summaryrefslogtreecommitdiffstats
path: root/arch/nios2/kernel/insnemu.S
blob: 1c6b651e770d34092aa98bca2a2f4da368e2647c (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
/*
 *  Copyright (C) 2003-2013 Altera Corporation
 *  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */


#include <linux/linkage.h>
#include <asm/entry.h>

.set noat
.set nobreak

/*
* Explicitly allow the use of r1 (the assembler temporary register)
* within this code. This register is normally reserved for the use of
* the compiler.
*/

ENTRY(instruction_trap)
	ldw	r1, PT_R1(sp)		// Restore registers
	ldw	r2, PT_R2(sp)
	ldw	r3, PT_R3(sp)
	ldw	r4, PT_R4(sp)
	ldw	r5, PT_R5(sp)
	ldw	r6, PT_R6(sp)
	ldw	r7, PT_R7(sp)
	ldw	r8, PT_R8(sp)
	ldw	r9, PT_R9(sp)
	ldw	r10, PT_R10(sp)
	ldw	r11, PT_R11(sp)
	ldw	r12, PT_R12(sp)
	ldw	r13, PT_R13(sp)
	ldw	r14, PT_R14(sp)
	ldw	r15, PT_R15(sp)
	ldw	ra, PT_RA(sp)
	ldw	fp, PT_FP(sp)
	ldw	gp, PT_GP(sp)
	ldw	et, PT_ESTATUS(sp)
	wrctl	estatus, et
	ldw	ea, PT_EA(sp)
	ldw	et, PT_SP(sp)		/* backup sp in et */

	addi	sp, sp, PT_REGS_SIZE

	/* INSTRUCTION EMULATION
	*  ---------------------
	*
	* Nios II processors generate exceptions for unimplemented instructions.
	* The routines below emulate these instructions.  Depending on the
	* processor core, the only instructions that might need to be emulated
	* are div, divu, mul, muli, mulxss, mulxsu, and mulxuu.
	*
	* The emulations match the instructions, except for the following
	* limitations:
	*
	* 1) The emulation routines do not emulate the use of the exception
	*    temporary register (et) as a source operand because the exception
	*    handler already has modified it.
	*
	* 2) The routines do not emulate the use of the stack pointer (sp) or
	*    the exception return address register (ea) as a destination because
	*    modifying these registers crashes the exception handler or the
	*    interrupted routine.
	*
	* Detailed Design
	* ---------------
	*
	* The emulation routines expect the contents of integer registers r0-r31
	* to be on the stack at addresses sp, 4(sp), 8(sp), ... 124(sp).  The
	* routines retrieve source operands from the stack and modify the
	* destination register's value on the stack prior to the end of the
	* exception handler.  Then all registers except the destination register
	* are restored to their previous values.
	*
	* The instruction that causes the exception is found at address -4(ea).
	* The instruction's OP and OPX fields identify the operation to be
	* performed.
	*
	* One instruction, muli, is an I-type instruction that is identified by
	* an OP field of 0x24.
	*
	* muli   AAAAA,BBBBB,IIIIIIIIIIIIIIII,-0x24-
	*           27    22                6      0    <-- LSB of field
	*
	* The remaining emulated instructions are R-type and have an OP field
	* of 0x3a.  Their OPX fields identify them.
	*
	* R-type AAAAA,BBBBB,CCCCC,XXXXXX,NNNNN,-0x3a-
	*           27    22    17     11     6      0  <-- LSB of field
	*
	*
	* Opcode Encoding.  muli is identified by its OP value.  Then OPX & 0x02
	* is used to differentiate between the division opcodes and the
	* remaining multiplication opcodes.
	*
	* Instruction   OP      OPX    OPX & 0x02
	* -----------   ----    ----   ----------
	* muli          0x24
	* divu          0x3a    0x24         0
	* div           0x3a    0x25         0
	* mul           0x3a    0x27      != 0
	* mulxuu        0x3a    0x07      != 0
	* mulxsu        0x3a    0x17      != 0
	* mulxss        0x3a    0x1f      != 0
	*/


	/*
	* Save everything on the stack to make it easy for the emulation
	* routines to retrieve the source register operands.
	*/

	addi sp, sp, -128
	stw zero, 0(sp)	/* Save zero on stack to avoid special case for r0. */
	stw r1, 4(sp)
	stw r2,  8(sp)
	stw r3, 12(sp)
	stw r4, 16(sp)
	stw r5, 20(sp)
	stw r6, 24(sp)
	stw r7, 28(sp)
	stw r8, 32(sp)
	stw r9, 36(sp)
	stw r10, 40(sp)
	stw r11, 44(sp)
	stw r12, 48(sp)
	stw r13, 52(sp)
	stw r14, 56(sp)
	stw r15, 60(sp)
	stw r16, 64(sp)
	stw r17, 68(sp)
	stw r18, 72(sp)
	stw r19, 76(sp)
	stw r20, 80(sp)
	stw r21, 84(sp)
	stw r22, 88(sp)
	stw r23, 92(sp)
		/* Don't bother to save et.  It's already been changed. */
	rdctl r5, estatus
	stw r5,  100(sp)

	stw gp, 104(sp)
	stw et, 108(sp)	/* et contains previous sp value. */
	stw fp, 112(sp)
	stw ea, 116(sp)
	stw ra, 120(sp)


	/*
	* Split the instruction into its fields.  We need 4*A, 4*B, and 4*C as
	* offsets to the stack pointer for access to the stored register values.
	*/
	ldw r2,-4(ea)	/* r2 = AAAAA,BBBBB,IIIIIIIIIIIIIIII,PPPPPP */
	roli r3, r2, 7	/* r3 = BBB,IIIIIIIIIIIIIIII,PPPPPP,AAAAA,BB */
	roli r4, r3, 3	/* r4 = IIIIIIIIIIIIIIII,PPPPPP,AAAAA,BBBBB */
	roli r5, r4, 2	/* r5 = IIIIIIIIIIIIII,PPPPPP,AAAAA,BBBBB,II */
	srai r4, r4, 16	/* r4 = (sign-extended) IMM16 */
	roli r6, r5, 5	/* r6 = XXXX,NNNNN,PPPPPP,AAAAA,BBBBB,CCCCC,XX */
	andi r2, r2, 0x3f	/* r2 = 00000000000000000000000000,PPPPPP */
	andi r3, r3, 0x7c	/* r3 = 0000000000000000000000000,AAAAA,00 */
	andi r5, r5, 0x7c	/* r5 = 0000000000000000000000000,BBBBB,00 */
	andi r6, r6, 0x7c	/* r6 = 0000000000000000000000000,CCCCC,00 */

	/* Now
	* r2 = OP
	* r3 = 4*A
	* r4 = IMM16 (sign extended)
	* r5 = 4*B
	* r6 = 4*C
	*/

	/*
	* Get the operands.
	*
	* It is necessary to check for muli because it uses an I-type
	* instruction format, while the other instructions are have an R-type
	* format.
	*
	*  Prepare for either multiplication or division loop.
	*  They both loop 32 times.
	*/
	movi r14, 32

	add  r3, r3, sp		/* r3 = address of A-operand. */
	ldw  r3, 0(r3)		/* r3 = A-operand. */
	movi r7, 0x24		/* muli opcode (I-type instruction format) */
	beq r2, r7, mul_immed /* muli doesn't use the B register as a source */

	add  r5, r5, sp		/* r5 = address of B-operand. */
	ldw  r5, 0(r5)		/* r5 = B-operand. */
				/* r4 = SSSSSSSSSSSSSSSS,-----IMM16------ */
				/* IMM16 not needed, align OPX portion */
				/* r4 = SSSSSSSSSSSSSSSS,CCCCC,-OPX--,00000 */
	srli r4, r4, 5		/* r4 = 00000,SSSSSSSSSSSSSSSS,CCCCC,-OPX-- */
	andi r4, r4, 0x3f	/* r4 = 00000000000000000000000000,-OPX-- */

	/* Now
	* r2 = OP
	* r3 = src1
	* r5 = src2
	* r4 = OPX (no longer can be muli)
	* r6 = 4*C
	*/


	/*
	*  Multiply or Divide?
	*/
	andi r7, r4, 0x02	/* For R-type multiply instructions,
				   OPX & 0x02 != 0 */
	bne r7, zero, multiply


	/* DIVISION
	*
	* Divide an unsigned dividend by an unsigned divisor using
	* a shift-and-subtract algorithm.  The example below shows
	* 43 div 7 = 6 for 8-bit integers.  This classic algorithm uses a
	* single register to store both the dividend and the quotient,
	* allowing both values to be shifted with a single instruction.
	*
	*                               remainder dividend:quotient
	*                               --------- -----------------
	*   initialize                   00000000     00101011:
	*   shift                        00000000     0101011:_
	*   remainder >= divisor? no     00000000     0101011:0
	*   shift                        00000000     101011:0_
	*   remainder >= divisor? no     00000000     101011:00
	*   shift                        00000001     01011:00_
	*   remainder >= divisor? no     00000001     01011:000
	*   shift                        00000010     1011:000_
	*   remainder >= divisor? no     00000010     1011:0000
	*   shift                        00000101     011:0000_
	*   remainder >= divisor? no     00000101     011:00000
	*   shift                        00001010     11:00000_
	*   remainder >= divisor? yes    00001010     11:000001
	*       remainder -= divisor   - 00000111
	*                              ----------
	*                                00000011     11:000001
	*   shift                        00000111     1:000001_
	*   remainder >= divisor? yes    00000111     1:0000011
	*       remainder -= divisor   - 00000111
	*                              ----------
	*                                00000000     1:0000011
	*   shift                        00000001     :0000011_
	*   remainder >= divisor? no     00000001     :00000110
	*
	* The quotient is 00000110.
	*/

divide:
	/*
	*  Prepare for division by assuming the result
	*  is unsigned, and storing its "sign" as 0.
	*/
	movi r17, 0


	/* Which division opcode? */
	xori r7, r4, 0x25		/* OPX of div */
	bne r7, zero, unsigned_division


	/*
	*  OPX is div.  Determine and store the sign of the quotient.
	*  Then take the absolute value of both operands.
	*/
	xor r17, r3, r5		/* MSB contains sign of quotient */
	bge r3,zero,dividend_is_nonnegative
	sub r3, zero, r3	/* -r3 */
dividend_is_nonnegative:
	bge r5, zero, divisor_is_nonnegative
	sub r5, zero, r5	/* -r5 */
divisor_is_nonnegative:


unsigned_division:
	/* Initialize the unsigned-division loop. */
	movi r13, 0	/* remainder = 0 */

	/* Now
	* r3 = dividend : quotient
	* r4 = 0x25 for div, 0x24 for divu
	* r5 = divisor
	* r13 = remainder
	* r14 = loop counter (already initialized to 32)
	* r17 = MSB contains sign of quotient
	*/


	/*
	*   for (count = 32; count > 0; --count)
	*   {
	*/
divide_loop:

	/*
	*       Division:
	*
	*       (remainder:dividend:quotient) <<= 1;
	*/
	slli r13, r13, 1
	cmplt r7, r3, zero	/* r7 = MSB of r3 */
	or r13, r13, r7
	slli r3, r3, 1


	/*
	*       if (remainder >= divisor)
	*       {
	*           set LSB of quotient
	*           remainder -= divisor;
	*       }
	*/
	bltu r13, r5, div_skip
	ori r3, r3, 1
	sub r13, r13, r5
div_skip:

	/*
	*   }
	*/
	subi r14, r14, 1
	bne r14, zero, divide_loop


	/* Now
	* r3 = quotient
	* r4 = 0x25 for div, 0x24 for divu
	* r6 = 4*C
	* r17 = MSB contains sign of quotient
	*/


	/*
	*  Conditionally negate signed quotient.  If quotient is unsigned,
	*  the sign already is initialized to 0.
	*/
	bge r17, zero, quotient_is_nonnegative
	sub r3, zero, r3		/* -r3 */
	quotient_is_nonnegative:


	/*
	*  Final quotient is in r3.
	*/
	add r6, r6, sp
	stw r3, 0(r6)	/* write quotient to stack */
	br restore_registers




	/* MULTIPLICATION
	*
	* A "product" is the number that one gets by summing a "multiplicand"
	* several times.  The "multiplier" specifies the number of copies of the
	* multiplicand that are summed.
	*
	* Actual multiplication algorithms don't use repeated addition, however.
	* Shift-and-add algorithms get the same answer as repeated addition, and
	* they are faster.  To compute the lower half of a product (pppp below)
	* one shifts the product left before adding in each of the partial
	* products (a * mmmm) through (d * mmmm).
	*
	* To compute the upper half of a product (PPPP below), one adds in the
	* partial products (d * mmmm) through (a * mmmm), each time following
	* the add by a right shift of the product.
	*
	*     mmmm
	*   * abcd
	*   ------
	*     ####  = d * mmmm
	*    ####   = c * mmmm
	*   ####    = b * mmmm
	*  ####     = a * mmmm
	* --------
	* PPPPpppp
	*
	* The example above shows 4 partial products.  Computing actual Nios II
	* products requires 32 partials.
	*
	* It is possible to compute the result of mulxsu from the result of
	* mulxuu because the only difference between the results of these two
	* opcodes is the value of the partial product associated with the sign
	* bit of rA.
	*
	*   mulxsu = mulxuu - (rA < 0) ? rB : 0;
	*
	* It is possible to compute the result of mulxss from the result of
	* mulxsu because the only difference between the results of these two
	* opcodes is the value of the partial product associated with the sign
	* bit of rB.
	*
	*   mulxss = mulxsu - (rB < 0) ? rA : 0;
	*
	*/

mul_immed:
	/* Opcode is muli.  Change it into mul for remainder of algorithm. */
	mov r6, r5		/* Field B is dest register, not field C. */
	mov r5, r4		/* Field IMM16 is src2, not field B. */
	movi r4, 0x27		/* OPX of mul is 0x27 */

multiply:
	/* Initialize the multiplication loop. */
	movi r9, 0	/* mul_product    = 0 */
	movi r10, 0	/* mulxuu_product = 0 */
	mov r11, r5	/* save original multiplier for mulxsu and mulxss */
	mov r12, r5	/* mulxuu_multiplier (will be shifted) */
	movi r16, 1	/* used to create "rori B,A,1" from "ror B,A,r16" */

	/* Now
	* r3 = multiplicand
	* r5 = mul_multiplier
	* r6 = 4 * dest_register (used later as offset to sp)
	* r7 = temp
	* r9 = mul_product
	* r10 = mulxuu_product
	* r11 = original multiplier
	* r12 = mulxuu_multiplier
	* r14 = loop counter (already initialized)
	* r16 = 1
	*/


	/*
	*   for (count = 32; count > 0; --count)
	*   {
	*/
multiply_loop:

	/*
	*       mul_product <<= 1;
	*       lsb = multiplier & 1;
	*/
	slli r9, r9, 1
	andi r7, r12, 1

	/*
	*       if (lsb == 1)
	*       {
	*           mulxuu_product += multiplicand;
	*       }
	*/
	beq r7, zero, mulx_skip
	add r10, r10, r3
	cmpltu r7, r10, r3 /* Save the carry from the MSB of mulxuu_product. */
	ror r7, r7, r16	/* r7 = 0x80000000 on carry, or else 0x00000000 */
mulx_skip:

	/*
	*       if (MSB of mul_multiplier == 1)
	*       {
	*           mul_product += multiplicand;
	*       }
	*/
	bge r5, zero, mul_skip
	add r9, r9, r3
mul_skip:

	/*
	*       mulxuu_product >>= 1;           logical shift
	*       mul_multiplier <<= 1;           done with MSB
	*       mulx_multiplier >>= 1;          done with LSB
	*/
	srli r10, r10, 1
	or r10, r10, r7		/* OR in the saved carry bit. */
	slli r5, r5, 1
	srli r12, r12, 1


	/*
	*   }
	*/
	subi r14, r14, 1
	bne r14, zero, multiply_loop


	/*
	*  Multiply emulation loop done.
	*/

	/* Now
	* r3 = multiplicand
	* r4 = OPX
	* r6 = 4 * dest_register (used later as offset to sp)
	* r7 = temp
	* r9 = mul_product
	* r10 = mulxuu_product
	* r11 = original multiplier
	*/


	/* Calculate address for result from 4 * dest_register */
	add r6, r6, sp


	/*
	* Select/compute the result based on OPX.
	*/


	/* OPX == mul?  Then store. */
	xori r7, r4, 0x27
	beq r7, zero, store_product

	/* It's one of the mulx.. opcodes.  Move over the result. */
	mov r9, r10

	/* OPX == mulxuu?  Then store. */
	xori r7, r4, 0x07
	beq r7, zero, store_product

	/* Compute mulxsu
	 *
	 * mulxsu = mulxuu - (rA < 0) ? rB : 0;
	 */
	bge r3, zero, mulxsu_skip
	sub r9, r9, r11
mulxsu_skip:

	/* OPX == mulxsu?  Then store. */
	xori r7, r4, 0x17
	beq r7, zero, store_product

	/* Compute mulxss
	 *
	 * mulxss = mulxsu - (rB < 0) ? rA : 0;
	 */
	bge r11,zero,mulxss_skip
	sub r9, r9, r3
mulxss_skip:
	/* At this point, assume that OPX is mulxss, so store*/


store_product:
	stw r9, 0(r6)


restore_registers:
			/* No need to restore r0. */
	ldw r5, 100(sp)
	wrctl estatus, r5

	ldw r1, 4(sp)
	ldw r2, 8(sp)
	ldw r3, 12(sp)
	ldw r4, 16(sp)
	ldw r5, 20(sp)
	ldw r6, 24(sp)
	ldw r7, 28(sp)
	ldw r8, 32(sp)
	ldw r9, 36(sp)
	ldw r10, 40(sp)
	ldw r11, 44(sp)
	ldw r12, 48(sp)
	ldw r13, 52(sp)
	ldw r14, 56(sp)
	ldw r15, 60(sp)
	ldw r16, 64(sp)
	ldw r17, 68(sp)
	ldw r18, 72(sp)
	ldw r19, 76(sp)
	ldw r20, 80(sp)
	ldw r21, 84(sp)
	ldw r22, 88(sp)
	ldw r23, 92(sp)
			/* Does not need to restore et */
	ldw gp, 104(sp)

	ldw fp, 112(sp)
	ldw ea, 116(sp)
	ldw ra, 120(sp)
	ldw sp, 108(sp)	/* last restore sp */
	eret

.set at
.set break