypto_AES256()) {
1732 __ bind(parmBlk_256);
1733 cv_len = VM_Version::Cipher::_AES256_dataBlk;
1734 __ z_mvc(0, cv_len-1, cv, 0, parmBlk); // Copy cv.
1735 // __ z_bru(parmBlk_set); // fallthru
1736 }
1737 __ bind(parmBlk_set);
1738 }
1739 __ z_lg(Z_SP, -16, parmBlk); // Revert resize_frame_absolute. Z_SP saved by push_parmBlk.
1740 BLOCK_COMMENT("} pop parmBlk");
1741 }
1742
1743 // Compute AES encrypt/decrypt function.
1744 void generate_AES_cipherBlock(bool is_decipher) {
1745 // Incoming arguments.
1746 Register from = Z_ARG1; // source byte array
1747 Register to = Z_ARG2; // destination byte array
1748 Register key = Z_ARG3; // expanded key array
1749
1750 const Register keylen = Z_R0; // Temporarily (until fCode is set) holds the expanded key array length.
1751
1752 // Register definitions as required by KM instruction.
1753 const Register fCode = Z_R0; // crypto function code
1754 const Register parmBlk = Z_R1; // parameter block address (points to crypto key)
1755 const Register src = Z_ARG1; // Must be even reg (KM requirement).
1756 const Register srclen = Z_ARG2; // Must be odd reg and pair with src. Overwrites destination address.
1757 const Register dst = Z_ARG3; // Must be even reg (KM requirement). Overwrites expanded key address.
1758
1759 // Read key len of expanded key (in 4-byte words).
1760 __ z_lgf(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1761
1762 // Copy arguments to registers as required by crypto instruction.
1763 __ z_lgr(parmBlk, key); // crypto key (in T_INT array).
1764 __ lgr_if_needed(src, from); // Copy src address. Will not emit, src/from are identical.
1765 __ z_lgr(dst, to); // Copy dst address, even register required.
1766
1767 // Construct function code into fCode(Z_R0), data block length into srclen(Z_ARG2).
1768 generate_load_AES_fCode(keylen, fCode, srclen, is_decipher);
1769
1770 __ km(dst, src); // Cipher the message.
1771
1772 __ z_br(Z_R14);
1773 }
1774
1775 // Compute AES encrypt function.
1776 address generate_AES_encryptBlock(const char* name) {
1777 __ align(CodeEntryAlignment);
1778 StubCodeMark mark(this, "StubRoutines", name);
1779 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1780
1781 generate_AES_cipherBlock(false);
1782
1783 return __ addr_at(start_off);
1784 }
1785
1786 // Compute AES decrypt function.
1787 address generate_AES_decryptBlock(const char* name) {
1788 __ align(CodeEntryAlignment);
1789 StubCodeMark mark(this, "StubRoutines", name);
1790 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1791
1792 generate_AES_cipherBlock(true);
1793
1794 return __ addr_at(start_off);
1795 }
1796
1797 // These stubs receive the addresses of the cryptographic key and of the chaining value as two separate
1798 // arguments (registers "key" and "cv", respectively). The KMC instruction, on the other hand, requires
1799 // chaining value and key to be, in this sequence, adjacent in storage. Thus, we need to allocate some
1800 // thread-local working storage. Using heap memory incurs all the hassles of allocating/freeing.
1801 // Stack space, on the contrary, is deallocated automatically when we return from the stub to the caller.
1802 // *** WARNING ***
1803 // Please note that we do not formally allocate stack space, nor do we
1804 // update the stack pointer. Therefore, no function calls are allowed
1805 // and nobody else must use the stack range where the parameter block
1806 // is located.
1807 // We align the parameter block to the next available octoword.
1808 //
1809 // Compute chained AES encrypt function.
1810 void generate_AES_cipherBlockChaining(bool is_decipher) {
1811
1812 Register from = Z_ARG1; // source byte array (clear text)
1813 Register to = Z_ARG2; // destination byte array (ciphered)
1814 Register key = Z_ARG3; // expanded key array.
1815 Register cv = Z_ARG4; // chaining value
1816 const Register msglen = Z_ARG5; // Total length of the msg to be encrypted. Value must be returned
1817 // in Z_RET upon completion of this stub. Is 32-bit integer.
1818
1819 const Register keylen = Z_R0; // Expanded key length, as read from key array. Temp only.
1820 const Register fCode = Z_R0; // crypto function code
1821 const Register parmBlk = Z_R1; // parameter block address (points to crypto key)
1822 const Register src = Z_ARG1; // is Z_R2
1823 const Register srclen = Z_ARG2; // Overwrites destination address.
1824 const Register dst = Z_ARG3; // Overwrites key address.
1825
1826 // Read key len of expanded key (in 4-byte words).
1827 __ z_lgf(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1828
1829 // Construct parm block address in parmBlk (== Z_R1), copy cv and key to parm block.
1830 // Construct function code in fCode (Z_R0).
1831 generate_push_parmBlk(keylen, fCode, parmBlk, key, cv, is_decipher);
1832
1833 // Prepare other registers for instruction.
1834 __ lgr_if_needed(src, from); // Copy src address. Will not emit, src/from are identical.
1835 __ z_lgr(dst, to);
1836 __ z_llgfr(srclen, msglen); // We pass the offsets as ints, not as longs as required.
1837
1838 __ kmc(dst, src); // Cipher the message.
1839
1840 generate_pop_parmBlk(keylen, parmBlk, key, cv);
1841
1842 __ z_llgfr(Z_RET, msglen); // We pass the offsets as ints, not as longs as required.
1843 __ z_br(Z_R14);
1844 }
1845
1846 // Compute chained AES encrypt function.
1847 address generate_cipherBlockChaining_AES_encrypt(const char* name) {
1848 __ align(CodeEntryAlignment);
1849 StubCodeMark mark(this, "StubRoutines", name);
1850 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1851
1852 generate_AES_cipherBlockChaining(false);
1853
1854 return __ addr_at(start_off);
1855 }
1856
1857 // Compute chained AES decrypt function.
1858 address generate_cipherBlockChaining_AES_decrypt(const char* name) {
1859 __ align(CodeEntryAlignment);
1860 StubCodeMark mark(this, "StubRoutines", name);
1861 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
1862
1863 generate_AES_cipherBlockChaining(true);
1864
1865 return __ addr_at(start_off);
1866 }
1867
1868
1869 // *****************************************************************************
1870
1871 // AES CounterMode
1872 // Push a parameter block for the cipher/decipher instruction on the stack.
1873 // Layout of the additional stack space allocated for counterMode_AES_cipherBlock
1874 //
1875 // | |
1876 // +--------+ <-- SP before expansion
1877 // | |
1878 // : : alignment loss (part 2), 0..(AES_parmBlk_align-1) bytes.
1879 // | |
1880 // +--------+ <-- gap = parmBlk + parmBlk_len + ctrArea_len
1881 // | |
1882 // : : byte[] ctr - kmctr expects a counter vector the size of the input vector.
1883 // : : The interface only provides byte[16] iv, the init vector.
1884 // : : The size of this area is a tradeoff between stack space, init effort, and speed.
1885 // | | Each counter is a 128bit int. Vector element [0] is a copy of iv.
1886 // | | Vector element [i] is formed by incrementing element [i-1].
1887 // +--------+ <-- ctr = parmBlk + parmBlk_len
1888 // | |
1889 // : : space for parameter block, size VM_Version::Cipher::_AES*_parmBlk_G
1890 // | |
1891 // +--------+ <-- parmBlk = Z_SP + (alignment loss (part 1+2)) + AES_dataBlk_space + AES_parmBlk_addSpace, octoword-aligned, start of parameter block
1892 // | |
1893 // : : additional stack space for spills etc., min. size AES_parmBlk_addspace, all bytes usable.
1894 // | |
1895 // +--------+ <-- Z_SP + alignment loss (part 1+2) + AES_dataBlk_space, octoword-aligned
1896 // | |
1897 // : : space for one source data block and one dest data block.
1898 // | |
1899 // +--------+ <-- Z_SP + alignment loss (part 1+2), octoword-aligned
1900 // | |
1901 // : : additional alignment loss. Blocks above can't tolerate unusable DW @SP.
1902 // | |
1903 // +--------+ <-- Z_SP + alignment loss (part 1), octoword-aligned
1904 // | |
1905 // : : alignment loss (part 1), 0..(AES_parmBlk_align-1) bytes. DW @ Z_SP holds frame ptr.
1906 // | |
1907 // +--------+ <-- Z_SP after expansion
1908 //
1909 // additional space allocation (per DW):
1910 // spillSpace = parmBlk - AES_parmBlk_addspace
1911 // dataBlocks = spillSpace - AES_dataBlk_space
1912 //
1913 // parmBlk-8 various fields of various lengths
1914 // parmBlk-1: key_len (only one byte is stored at parmBlk-1)
1915 // parmBlk-2: fCode (only one byte is stored at parmBlk-2)
1916 // parmBlk-4: ctrVal_len (as retrieved from iv array), in bytes, as HW
1917 // parmBlk-8: msglen length (in bytes) of crypto msg, as passed in by caller
1918 // return value is calculated from this: rv = msglen - processed.
1919 // parmBlk-16 old_SP (SP before resize)
1920 // parmBlk-24 temp values
1921 // up to and including main loop in generate_counterMode_AES
1922 // - parmBlk-20: remmsg_len remaining msg len (aka unprocessed msg bytes)
1923 // after main loop in generate_counterMode_AES
1924 // - parmBlk-24: spill slot for various address values
1925 //
1926 // parmBlk-40 free spill slot, used for local spills.
1927 // parmBlk-64 ARG2(dst) ptr spill slot
1928 // parmBlk-56 ARG3(crypto key) ptr spill slot
1929 // parmBlk-48 ARG4(icv value) ptr spill slot
1930 //
1931 // parmBlk-72
1932 // parmBlk-80
1933 // parmBlk-88 counter vector current position
1934 // parmBlk-96 reduced msg len (after preLoop processing)
1935 //
1936 // parmBlk-104 Z_R13 spill slot (preLoop only)
1937 // parmBlk-112 Z_R12 spill slot (preLoop only)
1938 // parmBlk-120 Z_R11 spill slot (preLoop only)
1939 // parmBlk-128 Z_R10 spill slot (preLoop only)
1940 //
1941 //
1942 // Layout of the parameter block (instruction KMCTR, function KMCTR-AES*
1943 //
1944 // +--------+ key_len: +16 (AES-128), +24 (AES-192), +32 (AES-256)
1945 // | |
1946 // | | cryptographic key
1947 // | |
1948 // +--------+ <-- parmBlk
1949 //
1950 // On exit:
1951 // Z_SP points to resized frame
1952 // Z_SP before resize available from -16(parmBlk)
1953 // parmBlk points to crypto instruction parameter block
1954 // parameter block is filled with crypto key.
1955 // msglen unchanged, saved for later at -24(parmBlk)
1956 // fCode contains function code for instruction
1957 // key unchanged
1958 //
1959 void generate_counterMode_prepare_Stack(Register parmBlk, Register ctr, Register counter, Register scratch) {
1960
1961 BLOCK_COMMENT("prepare stack counterMode_AESCrypt {");
1962
1963 // save argument registers.
1964 // ARG1(from) is Z_RET as well. Not saved or restored.
1965 // ARG5(msglen) is restored by other means.
1966 __ z_stmg(Z_ARG2, Z_ARG4, argsave_offset, parmBlk);
1967
1968 assert(AES_ctrVec_len > 0, "sanity. We need a counter vector");
1969 __ add2reg(counter, AES_parmBlk_align, parmBlk); // counter array is located behind crypto key. Available range is disp12 only.
1970 __ z_mvc(0, AES_ctrVal_len-1, counter, 0, ctr); // move first copy of iv
1971 for (int j = 1; j < AES_ctrVec_len; j+=j) { // j (and amount of moved data) doubles with every iteration
1972 int offset = j * AES_ctrVal_len;
1973 if (offset <= 256) {
1974 __ z_mvc(offset, offset-1, counter, 0, counter); // move iv
1975 } else {
1976 for (int k = 0; k < offset; k += 256) {
1977 __ z_mvc(offset+k, 255, counter, 0, counter);
1978 }
1979 }
1980 }
1981
1982 Label noCarry, done;
1983 __ z_lg(scratch, Address(ctr, 8)); // get low-order DW of initial counter.
1984 __ z_algfi(scratch, AES_ctrVec_len); // check if we will overflow during init.
1985 __ z_brc(Assembler::bcondLogNoCarry, noCarry); // No, 64-bit increment is sufficient.
1986
1987 for (int j = 1; j < AES_ctrVec_len; j++) { // start with j = 1; no need to add 0 to the first counter value.
1988 int offset = j * AES_ctrVal_len;
1989 generate_increment128(counter, offset, j, scratch); // increment iv by index value
1990 }
1991 __ z_bru(done);
1992
1993 __ bind(noCarry);
1994 for (int j = 1; j < AES_ctrVec_len; j++) { // start with j = 1; no need to add 0 to the first counter value.
1995 int offset = j * AES_ctrVal_len;
1996 generate_increment64(counter, offset, j); // increment iv by index value
1997 }
1998
1999 __ bind(done);
2000
2001 BLOCK_COMMENT("} prepare stack counterMode_AESCrypt");
2002 }
2003
2004
2005 void generate_counterMode_increment_ctrVector(Register parmBlk, Register counter, Register scratch, bool v0_only) {
2006
2007 BLOCK_COMMENT("increment ctrVector counterMode_AESCrypt {");
2008
2009 __ add2reg(counter, AES_parmBlk_align, parmBlk); // ptr to counter array needs to be restored
2010
2011 if (v0_only) {
2012 int offset = 0;
2013 generate_increment128(counter, offset, AES_ctrVec_len, scratch); // increment iv by # vector elements
2014 } else {
2015 int j = 0;
2016 if (VM_Version::has_VectorFacility()) {
2017 bool first_call = true;
2018 for (; j < (AES_ctrVec_len - 3); j+=4) { // increment blocks of 4 iv elements
2019 int offset = j * AES_ctrVal_len;
2020 generate_increment128x4(counter, offset, AES_ctrVec_len, first_call);
2021 first_call = false;
2022 }
2023 }
2024 for (; j < AES_ctrVec_len; j++) {
2025 int offset = j * AES_ctrVal_len;
2026 generate_increment128(counter, offset, AES_ctrVec_len, scratch); // increment iv by # vector elements
2027 }
2028 }
2029
2030 BLOCK_COMMENT("} increment ctrVector counterMode_AESCrypt");
2031 }
2032
2033 // IBM s390 (IBM z/Architecture, to be more exact) uses Big-Endian number representation.
2034 // Therefore, the bits are ordered from most significant to least significant. The address
2035 // of a number in memory points to its lowest location where the most significant bit is stored.
2036 void generate_increment64(Register counter, int offset, int increment) {
2037 __ z_algsi(offset + 8, counter, increment); // increment, no overflow check
2038 }
2039
2040 void generate_increment128(Register counter, int offset, int increment, Register scratch) {
2041 __ clear_reg(scratch); // prepare to add carry to high-order DW
2042 __ z_algsi(offset + 8, counter, increment); // increment low order DW
2043 __ z_alcg(scratch, Address(counter, offset)); // add carry to high-order DW
2044 __ z_stg(scratch, Address(counter, offset)); // store back
2045 }
2046
2047 void generate_increment128(Register counter, int offset, Register increment, Register scratch) {
2048 __ clear_reg(scratch); // prepare to add carry to high-order DW
2049 __ z_alg(increment, Address(counter, offset + 8)); // increment low order DW
2050 __ z_stg(increment, Address(counter, offset + 8)); // store back
2051 __ z_alcg(scratch, Address(counter, offset)); // add carry to high-order DW
2052 __ z_stg(scratch, Address(counter, offset)); // store back
2053 }
2054
2055 // This is the vector variant of increment128, incrementing 4 ctr vector elements per call.
2056 void generate_increment128x4(Register counter, int offset, int increment, bool init) {
2057 VectorRegister Vincr = Z_V16;
2058 VectorRegister Vctr0 = Z_V20;
2059 VectorRegister Vctr1 = Z_V21;
2060 VectorRegister Vctr2 = Z_V22;
2061 VectorRegister Vctr3 = Z_V23;
2062
2063 // Initialize the increment value only once for a series of increments.
2064 // It must be assured that the non-initializing generator calls are
2065 // immediately subsequent. Otherwise, there is no guarantee for Vincr to be unchanged.
2066 if (init) {
2067 __ z_vzero(Vincr); // preset VReg with constant increment
2068 __ z_vleih(Vincr, increment, 7); // rightmost HW has ix = 7
2069 }
2070
2071 __ z_vlm(Vctr0, Vctr3, offset, counter); // get the counter values
2072 __ z_vaq(Vctr0, Vctr0, Vincr); // increment them
2073 __ z_vaq(Vctr1, Vctr1, Vincr);
2074 __ z_vaq(Vctr2, Vctr2, Vincr);
2075 __ z_vaq(Vctr3, Vctr3, Vincr);
2076 __ z_vstm(Vctr0, Vctr3, offset, counter); // store the counter values
2077 }
2078
2079 unsigned int generate_counterMode_push_Block(int dataBlk_len, int parmBlk_len, int crypto_fCode,
2080 Register parmBlk, Register msglen, Register fCode, Register key) {
2081
2082 // space for data blocks (src and dst, one each) for partial block processing)
2083 AES_parmBlk_addspace = AES_stackSpace_incr // spill space (temp data)
2084 + AES_stackSpace_incr // for argument save/restore
2085 + AES_stackSpace_incr*2 // for work reg save/restore
2086 ;
2087 AES_dataBlk_space = roundup(2*dataBlk_len, AES_parmBlk_align);
2088 AES_dataBlk_offset = -(AES_parmBlk_addspace+AES_dataBlk_space);
2089 const int key_len = parmBlk_len; // The length of the unextended key (16, 24, 32)
2090
2091 assert((AES_ctrVal_len == 0) || (AES_ctrVal_len == dataBlk_len), "varying dataBlk_len is not supported.");
2092 AES_ctrVal_len = dataBlk_len; // ctr init value len (in bytes)
2093 AES_ctrArea_len = AES_ctrVec_len * AES_ctrVal_len; // space required on stack for ctr vector
2094
2095 // This len must be known at JIT compile time. Only then are we able to recalc the SP before resize.
2096 // We buy this knowledge by wasting some (up to AES_parmBlk_align) bytes of stack space.
2097 const int resize_len = AES_parmBlk_align // room for alignment of parmBlk
2098 + AES_parmBlk_align // extra room for alignment
2099 + AES_dataBlk_space // one src and one dst data blk
2100 + AES_parmBlk_addspace // spill space for local data
2101 + roundup(parmBlk_len, AES_parmBlk_align) // aligned length of parmBlk
2102 + AES_ctrArea_len // stack space for ctr vector
2103 ;
2104 Register scratch = fCode; // We can use fCode as a scratch register. It's contents on entry
2105 // is irrelevant and it is set at the very end of this code block.
2106
2107 assert(key_len < 256, "excessive crypto key len: %d, limit: 256", key_len);
2108
2109 BLOCK_COMMENT(err_msg("push_Block (%d bytes) counterMode_AESCrypt%d {", resize_len, parmBlk_len*8));
2110
2111 // After the frame is resized, the parmBlk is positioned such
2112 // that it is octoword-aligned. This potentially creates some
2113 // alignment waste in addspace and/or in the gap area.
2114 // After resize_frame, scratch contains the frame pointer.
2115 __ resize_frame(-resize_len, scratch, true);
2116 #ifdef ASSERT
2117 __ clear_mem(Address(Z_SP, (intptr_t)8), resize_len - 8);
2118 #endif
2119
2120 // calculate aligned parmBlk address from updated (resized) SP.
2121 __ add2reg(parmBlk, AES_parmBlk_addspace + AES_dataBlk_space + (2*AES_parmBlk_align-1), Z_SP);
2122 __ z_nill(parmBlk, (~(AES_parmBlk_align-1)) & 0xffff); // Align parameter block.
2123
2124 // There is room to spill stuff in the range [parmBlk-AES_parmBlk_addspace+8, parmBlk).
2125 __ z_mviy(keylen_offset, parmBlk, key_len - 1); // Spill crypto key length for later use. Decrement by one for direct use with xc template.
2126 __ z_mviy(fCode_offset, parmBlk, crypto_fCode); // Crypto function code, will be loaded into Z_R0 later.
2127 __ z_sty(msglen, msglen_offset, parmBlk); // full plaintext/ciphertext len.
2128 __ z_sty(msglen, msglen_red_offset, parmBlk); // save for main loop, may get updated in preLoop.
2129 __ z_sra(msglen, exact_log2(dataBlk_len)); // # full cipher blocks that can be formed from input text.
2130 __ z_sty(msglen, rem_msgblk_offset, parmBlk);
2131
2132 __ add2reg(scratch, resize_len, Z_SP); // calculate (SP before resize) from resized SP.
2133 __ z_stg(scratch, unextSP_offset, parmBlk); // Spill unextended SP for easy revert.
2134 __ z_stmg(Z_R10, Z_R13, regsave_offset, parmBlk); // make some regs available as work registers
2135
2136 // Fill parmBlk with all required data
2137 __ z_mvc(0, key_len-1, parmBlk, 0, key); // Copy key. Need to do it here - key_len is only known here.
2138 BLOCK_COMMENT(err_msg("} push_Block (%d bytes) counterMode_AESCrypt%d", resize_len, parmBlk_len*8));
2139 return resize_len;
2140 }
2141
2142
2143 void generate_counterMode_pop_Block(Register parmBlk, Register msglen, Label& eraser) {
2144 // For added safety, clear the stack area where the crypto key was stored.
2145 Register scratch = msglen;
2146 assert_different_registers(scratch, Z_R0); // can't use Z_R0 for exrl.
2147
2148 // wipe out key on stack
2149 __ z_llgc(scratch, keylen_offset, parmBlk); // get saved (key_len-1) value (we saved just one byte!)
2150 __ z_exrl(scratch, eraser); // template relies on parmBlk still pointing to key on stack
2151
2152 // restore argument registers.
2153 // ARG1(from) is Z_RET as well. Not restored - will hold return value anyway.
2154 // ARG5(msglen) is restored further down.
2155 __ z_lmg(Z_ARG2, Z_ARG4, argsave_offset, parmBlk);
2156
2157 // restore work registers
2158 __ z_lmg(Z_R10, Z_R13, regsave_offset, parmBlk); // make some regs available as work registers
2159
2160 __ z_lgf(msglen, msglen_offset, parmBlk); // Restore msglen, only low order FW is valid
2161 #ifdef ASSERT
2162 {
2163 Label skip2last, skip2done;
2164 // Z_RET (aka Z_R2) can be used as scratch as well. It will be set from msglen before return.
2165 __ z_lgr(Z_RET, Z_SP); // save extended SP
2166 __ z_lg(Z_SP, unextSP_offset, parmBlk); // trim stack back to unextended size
2167 __ z_sgrk(Z_R1, Z_SP, Z_RET);
2168
2169 __ z_cghi(Z_R1, 256);
2170 __ z_brl(skip2last);
2171 __ z_xc(0, 255, Z_RET, 0, Z_RET);
2172 __ z_aghi(Z_RET, 256);
2173 __ z_aghi(Z_R1, -256);
2174
2175 __ z_cghi(Z_R1, 256);
2176 __ z_brl(skip2last);
2177 __ z_xc(0, 255, Z_RET, 0, Z_RET);
2178 __ z_aghi(Z_RET, 256);
2179 __ z_aghi(Z_R1, -256);
2180
2181 __ z_cghi(Z_R1, 256);
2182 __ z_brl(skip2last);
2183 __ z_xc(0, 255, Z_RET, 0, Z_RET);
2184 __ z_aghi(Z_RET, 256);
2185 __ z_aghi(Z_R1, -256);
2186
2187 __ bind(skip2last);
2188 __ z_lgr(Z_R0, Z_RET);
2189 __ z_aghik(Z_RET, Z_R1, -1); // decrement for exrl
2190 __ z_brl(skip2done);
2191 __ z_lgr(parmBlk, Z_R0); // parmBlk == Z_R1, used in eraser template
2192 __ z_exrl(Z_RET, eraser);
2193
2194 __ bind(skip2done);
2195 }
2196 #else
2197 __ z_lg(Z_SP, unextSP_offset, parmBlk); // trim stack back to unextended size
2198 #endif
2199 }
2200
2201
2202 int generate_counterMode_push_parmBlk(Register parmBlk, Register msglen, Register fCode, Register key, bool is_decipher) {
2203 int resize_len = 0;
2204 int mode = is_decipher ? VM_Version::CipherMode::decipher : VM_Version::CipherMode::cipher;
2205 Label parmBlk_128, parmBlk_192, parmBlk_256, parmBlk_set;
2206 Register keylen = fCode; // Expanded key length, as read from key array, Temp only.
2207 // use fCode as scratch; fCode receives its final value later.
2208
2209 // Read key len of expanded key (in 4-byte words).
2210 __ z_lgf(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2211 __ z_cghi(keylen, 52);
2212 if (VM_Version::has_Crypto_AES_CTR256()) { __ z_brh(parmBlk_256); } // keyLen > 52: AES256. Assume: most frequent
2213 if (VM_Version::has_Crypto_AES_CTR128()) { __ z_brl(parmBlk_128); } // keyLen < 52: AES128.
2214 if (VM_Version::has_Crypto_AES_CTR192()) { __ z_bre(parmBlk_192); } // keyLen == 52: AES192. Assume: least frequent
2215
2216 // Safety net: requested AES_CTR function for requested keylen not available on this CPU.
2217 __ stop_static("AES key strength not supported by CPU. Use -XX:-UseAESCTRIntrinsics as remedy.", 0);
2218
2219 if (VM_Version::has_Crypto_AES_CTR128()) {
2220 __ bind(parmBlk_128);
2221 resize_len = generate_counterMode_push_Block(VM_Version::Cipher::_AES128_dataBlk,
2222 VM_Version::Cipher::_AES128_parmBlk_G,
2223 VM_Version::Cipher::_AES128 + mode,
2224 parmBlk, msglen, fCode, key);
2225 if (VM_Version::has_Crypto_AES_CTR256() || VM_Version::has_Crypto_AES_CTR192()) {
2226 __ z_bru(parmBlk_set); // Fallthru otherwise.
2227 }
2228 }
2229
2230 if (VM_Version::has_Crypto_AES_CTR192()) {
2231 __ bind(parmBlk_192);
2232 resize_len = generate_counterMode_push_Block(VM_Version::Cipher::_AES192_dataBlk,
2233 VM_Version::Cipher::_AES192_parmBlk_G,
2234 VM_Version::Cipher::_AES192 + mode,
2235 parmBlk, msglen, fCode, key);
2236 if (VM_Version::has_Crypto_AES_CTR256()) {
2237 __ z_bru(parmBlk_set); // Fallthru otherwise.
2238 }
2239 }
2240
2241 if (VM_Version::has_Crypto_AES_CTR256()) {
2242 __ bind(parmBlk_256);
2243 resize_len = generate_counterMode_push_Block(VM_Version::Cipher::_AES256_dataBlk,
2244 VM_Version::Cipher::_AES256_parmBlk_G,
2245 VM_Version::Cipher::_AES256 + mode,
2246 parmBlk, msglen, fCode, key);
2247 // Fallthru
2248 }
2249
2250 __ bind(parmBlk_set);
2251 return resize_len;
2252 }
2253
2254
2255 void generate_counterMode_pop_parmBlk(Register parmBlk, Register msglen, Label& eraser) {
2256
2257 BLOCK_COMMENT("pop parmBlk counterMode_AESCrypt {");
2258
2259 generate_counterMode_pop_Block(parmBlk, msglen, eraser);
2260
2261 BLOCK_COMMENT("} pop parmBlk counterMode_AESCrypt");
2262 }
2263
2264 // Implementation of counter-mode AES encrypt/decrypt function.
2265 //
2266 void generate_counterMode_AES_impl(bool is_decipher) {
2267
2268 // On entry:
2269 // if there was a previous call to update(), and this previous call did not fully use
2270 // the current encrypted counter, that counter is available at arg6_Offset(Z_SP).
2271 // The index of the first unused bayte in the encrypted counter is available at arg7_Offset(Z_SP).
2272 // The index is in the range [1..AES_ctrVal_len] ([1..16]), where index == 16 indicates a fully
2273 // used previous encrypted counter.
2274 // The unencrypted counter has already been incremented and is ready to be used for the next
2275 // data block, after the unused bytes from the previous call have been consumed.
2276 // The unencrypted counter follows the "increment-after use" principle.
2277
2278 // On exit:
2279 // The index of the first unused byte of the encrypted counter is written back to arg7_Offset(Z_SP).
2280 // A value of AES_ctrVal_len (16) indicates there is no leftover byte.
2281 // If there is at least one leftover byte (1 <= index < AES_ctrVal_len), the encrypted counter value
2282 // is written back to arg6_Offset(Z_SP). If there is no leftover, nothing is written back.
2283 // The unencrypted counter value is written back after having been incremented.
2284
2285 Register from = Z_ARG1; // byte[], source byte array (clear text)
2286 Register to = Z_ARG2; // byte[], destination byte array (ciphered)
2287 Register key = Z_ARG3; // byte[], expanded key array.
2288 Register ctr = Z_ARG4; // byte[], counter byte array.
2289 const Register msglen = Z_ARG5; // int, Total length of the msg to be encrypted. Value must be
2290 // returned in Z_RET upon completion of this stub.
2291 // This is a jint. Negative values are illegal, but technically possible.
2292 // Do not rely on high word. Contents is undefined.
2293 // encCtr = Z_ARG6 - encrypted counter (byte array),
2294 // address passed on stack at _z_abi(remaining_cargs) + 0 * WordSize
2295 // cvIndex = Z_ARG7 - # used (consumed) bytes of encrypted counter,
2296 // passed on stack at _z_abi(remaining_cargs) + 1 * WordSize
2297 // Caution:4-byte value, right-justified in 8-byte stack word
2298
2299 const Register fCode = Z_R0; // crypto function code
2300 const Register parmBlk = Z_R1; // parameter block address (points to crypto key)
2301 const Register src = Z_ARG1; // is Z_R2, forms even/odd pair with srclen
2302 const Register srclen = Z_ARG2; // Overwrites destination address.
2303 const Register dst = Z_ARG3; // Overwrites key address.
2304 const Register counter = Z_ARG5; // Overwrites msglen. Must have counter array in an even register.
2305
2306 Label srcMover, dstMover, fromMover, ctrXOR, dataEraser; // EXRL (execution) templates.
2307 Label CryptoLoop, CryptoLoop_doit, CryptoLoop_end, CryptoLoop_setupAndDoLast, CryptoLoop_ctrVal_inc;
2308 Label allDone, allDone_noInc, popAndExit, Exit;
2309
2310 int arg6_Offset = _z_abi(remaining_cargs) + 0 * HeapWordSize;
2311 int arg7_Offset = _z_abi(remaining_cargs) + 1 * HeapWordSize; // stack slot holds ptr to int value
2312 int oldSP_Offset = 0;
2313
2314 // Is there anything to do at all? Protect against negative len as well.
2315 __ z_ltr(msglen, msglen);
2316 __ z_brnh(Exit);
2317
2318 // Expand stack, load parm block address into parmBlk (== Z_R1), copy crypto key to parm block.
2319 oldSP_Offset = generate_counterMode_push_parmBlk(parmBlk, msglen, fCode, key, is_decipher);
2320 arg6_Offset += oldSP_Offset;
2321 arg7_Offset += oldSP_Offset;
2322
2323 // Check if there is a leftover, partially used encrypted counter from last invocation.
2324 // If so, use those leftover counter bytes first before starting the "normal" encryption.
2325
2326 // We do not have access to the encrypted counter value. It is generated and used only
2327 // internally within the previous kmctr instruction. But, at the end of call to this stub,
2328 // the last encrypted couner is extracted by ciphering a 0x00 byte stream. The result is
2329 // stored at the arg6 location for use with the subsequent call.
2330 //
2331 // The #used bytes of the encrypted counter (from a previous call) is provided via arg7.
2332 // It is used as index into the encrypted counter to access the first byte availabla for ciphering.
2333 // To cipher the input text, we move the number of remaining bytes in the encrypted counter from
2334 // input to output. Then we simply XOR the output bytes with the associated encrypted counter bytes.
2335
2336 Register cvIxAddr = Z_R10; // Address of index into encCtr. Preserved for use @CryptoLoop_end.
2337 __ z_lg(cvIxAddr, arg7_Offset, Z_SP); // arg7: addr of field encCTR_index.
2338
2339 {
2340 Register cvUnused = Z_R11; // # unused bytes of encrypted counter value (= 16 - cvIndex)
2341 Register encCtr = Z_R12; // encrypted counter value, points to first ununsed byte.
2342 Register cvIndex = Z_R13; // # index of first unused byte of encrypted counter value
2343 Label preLoop_end;
2344
2345 // preLoop is necessary only if there is a partially used encrypted counter (encCtr).
2346 // Partially used means cvIndex is in [1, dataBlk_len-1].
2347 // cvIndex == 0: encCtr is set up but not used at all. Should not occur.
2348 // cvIndex == dataBlk_len: encCtr is exhausted, all bytes used.
2349 // Using unsigned compare protects against cases where (cvIndex < 0).
2350 __ z_clfhsi(0, cvIxAddr, AES_ctrVal_len); // check #used bytes in encCtr against ctr len.
2351 __ z_brnl(preLoop_end); // if encCtr is fully used, skip to normal processing.
2352 __ z_ltgf(cvIndex, 0, Z_R0, cvIxAddr); // # used bytes in encCTR.
2353 __ z_brz(preLoop_end); // if encCtr has no used bytes, skip to normal processing.
2354
2355 __ z_lg(encCtr, arg6_Offset, Z_SP); // encrypted counter from last call to update()
2356 __ z_agr(encCtr, cvIndex); // now points to first unused byte
2357
2358 __ add2reg(cvUnused, -AES_ctrVal_len, cvIndex); // calculate #unused bytes in encCtr.
2359 __ z_lcgr(cvUnused, cvUnused); // previous checks ensure cvUnused in range [1, dataBlk_len-1]
2360
2361 __ z_lgf(msglen, msglen_offset, parmBlk); // Restore msglen (jint value)
2362 __ z_cr(cvUnused, msglen); // check if msg can consume all unused encCtr bytes
2363 __ z_locr(cvUnused, msglen, Assembler::bcondHigh); // take the shorter length
2364 __ z_aghi(cvUnused, -1); // decrement # unused bytes by 1 for exrl instruction
2365 // preceding checks ensure cvUnused in range [1, dataBlk_len-1]
2366 __ z_exrl(cvUnused, fromMover);
2367 __ z_exrl(cvUnused, ctrXOR);
2368
2369 __ z_aghi(cvUnused, 1); // revert decrement from above
2370 __ z_agr(cvIndex, cvUnused); // update index into encCtr (first unused byte)
2371 __ z_st(cvIndex, 0, cvIxAddr); // write back arg7, cvIxAddr is still valid
2372
2373 // update pointers and counters to prepare for main loop
2374 __ z_agr(from, cvUnused);
2375 __ z_agr(to, cvUnused);
2376 __ z_sr(msglen, cvUnused); // #bytes not yet processed
2377 __ z_sty(msglen, msglen_red_offset, parmBlk); // save for calculations in main loop
2378 __ z_srak(Z_R0, msglen, exact_log2(AES_ctrVal_len));// # full cipher blocks that can be formed from input text.
2379 __ z_sty(Z_R0, rem_msgblk_offset, parmBlk);
2380
2381 // check remaining msglen. If zero, all msg bytes were processed in preLoop.
2382 __ z_ltr(msglen, msglen);
2383 __ z_brnh(popAndExit);
2384
2385 __ bind(preLoop_end);
2386 }
2387
2388 // Create count vector on stack to accommodate up to AES_ctrVec_len blocks.
2389 generate_counterMode_prepare_Stack(parmBlk, ctr, counter, fCode);
2390
2391 // Prepare other registers for instruction.
2392 __ lgr_if_needed(src, from); // Copy src address. Will not emit, src/from are identical.
2393 __ z_lgr(dst, to);
2394 __ z_llgc(fCode, fCode_offset, Z_R0, parmBlk);
2395
2396 __ bind(CryptoLoop);
2397 __ z_lghi(srclen, AES_ctrArea_len); // preset len (#bytes) for next iteration: max possible.
2398 __ z_asi(rem_msgblk_offset, parmBlk, -AES_ctrVec_len); // decrement #remaining blocks (16 bytes each). Range: [+127..-128]
2399 __ z_brl(CryptoLoop_setupAndDoLast); // Handling the last iteration (using less than max #blocks) out-of-line
2400
2401 __ bind(CryptoLoop_doit);
2402 __ kmctr(dst, counter, src); // Cipher the message.
2403
2404 __ z_lt(srclen, rem_msgblk_offset, Z_R0, parmBlk); // check if this was the last iteration
2405 __ z_brz(CryptoLoop_ctrVal_inc); // == 0: ctrVector fully used. Need to increment the first
2406 // vector element to encrypt remaining unprocessed bytes.
2407 // __ z_brl(CryptoLoop_end); // < 0: this was detected before and handled at CryptoLoop_setupAndDoLast
2408 // > 0: this is the fallthru case, need another iteration
2409
2410 generate_counterMode_increment_ctrVector(parmBlk, counter, srclen, false); // srclen unused here (serves as scratch)
2411 __ z_bru(CryptoLoop);
2412
2413 __ bind(CryptoLoop_end);
2414
2415 // OK, when we arrive here, we have encrypted all of the "from" byte stream
2416 // except for the last few [0..dataBlk_len) bytes. In addition, we know that
2417 // there are no more unused bytes in the previously generated encrypted counter.
2418 // The (unencrypted) counter, however, is ready to use (it was incremented before).
2419
2420 // To encrypt the few remaining bytes, we need to form an extra src and dst
2421 // data block of dataBlk_len each. This is because we can only process full
2422 // blocks but we must not read or write beyond the boundaries of the argument
2423 // arrays. Here is what we do:
2424 // - The ctrVector has at least one unused element. This is ensured by CryptoLoop code.
2425 // - The (first) unused element is pointed at by the counter register.
2426 // - The src data block is filled with the remaining "from" bytes, remainder of block undefined.
2427 // - The single src data block is encrypted into the dst data block.
2428 // - The dst data block is copied into the "to" array, but only the leftmost few bytes
2429 // (as many as were left in the source byte stream).
2430 // - The counter value to be used is pointed at by the counter register.
2431 // - Fortunately, the crypto instruction (kmctr) has updated all related addresses such that
2432 // we know where to continue with "from" and "to" and which counter value to use next.
2433
2434 Register encCtr = Z_R12; // encrypted counter value, points to stub argument.
2435 Register tmpDst = Z_R12; // addr of temp destination (for last partial block encryption)
2436
2437 __ z_lgf(srclen, msglen_red_offset, parmBlk); // plaintext/ciphertext len after potential preLoop processing.
2438 __ z_nilf(srclen, AES_ctrVal_len - 1); // those rightmost bits indicate the unprocessed #bytes
2439 __ z_stg(srclen, localSpill_offset, parmBlk); // save for later reuse
2440 __ z_mvhi(0, cvIxAddr, 16); // write back arg7 (default 16 in case of allDone).
2441 __ z_braz(allDone_noInc); // no unprocessed bytes? Then we are done.
2442 // This also means the last block of data processed was
2443 // a full-sized block (AES_ctrVal_len bytes) which results
2444 // in no leftover encrypted counter bytes.
2445 __ z_st(srclen, 0, cvIxAddr); // This will be the index of the first unused byte in the encrypted counter.
2446 __ z_stg(counter, counter_offset, parmBlk); // save counter location for easy later restore
2447
2448 // calculate address (on stack) for final dst and src blocks.
2449 __ add2reg(tmpDst, AES_dataBlk_offset, parmBlk); // tmp dst (on stack) is right before tmp src
2450
2451 // We have a residue of [1..15] unprocessed bytes, srclen holds the exact number.
2452 // Residue == 0 was checked just above, residue == AES_ctrVal_len would be another
2453 // full-sized block and would have been handled by CryptoLoop.
2454
2455 __ add2reg(srclen, -1); // decrement for exrl
2456 __ z_exrl(srclen, srcMover); // copy remaining bytes of src byte stream
2457 __ load_const_optimized(srclen, AES_ctrVal_len); // kmctr processes only complete blocks
2458 __ add2reg(src, AES_ctrVal_len, tmpDst); // tmp dst is right before tmp src
2459
2460 __ kmctr(tmpDst, counter, src); // Cipher the remaining bytes.
2461
2462 __ add2reg(tmpDst, -AES_ctrVal_len, tmpDst); // restore tmp dst address
2463 __ z_lg(srclen, localSpill_offset, parmBlk); // residual len, saved above
2464 __ add2reg(srclen, -1); // decrement for exrl
2465 __ z_exrl(srclen, dstMover);
2466
2467 // Write back new encrypted counter
2468 __ add2reg(src, AES_dataBlk_offset, parmBlk);
2469 __ clear_mem(Address(src, RegisterOrConstant((intptr_t)0)), AES_ctrVal_len);
2470 __ load_const_optimized(srclen, AES_ctrVal_len); // kmctr processes only complete blocks
2471 __ z_lg(encCtr, arg6_Offset, Z_SP); // write encrypted counter to arg6
2472 __ z_lg(counter, counter_offset, parmBlk); // restore counter
2473 __ kmctr(encCtr, counter, src);
2474
2475 // The last used element of the counter vector contains the latest counter value that was used.
2476 // As described above, the counter value on exit must be the one to be used next.
2477 __ bind(allDone);
2478 __ z_lg(counter, counter_offset, parmBlk); // restore counter
2479 generate_increment128(counter, 0, 1, Z_R0);
2480
2481 __ bind(allDone_noInc);
2482 __ z_mvc(0, AES_ctrVal_len, ctr, 0, counter);
2483
2484 __ bind(popAndExit);
2485 generate_counterMode_pop_parmBlk(parmBlk, msglen, dataEraser);
2486
2487 __ bind(Exit);
2488 __ z_lgfr(Z_RET, msglen);
2489
2490 __ z_br(Z_R14);
2491
2492 //----------------------------
2493 //---< out-of-line code >---
2494 //----------------------------
2495 __ bind(CryptoLoop_setupAndDoLast);
2496 __ z_lgf(srclen, rem_msgblk_offset, parmBlk); // remaining #blocks in memory is < 0
2497 __ z_aghi(srclen, AES_ctrVec_len); // recalculate the actually remaining #blocks
2498 __ z_sllg(srclen, srclen, exact_log2(AES_ctrVal_len)); // convert to #bytes. Counter value is same length as data block
2499 __ kmctr(dst, counter, src); // Cipher the last integral blocks of the message.
2500 __ z_bru(CryptoLoop_end); // There is at least one unused counter vector element.
2501 // no need to increment.
2502
2503 __ bind(CryptoLoop_ctrVal_inc);
2504 generate_counterMode_increment_ctrVector(parmBlk, counter, srclen, true); // srclen unused here (serves as scratch)
2505 __ z_bru(CryptoLoop_end);
2506
2507 //-------------------------------------------
2508 //---< execution templates for preLoop >---
2509 //-------------------------------------------
2510 __ bind(fromMover);
2511 __ z_mvc(0, 0, to, 0, from); // Template instruction to move input data to dst.
2512 __ bind(ctrXOR);
2513 __ z_xc(0, 0, to, 0, encCtr); // Template instruction to XOR input data (now in to) with encrypted counter.
2514
2515 //-------------------------------
2516 //---< execution templates >---
2517 //-------------------------------
2518 __ bind(dataEraser);
2519 __ z_xc(0, 0, parmBlk, 0, parmBlk); // Template instruction to erase crypto key on stack.
2520 __ bind(dstMover);
2521 __ z_mvc(0, 0, dst, 0, tmpDst); // Template instruction to move encrypted reminder from stack to dst.
2522 __ bind(srcMover);
2523 __ z_mvc(AES_ctrVal_len, 0, tmpDst, 0, src); // Template instruction to move reminder of source byte stream to stack.
2524 }
2525
2526
2527 // Create two intrinsic variants, optimized for short and long plaintexts.
2528 void generate_counterMode_AES(bool is_decipher) {
2529
2530 const Register msglen = Z_ARG5; // int, Total length of the msg to be encrypted. Value must be
2531 // returned in Z_RET upon completion of this stub.
2532 const int threshold = 256; // above this length (in bytes), text is considered long.
2533 const int vec_short = threshold>>6; // that many blocks (16 bytes each) per iteration, max 4 loop iterations
2534 const int vec_long = threshold>>2; // that many blocks (16 bytes each) per iteration.
2535
2536 Label AESCTR_short, AESCTR_long;
2537
2538 __ z_chi(msglen, threshold);
2539 __ z_brh(AESCTR_long);
2540
2541 __ bind(AESCTR_short);
2542
2543 BLOCK_COMMENT(err_msg("counterMode_AESCrypt (text len <= %d, block size = %d) {", threshold, vec_short*16));
2544
2545 AES_ctrVec_len = vec_short;
2546 generate_counterMode_AES_impl(false); // control of generated code will not return
2547
2548 BLOCK_COMMENT(err_msg("} counterMode_AESCrypt (text len <= %d, block size = %d)", threshold, vec_short*16));
2549
2550 __ align(32); // Octoword alignment benefits branch targets.
2551
2552 BLOCK_COMMENT(err_msg("counterMode_AESCrypt (text len > %d, block size = %d) {", threshold, vec_long*16));
2553
2554 __ bind(AESCTR_long);
2555 AES_ctrVec_len = vec_long;
2556 generate_counterMode_AES_impl(false); // control of generated code will not return
2557
2558 BLOCK_COMMENT(err_msg("} counterMode_AESCrypt (text len > %d, block size = %d)", threshold, vec_long*16));
2559 }
2560
2561
2562 // Compute AES-CTR crypto function.
2563 // Encrypt or decrypt is selected via parameters. Only one stub is necessary.
2564 address generate_counterMode_AESCrypt(const char* name) {
2565 __ align(CodeEntryAlignment);
2566 StubCodeMark mark(this, "StubRoutines", name);
2567 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2568
2569 generate_counterMode_AES(false);
2570
2571 return __ addr_at(start_off);
2572 }
2573
2574 // *****************************************************************************
2575
2576 // Compute GHASH function.
2577 address generate_ghash_processBlocks() {
2578 __ align(CodeEntryAlignment);
2579 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
2580 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2581
2582 const Register state = Z_ARG1;
2583 const Register subkeyH = Z_ARG2;
2584 const Register data = Z_ARG3; // 1st of even-odd register pair.
2585 const Register blocks = Z_ARG4;
2586 const Register len = blocks; // 2nd of even-odd register pair.
2587
2588 const int param_block_size = 4 * 8;
2589 const int frame_resize = param_block_size + 8; // Extra space for copy of fp.
2590
2591 // Reserve stack space for parameter block (R1).
2592 __ z_lgr(Z_R1, Z_SP);
2593 __ resize_frame(-frame_resize, Z_R0, true);
2594 __ z_aghi(Z_R1, -param_block_size);
2595
2596 // Fill parameter block.
2597 __ z_mvc(Address(Z_R1) , Address(state) , 16);
2598 __ z_mvc(Address(Z_R1, 16), Address(subkeyH), 16);
2599
2600 // R4+5: data pointer + length
2601 __ z_llgfr(len, blocks); // Cast to 64-bit.
2602
2603 // R0: function code
2604 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_GHASH);
2605
2606 // Compute.
2607 __ z_sllg(len, len, 4); // In bytes.
2608 __ kimd(data);
2609
2610 // Copy back result and free parameter block.
2611 __ z_mvc(Address(state), Address(Z_R1), 16);
2612 __ z_xc(Address(Z_R1), param_block_size, Address(Z_R1));
2613 __ z_aghi(Z_SP, frame_resize);
2614
2615 __ z_br(Z_R14);
2616
2617 return __ addr_at(start_off);
2618 }
2619
2620
2621 // Call interface for all SHA* stubs.
2622 //
2623 // Z_ARG1 - source data block. Ptr to leftmost byte to be processed.
2624 // Z_ARG2 - current SHA state. Ptr to state area. This area serves as
2625 // parameter block as required by the crypto instruction.
2626 // Z_ARG3 - current byte offset in source data block.
2627 // Z_ARG4 - last byte offset in source data block.
2628 // (Z_ARG4 - Z_ARG3) gives the #bytes remaining to be processed.
2629 //
2630 // Z_RET - return value. First unprocessed byte offset in src buffer.
2631 //
2632 // A few notes on the call interface:
2633 // - All stubs, whether they are single-block or multi-block, are assumed to
2634 // digest an integer multiple of the data block length of data. All data
2635 // blocks are digested using the intermediate message digest (KIMD) instruction.
2636 // Special end processing, as done by the KLMD instruction, seems to be
2637 // emulated by the calling code.
2638 //
2639 // - Z_ARG1 addresses the first byte of source data. The offset (Z_ARG3) is
2640 // already accounted for.
2641 //
2642 // - The current SHA state (the intermediate message digest value) is contained
2643 // in an area addressed by Z_ARG2. The area size depends on the SHA variant
2644 // and is accessible via the enum VM_Version::MsgDigest::_SHA<n>_parmBlk_I
2645 //
2646 // - The single-block stub is expected to digest exactly one data block, starting
2647 // at the address passed in Z_ARG1.
2648 //
2649 // - The multi-block stub is expected to digest all data blocks which start in
2650 // the offset interval [srcOff(Z_ARG3), srcLimit(Z_ARG4)). The exact difference
2651 // (srcLimit-srcOff), rounded up to the next multiple of the data block length,
2652 // gives the number of blocks to digest. It must be assumed that the calling code
2653 // provides for a large enough source data buffer.
2654 //
2655 // Compute SHA-1 function.
2656 address generate_SHA1_stub(bool multiBlock, const char* name) {
2657 __ align(CodeEntryAlignment);
2658 StubCodeMark mark(this, "StubRoutines", name);
2659 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2660
2661 const Register srcBuff = Z_ARG1; // Points to first block to process (offset already added).
2662 const Register SHAState = Z_ARG2; // Only on entry. Reused soon thereafter for kimd register pairs.
2663 const Register srcOff = Z_ARG3; // int
2664 const Register srcLimit = Z_ARG4; // Only passed in multiBlock case. int
2665
2666 const Register SHAState_local = Z_R1;
2667 const Register SHAState_save = Z_ARG3;
2668 const Register srcBufLen = Z_ARG2; // Destroys state address, must be copied before.
2669 Label useKLMD, rtn;
2670
2671 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_SHA1); // function code
2672 __ z_lgr(SHAState_local, SHAState); // SHAState == parameter block
2673
2674 if (multiBlock) { // Process everything from offset to limit.
2675
2676 // The following description is valid if we get a raw (unpimped) source data buffer,
2677 // spanning the range between [srcOff(Z_ARG3), srcLimit(Z_ARG4)). As detailed above,
2678 // the calling convention for these stubs is different. We leave the description in
2679 // to inform the reader what must be happening hidden in the calling code.
2680 //
2681 // The data block to be processed can have arbitrary length, i.e. its length does not
2682 // need to be an integer multiple of SHA<n>_datablk. Therefore, we need to implement
2683 // two different paths. If the length is an integer multiple, we use KIMD, saving us
2684 // to copy the SHA state back and forth. If the length is odd, we copy the SHA state
2685 // to the stack, execute a KLMD instruction on it and copy the result back to the
2686 // caller's SHA state location.
2687
2688 // Total #srcBuff blocks to process.
2689 if (VM_Version::has_DistinctOpnds()) {
2690 __ z_srk(srcBufLen, srcLimit, srcOff); // exact difference
2691 __ z_ahi(srcBufLen, VM_Version::MsgDigest::_SHA1_dataBlk-1); // round up
2692 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA1_dataBlk-1)) & 0xffff);
2693 __ z_ark(srcLimit, srcOff, srcBufLen); // Srclimit temporarily holds return value.
2694 __ z_llgfr(srcBufLen, srcBufLen); // Cast to 64-bit.
2695 } else {
2696 __ z_lgfr(srcBufLen, srcLimit); // Exact difference. srcLimit passed as int.
2697 __ z_sgfr(srcBufLen, srcOff); // SrcOff passed as int, now properly casted to long.
2698 __ z_aghi(srcBufLen, VM_Version::MsgDigest::_SHA1_dataBlk-1); // round up
2699 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA1_dataBlk-1)) & 0xffff);
2700 __ z_lgr(srcLimit, srcOff); // SrcLimit temporarily holds return value.
2701 __ z_agr(srcLimit, srcBufLen);
2702 }
2703
2704 // Integral #blocks to digest?
2705 // As a result of the calculations above, srcBufLen MUST be an integer
2706 // multiple of _SHA1_dataBlk, or else we are in big trouble.
2707 // We insert an asm_assert into the KLMD case to guard against that.
2708 __ z_tmll(srcBufLen, VM_Version::MsgDigest::_SHA1_dataBlk-1);
2709 __ z_brc(Assembler::bcondNotAllZero, useKLMD);
2710
2711 // Process all full blocks.
2712 __ kimd(srcBuff);
2713
2714 __ z_lgr(Z_RET, srcLimit); // Offset of first unprocessed byte in buffer.
2715 } else { // Process one data block only.
2716 __ load_const_optimized(srcBufLen, (int)VM_Version::MsgDigest::_SHA1_dataBlk); // #srcBuff bytes to process
2717 __ kimd(srcBuff);
2718 __ add2reg(Z_RET, (int)VM_Version::MsgDigest::_SHA1_dataBlk, srcOff); // Offset of first unprocessed byte in buffer. No 32 to 64 bit extension needed.
2719 }
2720
2721 __ bind(rtn);
2722 __ z_br(Z_R14);
2723
2724 if (multiBlock) {
2725 __ bind(useKLMD);
2726
2727 #if 1
2728 // Security net: this stub is believed to be called for full-sized data blocks only
2729 // NOTE: The following code is believed to be correct, but is is not tested.
2730 __ stop_static("SHA128 stub can digest full data blocks only. Use -XX:-UseSHA as remedy.", 0);
2731 #endif
2732 }
2733
2734 return __ addr_at(start_off);
2735 }
2736
2737 // Compute SHA-256 function.
2738 address generate_SHA256_stub(bool multiBlock, const char* name) {
2739 __ align(CodeEntryAlignment);
2740 StubCodeMark mark(this, "StubRoutines", name);
2741 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2742
2743 const Register srcBuff = Z_ARG1;
2744 const Register SHAState = Z_ARG2; // Only on entry. Reused soon thereafter.
2745 const Register SHAState_local = Z_R1;
2746 const Register SHAState_save = Z_ARG3;
2747 const Register srcOff = Z_ARG3;
2748 const Register srcLimit = Z_ARG4;
2749 const Register srcBufLen = Z_ARG2; // Destroys state address, must be copied before.
2750 Label useKLMD, rtn;
2751
2752 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_SHA256); // function code
2753 __ z_lgr(SHAState_local, SHAState); // SHAState == parameter block
2754
2755 if (multiBlock) { // Process everything from offset to limit.
2756 // The following description is valid if we get a raw (unpimped) source data buffer,
2757 // spanning the range between [srcOff(Z_ARG3), srcLimit(Z_ARG4)). As detailed above,
2758 // the calling convention for these stubs is different. We leave the description in
2759 // to inform the reader what must be happening hidden in the calling code.
2760 //
2761 // The data block to be processed can have arbitrary length, i.e. its length does not
2762 // need to be an integer multiple of SHA<n>_datablk. Therefore, we need to implement
2763 // two different paths. If the length is an integer multiple, we use KIMD, saving us
2764 // to copy the SHA state back and forth. If the length is odd, we copy the SHA state
2765 // to the stack, execute a KLMD instruction on it and copy the result back to the
2766 // caller's SHA state location.
2767
2768 // total #srcBuff blocks to process
2769 if (VM_Version::has_DistinctOpnds()) {
2770 __ z_srk(srcBufLen, srcLimit, srcOff); // exact difference
2771 __ z_ahi(srcBufLen, VM_Version::MsgDigest::_SHA256_dataBlk-1); // round up
2772 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA256_dataBlk-1)) & 0xffff);
2773 __ z_ark(srcLimit, srcOff, srcBufLen); // Srclimit temporarily holds return value.
2774 __ z_llgfr(srcBufLen, srcBufLen); // Cast to 64-bit.
2775 } else {
2776 __ z_lgfr(srcBufLen, srcLimit); // exact difference
2777 __ z_sgfr(srcBufLen, srcOff);
2778 __ z_aghi(srcBufLen, VM_Version::MsgDigest::_SHA256_dataBlk-1); // round up
2779 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA256_dataBlk-1)) & 0xffff);
2780 __ z_lgr(srcLimit, srcOff); // Srclimit temporarily holds return value.
2781 __ z_agr(srcLimit, srcBufLen);
2782 }
2783
2784 // Integral #blocks to digest?
2785 // As a result of the calculations above, srcBufLen MUST be an integer
2786 // multiple of _SHA1_dataBlk, or else we are in big trouble.
2787 // We insert an asm_assert into the KLMD case to guard against that.
2788 __ z_tmll(srcBufLen, VM_Version::MsgDigest::_SHA256_dataBlk-1);
2789 __ z_brc(Assembler::bcondNotAllZero, useKLMD);
2790
2791 // Process all full blocks.
2792 __ kimd(srcBuff);
2793
2794 __ z_lgr(Z_RET, srcLimit); // Offset of first unprocessed byte in buffer.
2795 } else { // Process one data block only.
2796 __ load_const_optimized(srcBufLen, (int)VM_Version::MsgDigest::_SHA256_dataBlk); // #srcBuff bytes to process
2797 __ kimd(srcBuff);
2798 __ add2reg(Z_RET, (int)VM_Version::MsgDigest::_SHA256_dataBlk, srcOff); // Offset of first unprocessed byte in buffer.
2799 }
2800
2801 __ bind(rtn);
2802 __ z_br(Z_R14);
2803
2804 if (multiBlock) {
2805 __ bind(useKLMD);
2806 #if 1
2807 // Security net: this stub is believed to be called for full-sized data blocks only.
2808 // NOTE:
2809 // The following code is believed to be correct, but is is not tested.
2810 __ stop_static("SHA256 stub can digest full data blocks only. Use -XX:-UseSHA as remedy.", 0);
2811 #endif
2812 }
2813
2814 return __ addr_at(start_off);
2815 }
2816
2817 // Compute SHA-512 function.
2818 address generate_SHA512_stub(bool multiBlock, const char* name) {
2819 __ align(CodeEntryAlignment);
2820 StubCodeMark mark(this, "StubRoutines", name);
2821 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2822
2823 const Register srcBuff = Z_ARG1;
2824 const Register SHAState = Z_ARG2; // Only on entry. Reused soon thereafter.
2825 const Register SHAState_local = Z_R1;
2826 const Register SHAState_save = Z_ARG3;
2827 const Register srcOff = Z_ARG3;
2828 const Register srcLimit = Z_ARG4;
2829 const Register srcBufLen = Z_ARG2; // Destroys state address, must be copied before.
2830 Label useKLMD, rtn;
2831
2832 __ load_const_optimized(Z_R0, (int)VM_Version::MsgDigest::_SHA512); // function code
2833 __ z_lgr(SHAState_local, SHAState); // SHAState == parameter block
2834
2835 if (multiBlock) { // Process everything from offset to limit.
2836 // The following description is valid if we get a raw (unpimped) source data buffer,
2837 // spanning the range between [srcOff(Z_ARG3), srcLimit(Z_ARG4)). As detailed above,
2838 // the calling convention for these stubs is different. We leave the description in
2839 // to inform the reader what must be happening hidden in the calling code.
2840 //
2841 // The data block to be processed can have arbitrary length, i.e. its length does not
2842 // need to be an integer multiple of SHA<n>_datablk. Therefore, we need to implement
2843 // two different paths. If the length is an integer multiple, we use KIMD, saving us
2844 // to copy the SHA state back and forth. If the length is odd, we copy the SHA state
2845 // to the stack, execute a KLMD instruction on it and copy the result back to the
2846 // caller's SHA state location.
2847
2848 // total #srcBuff blocks to process
2849 if (VM_Version::has_DistinctOpnds()) {
2850 __ z_srk(srcBufLen, srcLimit, srcOff); // exact difference
2851 __ z_ahi(srcBufLen, VM_Version::MsgDigest::_SHA512_dataBlk-1); // round up
2852 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA512_dataBlk-1)) & 0xffff);
2853 __ z_ark(srcLimit, srcOff, srcBufLen); // Srclimit temporarily holds return value.
2854 __ z_llgfr(srcBufLen, srcBufLen); // Cast to 64-bit.
2855 } else {
2856 __ z_lgfr(srcBufLen, srcLimit); // exact difference
2857 __ z_sgfr(srcBufLen, srcOff);
2858 __ z_aghi(srcBufLen, VM_Version::MsgDigest::_SHA512_dataBlk-1); // round up
2859 __ z_nill(srcBufLen, (~(VM_Version::MsgDigest::_SHA512_dataBlk-1)) & 0xffff);
2860 __ z_lgr(srcLimit, srcOff); // Srclimit temporarily holds return value.
2861 __ z_agr(srcLimit, srcBufLen);
2862 }
2863
2864 // integral #blocks to digest?
2865 // As a result of the calculations above, srcBufLen MUST be an integer
2866 // multiple of _SHA1_dataBlk, or else we are in big trouble.
2867 // We insert an asm_assert into the KLMD case to guard against that.
2868 __ z_tmll(srcBufLen, VM_Version::MsgDigest::_SHA512_dataBlk-1);
2869 __ z_brc(Assembler::bcondNotAllZero, useKLMD);
2870
2871 // Process all full blocks.
2872 __ kimd(srcBuff);
2873
2874 __ z_lgr(Z_RET, srcLimit); // Offset of first unprocessed byte in buffer.
2875 } else { // Process one data block only.
2876 __ load_const_optimized(srcBufLen, (int)VM_Version::MsgDigest::_SHA512_dataBlk); // #srcBuff bytes to process
2877 __ kimd(srcBuff);
2878 __ add2reg(Z_RET, (int)VM_Version::MsgDigest::_SHA512_dataBlk, srcOff); // Offset of first unprocessed byte in buffer.
2879 }
2880
2881 __ bind(rtn);
2882 __ z_br(Z_R14);
2883
2884 if (multiBlock) {
2885 __ bind(useKLMD);
2886 #if 1
2887 // Security net: this stub is believed to be called for full-sized data blocks only
2888 // NOTE:
2889 // The following code is believed to be correct, but is is not tested.
2890 __ stop_static("SHA512 stub can digest full data blocks only. Use -XX:-UseSHA as remedy.", 0);
2891 #endif
2892 }
2893
2894 return __ addr_at(start_off);
2895 }
2896
2897
2898 /**
2899 * Arguments:
2900 *
2901 * Inputs:
2902 * Z_ARG1 - int crc
2903 * Z_ARG2 - byte* buf
2904 * Z_ARG3 - int length (of buffer)
2905 *
2906 * Result:
2907 * Z_RET - int crc result
2908 **/
2909 // Compute CRC function (generic, for all polynomials).
2910 void generate_CRC_updateBytes(const char* name, Register table, bool invertCRC) {
2911
2912 // arguments to kernel_crc32:
2913 Register crc = Z_ARG1; // Current checksum, preset by caller or result from previous call, int.
2914 Register data = Z_ARG2; // source byte array
2915 Register dataLen = Z_ARG3; // #bytes to process, int
2916 // Register table = Z_ARG4; // crc table address. Preloaded and passed in by caller.
2917 const Register t0 = Z_R10; // work reg for kernel* emitters
2918 const Register t1 = Z_R11; // work reg for kernel* emitters
2919 const Register t2 = Z_R12; // work reg for kernel* emitters
2920 const Register t3 = Z_R13; // work reg for kernel* emitters
2921
2922
2923 assert_different_registers(crc, data, dataLen, table);
2924
2925 // We pass these values as ints, not as longs as required by C calling convention.
2926 // Crc used as int.
2927 __ z_llgfr(dataLen, dataLen);
2928
2929 __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2930 __ z_stmg(Z_R10, Z_R13, 1*8, Z_SP); // Spill regs 10..11 to make them available as work registers.
2931 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, invertCRC);
2932 __ z_lmg(Z_R10, Z_R13, 1*8, Z_SP); // Spill regs 10..11 back from stack.
2933 __ resize_frame(+(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2934
2935 __ z_llgfr(Z_RET, crc); // Updated crc is function result. No copying required, just zero upper 32 bits.
2936 __ z_br(Z_R14); // Result already in Z_RET == Z_ARG1.
2937 }
2938
2939
2940 // Compute CRC32 function.
2941 address generate_CRC32_updateBytes(const char* name) {
2942 __ align(CodeEntryAlignment);
2943 StubCodeMark mark(this, "StubRoutines", name);
2944 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2945
2946 assert(UseCRC32Intrinsics, "should not generate this stub (%s) with CRC32 intrinsics disabled", name);
2947
2948 BLOCK_COMMENT("CRC32_updateBytes {");
2949 Register table = Z_ARG4; // crc32 table address.
2950 StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
2951
2952 generate_CRC_updateBytes(name, table, true);
2953 BLOCK_COMMENT("} CRC32_updateBytes");
2954
2955 return __ addr_at(start_off);
2956 }
2957
2958
2959 // Compute CRC32C function.
2960 address generate_CRC32C_updateBytes(const char* name) {
2961 __ align(CodeEntryAlignment);
2962 StubCodeMark mark(this, "StubRoutines", name);
2963 unsigned int start_off = __ offset(); // Remember stub start address (is rtn value).
2964
2965 assert(UseCRC32CIntrinsics, "should not generate this stub (%s) with CRC32C intrinsics disabled", name);
2966
2967 BLOCK_COMMENT("CRC32C_updateBytes {");
2968 Register table = Z_ARG4; // crc32c table address.
2969 StubRoutines::zarch::generate_load_crc32c_table_addr(_masm, table);
2970
2971 generate_CRC_updateBytes(name, table, false);
2972 BLOCK_COMMENT("} CRC32C_updateBytes");
2973
2974 return __ addr_at(start_off);
2975 }
2976
2977
2978 // Arguments:
2979 // Z_ARG1 - x address
2980 // Z_ARG2 - x length
2981 // Z_ARG3 - y address
2982 // Z_ARG4 - y length
2983 // Z_ARG5 - z address
2984 // 160[Z_SP] - z length
2985 address generate_multiplyToLen() {
2986 __ align(CodeEntryAlignment);
2987 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
2988
2989 address start = __ pc();
2990
2991 const Register x = Z_ARG1;
2992 const Register xlen = Z_ARG2;
2993 const Register y = Z_ARG3;
2994 const Register ylen = Z_ARG4;
2995 const Register z = Z_ARG5;
2996 // zlen is passed on the stack:
2997 // Address zlen(Z_SP, _z_abi(remaining_cargs));
2998
2999 // Next registers will be saved on stack in multiply_to_len().
3000 const Register tmp1 = Z_tmp_1;
3001 const Register tmp2 = Z_tmp_2;
3002 const Register tmp3 = Z_tmp_3;
3003 const Register tmp4 = Z_tmp_4;
3004 const Register tmp5 = Z_R9;
3005
3006 BLOCK_COMMENT("Entry:");
3007
3008 __ z_llgfr(xlen, xlen);
3009 __ z_llgfr(ylen, ylen);
3010
3011 __ multiply_to_len(x, xlen, y, ylen, z, tmp1, tmp2, tmp3, tmp4, tmp5);
3012
3013 __ z_br(Z_R14); // Return to caller.
3014
3015 return start;
3016 }
3017
3018 address generate_method_entry_barrier() {
3019 __ align(CodeEntryAlignment);
3020 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3021
3022 address start = __ pc();
3023
3024 int nbytes_volatile = (8 + 5) * BytesPerWord;
3025
3026 // VM-Call Prologue
3027 __ save_return_pc();
3028 __ push_frame_abi160(nbytes_volatile);
3029 __ save_volatile_regs(Z_SP, frame::z_abi_160_size, true, false);
3030
3031 // Prep arg for VM call
3032 // Create ptr to stored return_pc in caller frame.
3033 __ z_la(Z_ARG1, _z_abi(return_pc) + frame::z_abi_160_size + nbytes_volatile, Z_R0, Z_SP);
3034
3035 // VM-Call: BarrierSetNMethod::nmethod_stub_entry_barrier(address* return_address_ptr)
3036 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSetNMethod::nmethod_stub_entry_barrier));
3037 __ z_ltr(Z_R0_scratch, Z_RET);
3038
3039 // VM-Call Epilogue
3040 __ restore_volatile_regs(Z_SP, frame::z_abi_160_size, true, false);
3041 __ pop_frame();
3042 __ restore_return_pc();
3043
3044 // Check return val of VM-Call
3045 __ z_bcr(Assembler::bcondZero, Z_R14);
3046
3047 // Pop frame built in prologue.
3048 // Required so wrong_method_stub can deduce caller.
3049 __ pop_frame();
3050 __ restore_return_pc();
3051
3052 // VM-Call indicates deoptimization required
3053 __ load_const_optimized(Z_R1_scratch, SharedRuntime::get_handle_wrong_method_stub());
3054 __ z_br(Z_R1_scratch);
3055
3056 return start;
3057 }
3058
3059 address generate_cont_thaw(bool return_barrier, bool exception) {
3060 if (!Continuations::enabled()) return nullptr;
3061 Unimplemented();
3062 return nullptr;
3063 }
3064
3065 address generate_cont_thaw() {
3066 if (!Continuations::enabled()) return nullptr;
3067 Unimplemented();
3068 return nullptr;
3069 }
3070
3071 address generate_cont_returnBarrier() {
3072 if (!Continuations::enabled()) return nullptr;
3073 Unimplemented();
3074 return nullptr;
3075 }
3076
3077 address generate_cont_returnBarrier_exception() {
3078 if (!Continuations::enabled()) return nullptr;
3079 Unimplemented();
3080 return nullptr;
3081 }
3082
3083 #if INCLUDE_JFR
3084 RuntimeStub* generate_jfr_write_checkpoint() {
3085 if (!Continuations::enabled()) return nullptr;
3086 Unimplemented();
3087 return nullptr;
3088 }
3089
3090 RuntimeStub* generate_jfr_return_lease() {
3091 if (!Continuations::enabled()) return nullptr;
3092 Unimplemented();
3093 return nullptr;
3094 }
3095
3096 #endif // INCLUDE_JFR
3097
3098 // exception handler for upcall stubs
3099 address generate_upcall_stub_exception_handler() {
3100 StubCodeMark mark(this, "StubRoutines", "upcall stub exception handler");
3101 address start = __ pc();
3102
3103 // Native caller has no idea how to handle exceptions,
3104 // so we just crash here. Up to callee to catch exceptions.
3105 __ verify_oop(Z_ARG1);
3106 __ load_const_optimized(Z_R1_scratch, CAST_FROM_FN_PTR(uint64_t, UpcallLinker::handle_uncaught_exception));
3107 __ call_c(Z_R1_scratch);
3108 __ should_not_reach_here();
3109
3110 return start;
3111 }
3112
3113 void generate_initial_stubs() {
3114 // Generates all stubs and initializes the entry points.
3115
3116 // Entry points that exist in all platforms.
3117 // Note: This is code that could be shared among different
3118 // platforms - however the benefit seems to be smaller than the
3119 // disadvantage of having a much more complicated generator
3120 // structure. See also comment in stubRoutines.hpp.
3121 StubRoutines::_forward_exception_entry = generate_forward_exception();
3122
3123 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address);
3124 StubRoutines::_catch_exception_entry = generate_catch_exception();
3125
3126 // Build this early so it's available for the interpreter.
3127 StubRoutines::_throw_StackOverflowError_entry =
3128 generate_throw_exception("StackOverflowError throw_exception",
3129 CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
3130 StubRoutines::_throw_delayed_StackOverflowError_entry =
3131 generate_throw_exception("delayed StackOverflowError throw_exception",
3132 CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError), false);
3133
3134 //----------------------------------------------------------------------
3135 // Entry points that are platform specific.
3136
3137 if (UseCRC32Intrinsics) {
3138 StubRoutines::_crc_table_adr = (address)StubRoutines::zarch::_crc_table;
3139 StubRoutines::_updateBytesCRC32 = generate_CRC32_updateBytes("CRC32_updateBytes");
3140 }
3141
3142 if (UseCRC32CIntrinsics) {
3143 StubRoutines::_crc32c_table_addr = (address)StubRoutines::zarch::_crc32c_table;
3144 StubRoutines::_updateBytesCRC32C = generate_CRC32C_updateBytes("CRC32C_updateBytes");
3145 }
3146
3147 // Comapct string intrinsics: Translate table for string inflate intrinsic. Used by trot instruction.
3148 StubRoutines::zarch::_trot_table_addr = (address)StubRoutines::zarch::_trot_table;
3149 }
3150
3151 void generate_continuation_stubs() {
3152 if (!Continuations::enabled()) return;
3153
3154 // Continuation stubs:
3155 StubRoutines::_cont_thaw = generate_cont_thaw();
3156 StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
3157 StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
3158
3159 JFR_ONLY(generate_jfr_stubs();)
3160 }
3161
3162 #if INCLUDE_JFR
3163 void generate_jfr_stubs() {
3164 StubRoutines::_jfr_write_checkpoint_stub = generate_jfr_write_checkpoint();
3165 StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();
3166 StubRoutines::_jfr_return_lease_stub = generate_jfr_return_lease();
3167 StubRoutines::_jfr_return_lease = StubRoutines::_jfr_return_lease_stub->entry_point();
3168 }
3169 #endif // INCLUDE_JFR
3170
3171 void generate_final_stubs() {
3172 // Generates all stubs and initializes the entry points.
3173
3174 StubRoutines::zarch::_partial_subtype_check = generate_partial_subtype_check();
3175
3176 // These entry points require SharedInfo::stack0 to be set up in non-core builds.
3177 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
3178 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
3179 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call), false);
3180
3181 // Support for verify_oop (must happen after universe_init).
3182 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop_subroutine();
3183
3184 // Arraycopy stubs used by compilers.
3185 generate_arraycopy_stubs();
3186
3187 // nmethod entry barriers for concurrent class unloading
3188 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
3189 if (bs_nm != nullptr) {
3190 StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
3191 }
3192
3193 StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
3194 }
3195
3196 void generate_compiler_stubs() {
3197 #if COMPILER2_OR_JVMCI
3198 // Generate AES intrinsics code.
3199 if (UseAESIntrinsics) {
3200 if (VM_Version::has_Crypto_AES()) {
3201 StubRoutines::_aescrypt_encryptBlock = generate_AES_encryptBlock("AES_encryptBlock");
3202 StubRoutines::_aescrypt_decryptBlock = generate_AES_decryptBlock("AES_decryptBlock");
3203 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_AES_encrypt("AES_encryptBlock_chaining");
3204 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_AES_decrypt("AES_decryptBlock_chaining");
3205 } else {
3206 // In PRODUCT builds, the function pointers will keep their initial (null) value.
3207 // LibraryCallKit::try_to_inline() will return false then, preventing the intrinsic to be called.
3208 assert(VM_Version::has_Crypto_AES(), "Inconsistent settings. Check vm_version_s390.cpp");
3209 }
3210 }
3211
3212 if (UseAESCTRIntrinsics) {
3213 if (VM_Version::has_Crypto_AES_CTR()) {
3214 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt("counterMode_AESCrypt");
3215 } else {
3216 // In PRODUCT builds, the function pointers will keep their initial (null) value.
3217 // LibraryCallKit::try_to_inline() will return false then, preventing the intrinsic to be called.
3218 assert(VM_Version::has_Crypto_AES_CTR(), "Inconsistent settings. Check vm_version_s390.cpp");
3219 }
3220 }
3221
3222 // Generate GHASH intrinsics code
3223 if (UseGHASHIntrinsics) {
3224 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
3225 }
3226
3227 // Generate SHA1/SHA256/SHA512 intrinsics code.
3228 if (UseSHA1Intrinsics) {
3229 StubRoutines::_sha1_implCompress = generate_SHA1_stub(false, "SHA1_singleBlock");
3230 StubRoutines::_sha1_implCompressMB = generate_SHA1_stub(true, "SHA1_multiBlock");
3231 }
3232 if (UseSHA256Intrinsics) {
3233 StubRoutines::_sha256_implCompress = generate_SHA256_stub(false, "SHA256_singleBlock");
3234 StubRoutines::_sha256_implCompressMB = generate_SHA256_stub(true, "SHA256_multiBlock");
3235 }
3236 if (UseSHA512Intrinsics) {
3237 StubRoutines::_sha512_implCompress = generate_SHA512_stub(false, "SHA512_singleBlock");
3238 StubRoutines::_sha512_implCompressMB = generate_SHA512_stub(true, "SHA512_multiBlock");
3239 }
3240
3241 #ifdef COMPILER2
3242 if (UseMultiplyToLenIntrinsic) {
3243 StubRoutines::_multiplyToLen = generate_multiplyToLen();
3244 }
3245 if (UseMontgomeryMultiplyIntrinsic) {
3246 StubRoutines::_montgomeryMultiply
3247 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
3248 }
3249 if (UseMontgomerySquareIntrinsic) {
3250 StubRoutines::_montgomerySquare
3251 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
3252 }
3253 #endif
3254 #endif // COMPILER2_OR_JVMCI
3255 }
3256
3257 public:
3258 StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
3259 switch(kind) {
3260 case Initial_stubs:
3261 generate_initial_stubs();
3262 break;
3263 case Continuation_stubs:
3264 generate_continuation_stubs();
3265 break;
3266 case Compiler_stubs:
3267 generate_compiler_stubs();
3268 break;
3269 case Final_stubs:
3270 generate_final_stubs();
3271 break;
3272 default:
3273 fatal("unexpected stubs kind: %d", kind);
3274 break;
3275 };
3276 }
3277
3278 private:
3279 int _stub_count;
3280 void stub_prolog(StubCodeDesc* cdesc) {
3281 #ifdef ASSERT
3282 // Put extra information in the stub code, to make it more readable.
3283 // Write the high part of the address.
3284 // [RGV] Check if there is a dependency on the size of this prolog.
3285 __ emit_data((intptr_t)cdesc >> 32);
3286 __ emit_data((intptr_t)cdesc);
3287 __ emit_data(++_stub_count);
3288 #endif
3289 align(true);
3290 }
3291
3292 void align(bool at_header = false) {
3293 // z/Architecture cache line size is 256 bytes.
3294 // There is no obvious benefit in aligning stub
3295 // code to cache lines. Use CodeEntryAlignment instead.
3296 const unsigned int icache_line_size = CodeEntryAlignment;
3297 const unsigned int icache_half_line_size = MIN2<unsigned int>(32, CodeEntryAlignment);
3298
3299 if (at_header) {
3300 while ((intptr_t)(__ pc()) % icache_line_size != 0) {
3301 __ z_illtrap();
3302 }
3303 } else {
3304 while ((intptr_t)(__ pc()) % icache_half_line_size != 0) {
3305 __ z_nop();
3306 }
3307 }
3308 }
3309
3310 };
3311
3312 void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
3313 StubGenerator g(code, kind);
3314 }