1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux/PA-RISC Project (http://www.parisc-linux.org/)
5 * Floating-point emulation code
12 * @(#) pa/fp/fpudispatch.c $Revision: 1.1 $
15 * <<please update with a synopsis of the functionality provided by this file>>
17 * External Interfaces:
18 * <<the following list was autogenerated, please review>>
19 * emfpudispatch(ir, dummy1, dummy2, fpregs)
20 * fpudispatch(ir, excp_code, holder, fpregs)
22 * Internal Interfaces:
23 * <<the following list was autogenerated, please review>>
24 * static u_int decode_06(u_int, u_int *)
25 * static u_int decode_0c(u_int, u_int, u_int, u_int *)
26 * static u_int decode_0e(u_int, u_int, u_int, u_int *)
27 * static u_int decode_26(u_int, u_int *)
28 * static u_int decode_2e(u_int, u_int *)
29 * static void update_status_cbit(u_int *, u_int, u_int, u_int)
32 * <<please update with a overview of the operation of this file>>
40 #include <linux/bug.h>
41 #include <linux/kernel.h>
42 #include <asm/processor.h>
43 /* #include <sys/debug.h> */
44 /* #include <machine/sys/mdep_private.h> */
46 #define COPR_INST 0x30000000
49 * definition of extru macro. If pos and len are constants, the compiler
50 * will generate an extru instruction when optimized
52 #define extru(r,pos,len) (((r) >> (31-(pos))) & (( 1 << (len)) - 1))
53 /* definitions of bit field locations in the instruction */
59 #define fpclass1subpos 16
65 * the following are the extra bits for the 0E major op
73 * the following are for the multi-ops
82 * the following are for the fused FP instructions
92 #define fpfusedsubop 26
96 * offset to constant zero in the FP emulation registers
98 #define fpzeroreg (32*sizeof(double)/sizeof(u_int))
101 * extract the major opcode from the instruction
103 #define get_major(op) extru(op,fpmajorpos,6)
105 * extract the two bit class field from the FP instruction. The class is at bit
108 #define get_class(op) extru(op,fpclasspos,2)
110 * extract the 3 bit subop field. For all but class 1 instructions, it is
111 * located at bit positions 16-18
113 #define get_subop(op) extru(op,fpsubpos,3)
115 * extract the 2 or 3 bit subop field from class 1 instructions. It is located
116 * at bit positions 15-16 (PA1.1) or 14-16 (PA2.0)
118 #define get_subop1_PA1_1(op) extru(op,fpclass1subpos,2) /* PA89 (1.1) fmt */
119 #define get_subop1_PA2_0(op) extru(op,fpclass1subpos,3) /* PA 2.0 fmt */
121 /* definitions of unimplemented exceptions */
122 #define MAJOR_0C_EXCP 0x09
123 #define MAJOR_0E_EXCP 0x0b
124 #define MAJOR_06_EXCP 0x03
125 #define MAJOR_26_EXCP 0x23
126 #define MAJOR_2E_EXCP 0x2b
127 #define PA83_UNIMP_EXCP 0x01
130 * Special Defines for TIMEX specific code
133 #define FPU_TYPE_FLAG_POS (EM_FPU_TYPE_OFFSET>>2)
134 #define TIMEX_ROLEX_FPU_MASK (TIMEX_EXTEN_FLAG|ROLEX_EXTEN_FLAG)
137 * Static function definitions
140 #if defined(_PROTOTYPES) || defined(_lint)
141 static u_int decode_0c(u_int, u_int, u_int, u_int *);
142 static u_int decode_0e(u_int, u_int, u_int, u_int *);
143 static u_int decode_06(u_int, u_int *);
144 static u_int decode_26(u_int, u_int *);
145 static u_int decode_2e(u_int, u_int *);
146 static void update_status_cbit(u_int *, u_int, u_int, u_int);
147 #else /* !_PROTOTYPES&&!_lint */
148 static u_int decode_0c();
149 static u_int decode_0e();
150 static u_int decode_06();
151 static u_int decode_26();
152 static u_int decode_2e();
153 static void update_status_cbit();
154 #endif /* _PROTOTYPES&&!_lint */
158 static void parisc_linux_get_fpu_type(u_int fpregs[])
160 /* on pa-linux the fpu type is not filled in by the
161 * caller; it is constructed here
163 if (boot_cpu_data.cpu_type == pcxs)
164 fpregs[FPU_TYPE_FLAG_POS] = TIMEX_EXTEN_FLAG;
165 else if (boot_cpu_data.cpu_type == pcxt ||
166 boot_cpu_data.cpu_type == pcxt_)
167 fpregs[FPU_TYPE_FLAG_POS] = ROLEX_EXTEN_FLAG;
168 else if (boot_cpu_data.cpu_type >= pcxu)
169 fpregs[FPU_TYPE_FLAG_POS] = PA2_0_FPU_FLAG;
173 * this routine will decode the excepting floating point instruction and
174 * call the appropriate emulation routine.
175 * It is called by decode_fpu with the following parameters:
176 * fpudispatch(current_ir, unimplemented_code, 0, &Fpu_register)
177 * where current_ir is the instruction to be emulated,
178 * unimplemented_code is the exception_code that the hardware generated
179 * and &Fpu_register is the address of emulated FP reg 0.
182 fpudispatch(u_int ir, u_int excp_code, u_int holder, u_int fpregs[])
185 u_int fpu_type_flags;
187 /* All FP emulation code assumes that ints are 4-bytes in length */
188 VASSERT(sizeof(int) == 4);
190 parisc_linux_get_fpu_type(fpregs);
192 fpu_type_flags=fpregs[FPU_TYPE_FLAG_POS]; /* get fpu type flags */
194 class = get_class(ir);
196 if (fpu_type_flags & PA2_0_FPU_FLAG)
197 subop = get_subop1_PA2_0(ir);
199 subop = get_subop1_PA1_1(ir);
202 subop = get_subop(ir);
204 if (FPUDEBUG) printk("class %d subop %d\n", class, subop);
208 case PA83_UNIMP_EXCP:
209 return(decode_0c(ir,class,subop,fpregs));
211 return(decode_0e(ir,class,subop,fpregs));
213 return(decode_06(ir,fpregs));
215 return(decode_26(ir,fpregs));
217 return(decode_2e(ir,fpregs));
219 /* "crashme Night Gallery painting nr 2. (asm_crash.s).
220 * This was fixed for multi-user kernels, but
221 * workstation kernels had a panic here. This allowed
222 * any arbitrary user to panic the kernel by executing
223 * setting the FP exception registers to strange values
224 * and generating an emulation trap. The emulation and
225 * exception code must never be able to panic the
228 return(UNIMPLEMENTEDEXCEPTION);
233 * this routine is called by $emulation_trap to emulate a coprocessor
234 * instruction if one doesn't exist
237 emfpudispatch(u_int ir, u_int dummy1, u_int dummy2, u_int fpregs[])
239 u_int class, subop, major;
240 u_int fpu_type_flags;
242 /* All FP emulation code assumes that ints are 4-bytes in length */
243 VASSERT(sizeof(int) == 4);
245 fpu_type_flags=fpregs[FPU_TYPE_FLAG_POS]; /* get fpu type flags */
247 major = get_major(ir);
248 class = get_class(ir);
250 if (fpu_type_flags & PA2_0_FPU_FLAG)
251 subop = get_subop1_PA2_0(ir);
253 subop = get_subop1_PA1_1(ir);
256 subop = get_subop(ir);
259 return(decode_0c(ir,class,subop,fpregs));
261 return(decode_0e(ir,class,subop,fpregs));
263 return(decode_06(ir,fpregs));
265 return(decode_26(ir,fpregs));
267 return(decode_2e(ir,fpregs));
269 return(PA83_UNIMP_EXCP);
275 decode_0c(u_int ir, u_int class, u_int subop, u_int fpregs[])
277 u_int r1,r2,t; /* operand register offsets */
278 u_int fmt; /* also sf for class 1 conversions */
279 u_int df; /* for class 1 conversions */
281 u_int retval, local_status;
282 u_int fpu_type_flags;
284 if (ir == COPR_INST) {
285 fpregs[0] = EMULATION_VERSION << 11;
288 status = &fpregs[0]; /* fp status register */
289 local_status = fpregs[0]; /* and local copy */
290 r1 = extru(ir,fpr1pos,5) * sizeof(double)/sizeof(u_int);
291 if (r1 == 0) /* map fr0 source to constant zero */
293 t = extru(ir,fptpos,5) * sizeof(double)/sizeof(u_int);
294 if (t == 0 && class != 2) /* don't allow fr0 as a dest */
295 return(MAJOR_0C_EXCP);
296 fmt = extru(ir,fpfmtpos,2); /* get fmt completer */
301 case 0: /* COPR 0,0 emulated above*/
303 return(MAJOR_0C_EXCP);
306 case 2: /* illegal */
307 return(MAJOR_0C_EXCP);
309 t &= ~3; /* force to even reg #s */
311 fpregs[t+3] = fpregs[r1+3];
312 fpregs[t+2] = fpregs[r1+2];
314 fpregs[t+1] = fpregs[r1+1];
316 fpregs[t] = fpregs[r1];
321 case 2: /* illegal */
322 return(MAJOR_0C_EXCP);
324 t &= ~3; /* force to even reg #s */
326 fpregs[t+3] = fpregs[r1+3];
327 fpregs[t+2] = fpregs[r1+2];
329 fpregs[t+1] = fpregs[r1+1];
331 /* copy and clear sign bit */
332 fpregs[t] = fpregs[r1] & 0x7fffffff;
337 case 2: /* illegal */
338 return(MAJOR_0C_EXCP);
340 t &= ~3; /* force to even reg #s */
342 fpregs[t+3] = fpregs[r1+3];
343 fpregs[t+2] = fpregs[r1+2];
345 fpregs[t+1] = fpregs[r1+1];
347 /* copy and invert sign bit */
348 fpregs[t] = fpregs[r1] ^ 0x80000000;
351 case 7: /* FNEGABS */
353 case 2: /* illegal */
354 return(MAJOR_0C_EXCP);
356 t &= ~3; /* force to even reg #s */
358 fpregs[t+3] = fpregs[r1+3];
359 fpregs[t+2] = fpregs[r1+2];
361 fpregs[t+1] = fpregs[r1+1];
363 /* copy and set sign bit */
364 fpregs[t] = fpregs[r1] | 0x80000000;
370 return(sgl_fsqrt(&fpregs[r1],0,
373 return(dbl_fsqrt(&fpregs[r1],0,
376 case 3: /* quad not implemented */
377 return(MAJOR_0C_EXCP);
382 return(sgl_frnd(&fpregs[r1],0,
385 return(dbl_frnd(&fpregs[r1],0,
388 case 3: /* quad not implemented */
389 return(MAJOR_0C_EXCP);
391 } /* end of switch (subop) */
393 case 1: /* class 1 */
394 df = extru(ir,fpdfpos,2); /* get dest format */
395 if ((df & 2) || (fmt & 2)) {
397 * fmt's 2 and 3 are illegal of not implemented
400 return(MAJOR_0C_EXCP);
403 * encode source and dest formats into 2 bits.
404 * high bit is source, low bit is dest.
405 * bit = 1 --> double precision
407 fmt = (fmt << 1) | df;
411 case 0: /* sgl/sgl */
412 return(MAJOR_0C_EXCP);
413 case 1: /* sgl/dbl */
414 return(sgl_to_dbl_fcnvff(&fpregs[r1],0,
416 case 2: /* dbl/sgl */
417 return(dbl_to_sgl_fcnvff(&fpregs[r1],0,
419 case 3: /* dbl/dbl */
420 return(MAJOR_0C_EXCP);
424 case 0: /* sgl/sgl */
425 return(sgl_to_sgl_fcnvxf(&fpregs[r1],0,
427 case 1: /* sgl/dbl */
428 return(sgl_to_dbl_fcnvxf(&fpregs[r1],0,
430 case 2: /* dbl/sgl */
431 return(dbl_to_sgl_fcnvxf(&fpregs[r1],0,
433 case 3: /* dbl/dbl */
434 return(dbl_to_dbl_fcnvxf(&fpregs[r1],0,
439 case 0: /* sgl/sgl */
440 return(sgl_to_sgl_fcnvfx(&fpregs[r1],0,
442 case 1: /* sgl/dbl */
443 return(sgl_to_dbl_fcnvfx(&fpregs[r1],0,
445 case 2: /* dbl/sgl */
446 return(dbl_to_sgl_fcnvfx(&fpregs[r1],0,
448 case 3: /* dbl/dbl */
449 return(dbl_to_dbl_fcnvfx(&fpregs[r1],0,
452 case 3: /* FCNVFXT */
454 case 0: /* sgl/sgl */
455 return(sgl_to_sgl_fcnvfxt(&fpregs[r1],0,
457 case 1: /* sgl/dbl */
458 return(sgl_to_dbl_fcnvfxt(&fpregs[r1],0,
460 case 2: /* dbl/sgl */
461 return(dbl_to_sgl_fcnvfxt(&fpregs[r1],0,
463 case 3: /* dbl/dbl */
464 return(dbl_to_dbl_fcnvfxt(&fpregs[r1],0,
467 case 5: /* FCNVUF (PA2.0 only) */
469 case 0: /* sgl/sgl */
470 return(sgl_to_sgl_fcnvuf(&fpregs[r1],0,
472 case 1: /* sgl/dbl */
473 return(sgl_to_dbl_fcnvuf(&fpregs[r1],0,
475 case 2: /* dbl/sgl */
476 return(dbl_to_sgl_fcnvuf(&fpregs[r1],0,
478 case 3: /* dbl/dbl */
479 return(dbl_to_dbl_fcnvuf(&fpregs[r1],0,
482 case 6: /* FCNVFU (PA2.0 only) */
484 case 0: /* sgl/sgl */
485 return(sgl_to_sgl_fcnvfu(&fpregs[r1],0,
487 case 1: /* sgl/dbl */
488 return(sgl_to_dbl_fcnvfu(&fpregs[r1],0,
490 case 2: /* dbl/sgl */
491 return(dbl_to_sgl_fcnvfu(&fpregs[r1],0,
493 case 3: /* dbl/dbl */
494 return(dbl_to_dbl_fcnvfu(&fpregs[r1],0,
497 case 7: /* FCNVFUT (PA2.0 only) */
499 case 0: /* sgl/sgl */
500 return(sgl_to_sgl_fcnvfut(&fpregs[r1],0,
502 case 1: /* sgl/dbl */
503 return(sgl_to_dbl_fcnvfut(&fpregs[r1],0,
505 case 2: /* dbl/sgl */
506 return(dbl_to_sgl_fcnvfut(&fpregs[r1],0,
508 case 3: /* dbl/dbl */
509 return(dbl_to_dbl_fcnvfut(&fpregs[r1],0,
512 case 4: /* undefined */
513 return(MAJOR_0C_EXCP);
514 } /* end of switch subop */
516 case 2: /* class 2 */
517 fpu_type_flags=fpregs[FPU_TYPE_FLAG_POS];
518 r2 = extru(ir, fpr2pos, 5) * sizeof(double)/sizeof(u_int);
521 if (fpu_type_flags & PA2_0_FPU_FLAG) {
522 /* FTEST if nullify bit set, otherwise FCMP */
523 if (extru(ir, fpnulpos, 1)) { /* FTEST */
528 * second param is the t field used for
529 * ftest,acc and ftest,rej
530 * third param is the subop (y-field)
534 * return(ftest(0L,extru(ir,fptpos,5),
535 * &fpregs[0],subop));
540 return(MAJOR_0C_EXCP);
545 retval = sgl_fcmp(&fpregs[r1],
546 &fpregs[r2],extru(ir,fptpos,5),
548 update_status_cbit(status,local_status,
549 fpu_type_flags, subop);
552 retval = dbl_fcmp(&fpregs[r1],
553 &fpregs[r2],extru(ir,fptpos,5),
555 update_status_cbit(status,local_status,
556 fpu_type_flags, subop);
558 case 2: /* illegal */
559 case 3: /* quad not implemented */
560 return(MAJOR_0C_EXCP);
563 } /* end of if for PA2.0 */
564 else { /* PA1.0 & PA1.1 */
572 return(MAJOR_0C_EXCP);
576 retval = sgl_fcmp(&fpregs[r1],
577 &fpregs[r2],extru(ir,fptpos,5),
579 update_status_cbit(status,local_status,
580 fpu_type_flags, subop);
583 retval = dbl_fcmp(&fpregs[r1],
584 &fpregs[r2],extru(ir,fptpos,5),
586 update_status_cbit(status,local_status,
587 fpu_type_flags, subop);
589 case 2: /* illegal */
590 case 3: /* quad not implemented */
591 return(MAJOR_0C_EXCP);
598 * second param is the t field used for
599 * ftest,acc and ftest,rej
600 * third param is the subop (y-field)
604 * return(ftest(0L,extru(ir,fptpos,5),
605 * &fpregs[0],subop));
610 return(MAJOR_0C_EXCP);
612 } /* end of switch subop */
613 } /* end of else for PA1.0 & PA1.1 */
614 case 3: /* class 3 */
615 r2 = extru(ir,fpr2pos,5) * sizeof(double)/sizeof(u_int);
622 return(MAJOR_0C_EXCP);
627 return(sgl_fadd(&fpregs[r1],&fpregs[r2],
630 return(dbl_fadd(&fpregs[r1],&fpregs[r2],
632 case 2: /* illegal */
633 case 3: /* quad not implemented */
634 return(MAJOR_0C_EXCP);
639 return(sgl_fsub(&fpregs[r1],&fpregs[r2],
642 return(dbl_fsub(&fpregs[r1],&fpregs[r2],
644 case 2: /* illegal */
645 case 3: /* quad not implemented */
646 return(MAJOR_0C_EXCP);
651 return(sgl_fmpy(&fpregs[r1],&fpregs[r2],
654 return(dbl_fmpy(&fpregs[r1],&fpregs[r2],
656 case 2: /* illegal */
657 case 3: /* quad not implemented */
658 return(MAJOR_0C_EXCP);
663 return(sgl_fdiv(&fpregs[r1],&fpregs[r2],
666 return(dbl_fdiv(&fpregs[r1],&fpregs[r2],
668 case 2: /* illegal */
669 case 3: /* quad not implemented */
670 return(MAJOR_0C_EXCP);
675 return(sgl_frem(&fpregs[r1],&fpregs[r2],
678 return(dbl_frem(&fpregs[r1],&fpregs[r2],
680 case 2: /* illegal */
681 case 3: /* quad not implemented */
682 return(MAJOR_0C_EXCP);
684 } /* end of class 3 switch */
685 } /* end of switch(class) */
687 /* If we get here, something is really wrong! */
688 return(MAJOR_0C_EXCP);
692 decode_0e(ir,class,subop,fpregs)
693 u_int ir,class,subop;
696 u_int r1,r2,t; /* operand register offsets */
697 u_int fmt; /* also sf for class 1 conversions */
698 u_int df; /* dest format for class 1 conversions */
700 u_int retval, local_status;
701 u_int fpu_type_flags;
704 local_status = fpregs[0];
705 r1 = ((extru(ir,fpr1pos,5)<<1)|(extru(ir,fpxr1pos,1)));
708 t = ((extru(ir,fptpos,5)<<1)|(extru(ir,fpxtpos,1)));
709 if (t == 0 && class != 2)
710 return(MAJOR_0E_EXCP);
711 if (class < 2) /* class 0 or 1 has 2 bit fmt */
712 fmt = extru(ir,fpfmtpos,2);
713 else /* class 2 and 3 have 1 bit fmt */
714 fmt = extru(ir,fp0efmtpos,1);
716 * An undefined combination, double precision accessing the
717 * right half of a FPR, can get us into trouble.
718 * Let's just force proper alignment on it.
729 case 0: /* unimplemented */
731 return(MAJOR_0E_EXCP);
736 return(MAJOR_0E_EXCP);
738 fpregs[t+1] = fpregs[r1+1];
740 fpregs[t] = fpregs[r1];
747 return(MAJOR_0E_EXCP);
749 fpregs[t+1] = fpregs[r1+1];
751 fpregs[t] = fpregs[r1] & 0x7fffffff;
758 return(MAJOR_0E_EXCP);
760 fpregs[t+1] = fpregs[r1+1];
762 fpregs[t] = fpregs[r1] ^ 0x80000000;
765 case 7: /* FNEGABS */
769 return(MAJOR_0E_EXCP);
771 fpregs[t+1] = fpregs[r1+1];
773 fpregs[t] = fpregs[r1] | 0x80000000;
779 return(sgl_fsqrt(&fpregs[r1],0,
780 &fpregs[t], status));
782 return(dbl_fsqrt(&fpregs[r1],0,
783 &fpregs[t], status));
786 return(MAJOR_0E_EXCP);
791 return(sgl_frnd(&fpregs[r1],0,
792 &fpregs[t], status));
794 return(dbl_frnd(&fpregs[r1],0,
795 &fpregs[t], status));
798 return(MAJOR_0E_EXCP);
800 } /* end of switch (subop */
802 case 1: /* class 1 */
803 df = extru(ir,fpdfpos,2); /* get dest format */
805 * Fix Crashme problem (writing to 31R in double precision)
811 if ((df & 2) || (fmt & 2))
812 return(MAJOR_0E_EXCP);
814 fmt = (fmt << 1) | df;
818 case 0: /* sgl/sgl */
819 return(MAJOR_0E_EXCP);
820 case 1: /* sgl/dbl */
821 return(sgl_to_dbl_fcnvff(&fpregs[r1],0,
823 case 2: /* dbl/sgl */
824 return(dbl_to_sgl_fcnvff(&fpregs[r1],0,
826 case 3: /* dbl/dbl */
827 return(MAJOR_0E_EXCP);
831 case 0: /* sgl/sgl */
832 return(sgl_to_sgl_fcnvxf(&fpregs[r1],0,
834 case 1: /* sgl/dbl */
835 return(sgl_to_dbl_fcnvxf(&fpregs[r1],0,
837 case 2: /* dbl/sgl */
838 return(dbl_to_sgl_fcnvxf(&fpregs[r1],0,
840 case 3: /* dbl/dbl */
841 return(dbl_to_dbl_fcnvxf(&fpregs[r1],0,
846 case 0: /* sgl/sgl */
847 return(sgl_to_sgl_fcnvfx(&fpregs[r1],0,
849 case 1: /* sgl/dbl */
850 return(sgl_to_dbl_fcnvfx(&fpregs[r1],0,
852 case 2: /* dbl/sgl */
853 return(dbl_to_sgl_fcnvfx(&fpregs[r1],0,
855 case 3: /* dbl/dbl */
856 return(dbl_to_dbl_fcnvfx(&fpregs[r1],0,
859 case 3: /* FCNVFXT */
861 case 0: /* sgl/sgl */
862 return(sgl_to_sgl_fcnvfxt(&fpregs[r1],0,
864 case 1: /* sgl/dbl */
865 return(sgl_to_dbl_fcnvfxt(&fpregs[r1],0,
867 case 2: /* dbl/sgl */
868 return(dbl_to_sgl_fcnvfxt(&fpregs[r1],0,
870 case 3: /* dbl/dbl */
871 return(dbl_to_dbl_fcnvfxt(&fpregs[r1],0,
874 case 5: /* FCNVUF (PA2.0 only) */
876 case 0: /* sgl/sgl */
877 return(sgl_to_sgl_fcnvuf(&fpregs[r1],0,
879 case 1: /* sgl/dbl */
880 return(sgl_to_dbl_fcnvuf(&fpregs[r1],0,
882 case 2: /* dbl/sgl */
883 return(dbl_to_sgl_fcnvuf(&fpregs[r1],0,
885 case 3: /* dbl/dbl */
886 return(dbl_to_dbl_fcnvuf(&fpregs[r1],0,
889 case 6: /* FCNVFU (PA2.0 only) */
891 case 0: /* sgl/sgl */
892 return(sgl_to_sgl_fcnvfu(&fpregs[r1],0,
894 case 1: /* sgl/dbl */
895 return(sgl_to_dbl_fcnvfu(&fpregs[r1],0,
897 case 2: /* dbl/sgl */
898 return(dbl_to_sgl_fcnvfu(&fpregs[r1],0,
900 case 3: /* dbl/dbl */
901 return(dbl_to_dbl_fcnvfu(&fpregs[r1],0,
904 case 7: /* FCNVFUT (PA2.0 only) */
906 case 0: /* sgl/sgl */
907 return(sgl_to_sgl_fcnvfut(&fpregs[r1],0,
909 case 1: /* sgl/dbl */
910 return(sgl_to_dbl_fcnvfut(&fpregs[r1],0,
912 case 2: /* dbl/sgl */
913 return(dbl_to_sgl_fcnvfut(&fpregs[r1],0,
915 case 3: /* dbl/dbl */
916 return(dbl_to_dbl_fcnvfut(&fpregs[r1],0,
919 case 4: /* undefined */
920 return(MAJOR_0C_EXCP);
921 } /* end of switch subop */
922 case 2: /* class 2 */
924 * Be careful out there.
925 * Crashme can generate cases where FR31R is specified
926 * as the source or target of a double precision operation.
927 * Since we just pass the address of the floating-point
928 * register to the emulation routines, this can cause
929 * corruption of fpzeroreg.
932 r2 = (extru(ir,fpr2pos,5)<<1);
934 r2 = ((extru(ir,fpr2pos,5)<<1)|(extru(ir,fpxr2pos,1)));
935 fpu_type_flags=fpregs[FPU_TYPE_FLAG_POS];
938 if (fpu_type_flags & PA2_0_FPU_FLAG) {
939 /* FTEST if nullify bit set, otherwise FCMP */
940 if (extru(ir, fpnulpos, 1)) { /* FTEST */
942 return(MAJOR_0E_EXCP);
946 * fmt is only 1 bit long
949 retval = sgl_fcmp(&fpregs[r1],
950 &fpregs[r2],extru(ir,fptpos,5),
952 update_status_cbit(status,local_status,
953 fpu_type_flags, subop);
956 retval = dbl_fcmp(&fpregs[r1],
957 &fpregs[r2],extru(ir,fptpos,5),
959 update_status_cbit(status,local_status,
960 fpu_type_flags, subop);
964 } /* end of if for PA2.0 */
965 else { /* PA1.0 & PA1.1 */
974 return(MAJOR_0E_EXCP);
978 * fmt is only 1 bit long
981 retval = sgl_fcmp(&fpregs[r1],
982 &fpregs[r2],extru(ir,fptpos,5),
984 update_status_cbit(status,local_status,
985 fpu_type_flags, subop);
988 retval = dbl_fcmp(&fpregs[r1],
989 &fpregs[r2],extru(ir,fptpos,5),
991 update_status_cbit(status,local_status,
992 fpu_type_flags, subop);
995 } /* end of switch subop */
996 } /* end of else for PA1.0 & PA1.1 */
997 case 3: /* class 3 */
999 * Be careful out there.
1000 * Crashme can generate cases where FR31R is specified
1001 * as the source or target of a double precision operation.
1002 * Since we just pass the address of the floating-point
1003 * register to the emulation routines, this can cause
1004 * corruption of fpzeroreg.
1007 r2 = (extru(ir,fpr2pos,5)<<1);
1009 r2 = ((extru(ir,fpr2pos,5)<<1)|(extru(ir,fpxr2pos,1)));
1016 return(MAJOR_0E_EXCP);
1019 * Note that fmt is only 1 bit for class 3 */
1023 return(sgl_fadd(&fpregs[r1],&fpregs[r2],
1024 &fpregs[t],status));
1026 return(dbl_fadd(&fpregs[r1],&fpregs[r2],
1027 &fpregs[t],status));
1032 return(sgl_fsub(&fpregs[r1],&fpregs[r2],
1033 &fpregs[t],status));
1035 return(dbl_fsub(&fpregs[r1],&fpregs[r2],
1036 &fpregs[t],status));
1038 case 2: /* FMPY or XMPYU */
1040 * check for integer multiply (x bit set)
1042 if (extru(ir,fpxpos,1)) {
1049 * bad instruction if t specifies
1050 * the right half of a register
1053 return(MAJOR_0E_EXCP);
1056 * impyu(&fpregs[r1],&fpregs[r2],
1059 return(NOEXCEPTION);
1061 return(MAJOR_0E_EXCP);
1067 return(sgl_fmpy(&fpregs[r1],
1068 &fpregs[r2],&fpregs[t],status));
1070 return(dbl_fmpy(&fpregs[r1],
1071 &fpregs[r2],&fpregs[t],status));
1077 return(sgl_fdiv(&fpregs[r1],&fpregs[r2],
1078 &fpregs[t],status));
1080 return(dbl_fdiv(&fpregs[r1],&fpregs[r2],
1081 &fpregs[t],status));
1086 return(sgl_frem(&fpregs[r1],&fpregs[r2],
1087 &fpregs[t],status));
1089 return(dbl_frem(&fpregs[r1],&fpregs[r2],
1090 &fpregs[t],status));
1092 } /* end of class 3 switch */
1093 } /* end of switch(class) */
1095 /* If we get here, something is really wrong! */
1096 return(MAJOR_0E_EXCP);
1101 * routine to decode the 06 (FMPYADD and FMPYCFXT) instruction
1104 decode_06(ir,fpregs)
1108 u_int rm1, rm2, tm, ra, ta; /* operands */
1112 u_int fpu_type_flags;
1116 struct { u_int i1; u_int i2; } ints;
1120 status = fpregs[0]; /* use a local copy of status reg */
1121 fpu_type_flags=fpregs[FPU_TYPE_FLAG_POS]; /* get fpu type flags */
1122 fmt = extru(ir, fpmultifmt, 1); /* get sgl/dbl flag */
1123 if (fmt == 0) { /* DBL */
1124 rm1 = extru(ir, fprm1pos, 5) * sizeof(double)/sizeof(u_int);
1127 rm2 = extru(ir, fprm2pos, 5) * sizeof(double)/sizeof(u_int);
1130 tm = extru(ir, fptmpos, 5) * sizeof(double)/sizeof(u_int);
1132 return(MAJOR_06_EXCP);
1133 ra = extru(ir, fprapos, 5) * sizeof(double)/sizeof(u_int);
1134 ta = extru(ir, fptapos, 5) * sizeof(double)/sizeof(u_int);
1136 return(MAJOR_06_EXCP);
1138 if (fpu_type_flags & TIMEX_ROLEX_FPU_MASK) {
1141 /* special case FMPYCFXT, see sgl case below */
1142 if (dbl_fmpy(&fpregs[rm1],&fpregs[rm2],
1143 &mtmp.ints.i1,&status))
1145 if (dbl_to_sgl_fcnvfxt(&fpregs[ta],
1146 &atmp.ints.i1,&atmp.ints.i1,&status))
1151 if (dbl_fmpy(&fpregs[rm1],&fpregs[rm2],&mtmp.ints.i1,
1154 if (dbl_fadd(&fpregs[ta], &fpregs[ra], &atmp.ints.i1,
1166 if (dbl_fmpy(&fpregs[rm1],&fpregs[rm2],&mtmp.ints.i1,
1169 if (dbl_fadd(&fpregs[ta], &fpregs[ra], &atmp.ints.i1,
1176 return(MAJOR_06_EXCP);
1179 fpregs[tm] = mtmp.ints.i1;
1180 fpregs[tm+1] = mtmp.ints.i2;
1181 fpregs[ta] = atmp.ints.i1;
1182 fpregs[ta+1] = atmp.ints.i2;
1184 return(NOEXCEPTION);
1189 * calculate offsets for single precision numbers
1190 * See table 6-14 in PA-89 architecture for mapping
1192 rm1 = (extru(ir,fprm1pos,4) | 0x10 ) << 1; /* get offset */
1193 rm1 |= extru(ir,fprm1pos-4,1); /* add right word offset */
1195 rm2 = (extru(ir,fprm2pos,4) | 0x10 ) << 1; /* get offset */
1196 rm2 |= extru(ir,fprm2pos-4,1); /* add right word offset */
1198 tm = (extru(ir,fptmpos,4) | 0x10 ) << 1; /* get offset */
1199 tm |= extru(ir,fptmpos-4,1); /* add right word offset */
1201 ra = (extru(ir,fprapos,4) | 0x10 ) << 1; /* get offset */
1202 ra |= extru(ir,fprapos-4,1); /* add right word offset */
1204 ta = (extru(ir,fptapos,4) | 0x10 ) << 1; /* get offset */
1205 ta |= extru(ir,fptapos-4,1); /* add right word offset */
1207 if (ra == 0x20 &&(fpu_type_flags & TIMEX_ROLEX_FPU_MASK)) {
1208 /* special case FMPYCFXT (really 0)
1209 * This instruction is only present on the Timex and
1210 * Rolex fpu's in so if it is the special case and
1211 * one of these fpu's we run the FMPYCFXT instruction
1213 if (sgl_fmpy(&fpregs[rm1],&fpregs[rm2],&mtmp.ints.i1,
1216 if (sgl_to_sgl_fcnvfxt(&fpregs[ta],&atmp.ints.i1,
1217 &atmp.ints.i1,&status))
1221 if (sgl_fmpy(&fpregs[rm1],&fpregs[rm2],&mtmp.ints.i1,
1224 if (sgl_fadd(&fpregs[ta], &fpregs[ra], &atmp.ints.i1,
1229 return(MAJOR_06_EXCP);
1232 fpregs[tm] = mtmp.ints.i1;
1233 fpregs[ta] = atmp.ints.i1;
1235 return(NOEXCEPTION);
1241 * routine to decode the 26 (FMPYSUB) instruction
1244 decode_26(ir,fpregs)
1248 u_int rm1, rm2, tm, ra, ta; /* operands */
1255 struct { u_int i1; u_int i2; } ints;
1260 fmt = extru(ir, fpmultifmt, 1); /* get sgl/dbl flag */
1261 if (fmt == 0) { /* DBL */
1262 rm1 = extru(ir, fprm1pos, 5) * sizeof(double)/sizeof(u_int);
1265 rm2 = extru(ir, fprm2pos, 5) * sizeof(double)/sizeof(u_int);
1268 tm = extru(ir, fptmpos, 5) * sizeof(double)/sizeof(u_int);
1270 return(MAJOR_26_EXCP);
1271 ra = extru(ir, fprapos, 5) * sizeof(double)/sizeof(u_int);
1273 return(MAJOR_26_EXCP);
1274 ta = extru(ir, fptapos, 5) * sizeof(double)/sizeof(u_int);
1276 return(MAJOR_26_EXCP);
1278 if (dbl_fmpy(&fpregs[rm1],&fpregs[rm2],&mtmp.ints.i1,&status))
1280 if (dbl_fsub(&fpregs[ta], &fpregs[ra], &atmp.ints.i1,&status))
1283 return(MAJOR_26_EXCP);
1286 fpregs[tm] = mtmp.ints.i1;
1287 fpregs[tm+1] = mtmp.ints.i2;
1288 fpregs[ta] = atmp.ints.i1;
1289 fpregs[ta+1] = atmp.ints.i2;
1291 return(NOEXCEPTION);
1296 * calculate offsets for single precision numbers
1297 * See table 6-14 in PA-89 architecture for mapping
1299 rm1 = (extru(ir,fprm1pos,4) | 0x10 ) << 1; /* get offset */
1300 rm1 |= extru(ir,fprm1pos-4,1); /* add right word offset */
1302 rm2 = (extru(ir,fprm2pos,4) | 0x10 ) << 1; /* get offset */
1303 rm2 |= extru(ir,fprm2pos-4,1); /* add right word offset */
1305 tm = (extru(ir,fptmpos,4) | 0x10 ) << 1; /* get offset */
1306 tm |= extru(ir,fptmpos-4,1); /* add right word offset */
1308 ra = (extru(ir,fprapos,4) | 0x10 ) << 1; /* get offset */
1309 ra |= extru(ir,fprapos-4,1); /* add right word offset */
1311 ta = (extru(ir,fptapos,4) | 0x10 ) << 1; /* get offset */
1312 ta |= extru(ir,fptapos-4,1); /* add right word offset */
1314 if (sgl_fmpy(&fpregs[rm1],&fpregs[rm2],&mtmp.ints.i1,&status))
1316 if (sgl_fsub(&fpregs[ta], &fpregs[ra], &atmp.ints.i1,&status))
1319 return(MAJOR_26_EXCP);
1322 fpregs[tm] = mtmp.ints.i1;
1323 fpregs[ta] = atmp.ints.i1;
1325 return(NOEXCEPTION);
1332 * routine to decode the 2E (FMPYFADD,FMPYNFADD) instructions
1335 decode_2e(ir,fpregs)
1339 u_int rm1, rm2, ra, t; /* operands */
1342 fmt = extru(ir,fpfmtpos,1); /* get fmt completer */
1343 if (fmt == DBL) { /* DBL */
1344 rm1 = extru(ir,fprm1pos,5) * sizeof(double)/sizeof(u_int);
1347 rm2 = extru(ir,fprm2pos,5) * sizeof(double)/sizeof(u_int);
1350 ra = ((extru(ir,fpraupos,3)<<2)|(extru(ir,fpralpos,3)>>1)) *
1351 sizeof(double)/sizeof(u_int);
1354 t = extru(ir,fptpos,5) * sizeof(double)/sizeof(u_int);
1356 return(MAJOR_2E_EXCP);
1358 if (extru(ir,fpfusedsubop,1)) { /* fmpyfadd or fmpynfadd? */
1359 return(dbl_fmpynfadd(&fpregs[rm1], &fpregs[rm2],
1360 &fpregs[ra], &fpregs[0], &fpregs[t]));
1362 return(dbl_fmpyfadd(&fpregs[rm1], &fpregs[rm2],
1363 &fpregs[ra], &fpregs[0], &fpregs[t]));
1367 rm1 = (extru(ir,fprm1pos,5)<<1)|(extru(ir,fpxrm1pos,1));
1370 rm2 = (extru(ir,fprm2pos,5)<<1)|(extru(ir,fpxrm2pos,1));
1373 ra = (extru(ir,fpraupos,3)<<3)|extru(ir,fpralpos,3);
1376 t = ((extru(ir,fptpos,5)<<1)|(extru(ir,fpxtpos,1)));
1378 return(MAJOR_2E_EXCP);
1380 if (extru(ir,fpfusedsubop,1)) { /* fmpyfadd or fmpynfadd? */
1381 return(sgl_fmpynfadd(&fpregs[rm1], &fpregs[rm2],
1382 &fpregs[ra], &fpregs[0], &fpregs[t]));
1384 return(sgl_fmpyfadd(&fpregs[rm1], &fpregs[rm2],
1385 &fpregs[ra], &fpregs[0], &fpregs[t]));
1391 * update_status_cbit
1393 * This routine returns the correct FP status register value in
1394 * *status, based on the C-bit & V-bit returned by the FCMP
1395 * emulation routine in new_status. The architecture type
1396 * (PA83, PA89 or PA2.0) is available in fpu_type. The y_field
1397 * and the architecture type are used to determine what flavor
1398 * of FCMP is being emulated.
1401 update_status_cbit(status, new_status, fpu_type, y_field)
1402 u_int *status, new_status;
1407 * For PA89 FPU's which implement the Compare Queue and
1408 * for PA2.0 FPU's, update the Compare Queue if the y-field = 0,
1409 * otherwise update the specified bit in the Compare Array.
1410 * Note that the y-field will always be 0 for non-PA2.0 FPU's.
1412 if ((fpu_type & TIMEX_EXTEN_FLAG) ||
1413 (fpu_type & ROLEX_EXTEN_FLAG) ||
1414 (fpu_type & PA2_0_FPU_FLAG)) {
1416 *status = ((*status & 0x04000000) >> 5) | /* old Cbit */
1417 ((*status & 0x003ff000) >> 1) | /* old CQ */
1418 (new_status & 0xffc007ff); /* all other bits*/
1420 *status = (*status & 0x04000000) | /* old Cbit */
1421 ((new_status & 0x04000000) >> (y_field+4)) |
1422 (new_status & ~0x04000000 & /* other bits */
1423 ~(0x04000000 >> (y_field+4)));
1426 /* if PA83, just update the C-bit */
1428 *status = new_status;