diff options
author | Takashi YOSHII <takasi-y@ops.dti.ne.jp> | 2006-09-27 04:15:32 -0400 |
---|---|---|
committer | Paul Mundt <lethal@linux-sh.org> | 2006-09-27 04:15:32 -0400 |
commit | 4b565680d16300acab0ff167e24f0ea289a6bd5d (patch) | |
tree | d233188a29c0a0ccfdf72c84899e926df2fc2ba8 /arch | |
parent | 317a6104a99f87c0b35c0d9f19ec23ee7429b33e (diff) |
sh: math-emu support
This implements initial math-emu support, aimed primarily at SH-3.
Signed-off-by: Takashi YOSHII <takasi-y@ops.dti.ne.jp>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Diffstat (limited to 'arch')
-rw-r--r-- | arch/sh/Kconfig | 9 | ||||
-rw-r--r-- | arch/sh/Makefile | 1 | ||||
-rw-r--r-- | arch/sh/kernel/traps.c | 164 | ||||
-rw-r--r-- | arch/sh/math-emu/Makefile | 1 | ||||
-rw-r--r-- | arch/sh/math-emu/math.c | 624 | ||||
-rw-r--r-- | arch/sh/math-emu/sfp-util.h | 72 |
6 files changed, 832 insertions, 39 deletions
diff --git a/arch/sh/Kconfig b/arch/sh/Kconfig index d1d724447826..07d6d699e9c2 100644 --- a/arch/sh/Kconfig +++ b/arch/sh/Kconfig | |||
@@ -339,6 +339,15 @@ config SH_FPU | |||
339 | 339 | ||
340 | This option must be set in order to enable the FPU. | 340 | This option must be set in order to enable the FPU. |
341 | 341 | ||
342 | config SH_FPU_EMU | ||
343 | bool "FPU emulation support" | ||
344 | depends on !SH_FPU && EXPERIMENTAL | ||
345 | default n | ||
346 | help | ||
347 | Selecting this option will enable support for software FPU emulation. | ||
348 | Most SH-3 users will want to say Y here, whereas most SH-4 users will | ||
349 | want to say N. | ||
350 | |||
342 | config SH_DSP | 351 | config SH_DSP |
343 | bool "DSP support" | 352 | bool "DSP support" |
344 | depends on !CPU_SH4 | 353 | depends on !CPU_SH4 |
diff --git a/arch/sh/Makefile b/arch/sh/Makefile index 859285d47a29..13b688b13e1c 100644 --- a/arch/sh/Makefile +++ b/arch/sh/Makefile | |||
@@ -79,6 +79,7 @@ head-y := arch/sh/kernel/head.o arch/sh/kernel/init_task.o | |||
79 | LIBGCC := $(shell $(CC) $(CFLAGS) -print-libgcc-file-name) | 79 | LIBGCC := $(shell $(CC) $(CFLAGS) -print-libgcc-file-name) |
80 | 80 | ||
81 | core-y += arch/sh/kernel/ arch/sh/mm/ | 81 | core-y += arch/sh/kernel/ arch/sh/mm/ |
82 | core-$(CONFIG_SH_FPU_EMU) += arch/sh/math-emu/ | ||
82 | 83 | ||
83 | # Boards | 84 | # Boards |
84 | machdir-$(CONFIG_SH_SOLUTION_ENGINE) := se/770x | 85 | machdir-$(CONFIG_SH_SOLUTION_ENGINE) := se/770x |
diff --git a/arch/sh/kernel/traps.c b/arch/sh/kernel/traps.c index 21bef1b5f991..95c810b3c97e 100644 --- a/arch/sh/kernel/traps.c +++ b/arch/sh/kernel/traps.c | |||
@@ -36,40 +36,15 @@ | |||
36 | 36 | ||
37 | #ifdef CONFIG_SH_KGDB | 37 | #ifdef CONFIG_SH_KGDB |
38 | #include <asm/kgdb.h> | 38 | #include <asm/kgdb.h> |
39 | #define CHK_REMOTE_DEBUG(regs) \ | 39 | #define CHK_REMOTE_DEBUG(regs) \ |
40 | { \ | 40 | { \ |
41 | if ((kgdb_debug_hook != (kgdb_debug_hook_t *) NULL) && (!user_mode(regs))) \ | 41 | if (kgdb_debug_hook && !user_mode(regs))\ |
42 | { \ | 42 | (*kgdb_debug_hook)(regs); \ |
43 | (*kgdb_debug_hook)(regs); \ | ||
44 | } \ | ||
45 | } | 43 | } |
46 | #else | 44 | #else |
47 | #define CHK_REMOTE_DEBUG(regs) | 45 | #define CHK_REMOTE_DEBUG(regs) |
48 | #endif | 46 | #endif |
49 | 47 | ||
50 | #define DO_ERROR(trapnr, signr, str, name, tsk) \ | ||
51 | asmlinkage void do_##name(unsigned long r4, unsigned long r5, \ | ||
52 | unsigned long r6, unsigned long r7, \ | ||
53 | struct pt_regs regs) \ | ||
54 | { \ | ||
55 | unsigned long error_code; \ | ||
56 | \ | ||
57 | /* Check if it's a DSP instruction */ \ | ||
58 | if (is_dsp_inst(®s)) { \ | ||
59 | /* Enable DSP mode, and restart instruction. */ \ | ||
60 | regs.sr |= SR_DSP; \ | ||
61 | return; \ | ||
62 | } \ | ||
63 | \ | ||
64 | asm volatile("stc r2_bank, %0": "=r" (error_code)); \ | ||
65 | local_irq_enable(); \ | ||
66 | tsk->thread.error_code = error_code; \ | ||
67 | tsk->thread.trap_no = trapnr; \ | ||
68 | CHK_REMOTE_DEBUG(®s); \ | ||
69 | force_sig(signr, tsk); \ | ||
70 | die_if_no_fixup(str,®s,error_code); \ | ||
71 | } | ||
72 | |||
73 | #ifdef CONFIG_CPU_SH2 | 48 | #ifdef CONFIG_CPU_SH2 |
74 | #define TRAP_RESERVED_INST 4 | 49 | #define TRAP_RESERVED_INST 4 |
75 | #define TRAP_ILLEGAL_SLOT_INST 6 | 50 | #define TRAP_ILLEGAL_SLOT_INST 6 |
@@ -575,8 +550,117 @@ int is_dsp_inst(struct pt_regs *regs) | |||
575 | #define is_dsp_inst(regs) (0) | 550 | #define is_dsp_inst(regs) (0) |
576 | #endif /* CONFIG_SH_DSP */ | 551 | #endif /* CONFIG_SH_DSP */ |
577 | 552 | ||
578 | DO_ERROR(TRAP_RESERVED_INST, SIGILL, "reserved instruction", reserved_inst, current) | 553 | extern int do_fpu_inst(unsigned short, struct pt_regs*); |
579 | DO_ERROR(TRAP_ILLEGAL_SLOT_INST, SIGILL, "illegal slot instruction", illegal_slot_inst, current) | 554 | |
555 | asmlinkage void do_reserved_inst(unsigned long r4, unsigned long r5, | ||
556 | unsigned long r6, unsigned long r7, | ||
557 | struct pt_regs regs) | ||
558 | { | ||
559 | unsigned long error_code; | ||
560 | struct task_struct *tsk = current; | ||
561 | |||
562 | #ifdef CONFIG_SH_FPU_EMU | ||
563 | unsigned short inst; | ||
564 | int err; | ||
565 | |||
566 | get_user(inst, (unsigned short*)regs.pc); | ||
567 | |||
568 | err = do_fpu_inst(inst, ®s); | ||
569 | if (!err) { | ||
570 | regs.pc += 2; | ||
571 | return; | ||
572 | } | ||
573 | /* not a FPU inst. */ | ||
574 | #endif | ||
575 | |||
576 | #ifdef CONFIG_SH_DSP | ||
577 | /* Check if it's a DSP instruction */ | ||
578 | if (is_dsp_inst(®s)) { | ||
579 | /* Enable DSP mode, and restart instruction. */ | ||
580 | regs.sr |= SR_DSP; | ||
581 | return; | ||
582 | } | ||
583 | #endif | ||
584 | |||
585 | asm volatile("stc r2_bank, %0": "=r" (error_code)); | ||
586 | local_irq_enable(); | ||
587 | tsk->thread.error_code = error_code; | ||
588 | tsk->thread.trap_no = TRAP_RESERVED_INST; | ||
589 | CHK_REMOTE_DEBUG(®s); | ||
590 | force_sig(SIGILL, tsk); | ||
591 | die_if_no_fixup("reserved instruction", ®s, error_code); | ||
592 | } | ||
593 | |||
594 | #ifdef CONFIG_SH_FPU_EMU | ||
595 | static int emulate_branch(unsigned short inst, struct pt_regs* regs) | ||
596 | { | ||
597 | /* | ||
598 | * bfs: 8fxx: PC+=d*2+4; | ||
599 | * bts: 8dxx: PC+=d*2+4; | ||
600 | * bra: axxx: PC+=D*2+4; | ||
601 | * bsr: bxxx: PC+=D*2+4 after PR=PC+4; | ||
602 | * braf:0x23: PC+=Rn*2+4; | ||
603 | * bsrf:0x03: PC+=Rn*2+4 after PR=PC+4; | ||
604 | * jmp: 4x2b: PC=Rn; | ||
605 | * jsr: 4x0b: PC=Rn after PR=PC+4; | ||
606 | * rts: 000b: PC=PR; | ||
607 | */ | ||
608 | if ((inst & 0xfd00) == 0x8d00) { | ||
609 | regs->pc += SH_PC_8BIT_OFFSET(inst); | ||
610 | return 0; | ||
611 | } | ||
612 | |||
613 | if ((inst & 0xe000) == 0xa000) { | ||
614 | regs->pc += SH_PC_12BIT_OFFSET(inst); | ||
615 | return 0; | ||
616 | } | ||
617 | |||
618 | if ((inst & 0xf0df) == 0x0003) { | ||
619 | regs->pc += regs->regs[(inst & 0x0f00) >> 8] + 4; | ||
620 | return 0; | ||
621 | } | ||
622 | |||
623 | if ((inst & 0xf0df) == 0x400b) { | ||
624 | regs->pc = regs->regs[(inst & 0x0f00) >> 8]; | ||
625 | return 0; | ||
626 | } | ||
627 | |||
628 | if ((inst & 0xffff) == 0x000b) { | ||
629 | regs->pc = regs->pr; | ||
630 | return 0; | ||
631 | } | ||
632 | |||
633 | return 1; | ||
634 | } | ||
635 | #endif | ||
636 | |||
637 | asmlinkage void do_illegal_slot_inst(unsigned long r4, unsigned long r5, | ||
638 | unsigned long r6, unsigned long r7, | ||
639 | struct pt_regs regs) | ||
640 | { | ||
641 | unsigned long error_code; | ||
642 | struct task_struct *tsk = current; | ||
643 | #ifdef CONFIG_SH_FPU_EMU | ||
644 | unsigned short inst; | ||
645 | |||
646 | get_user(inst, (unsigned short *)regs.pc + 1); | ||
647 | if (!do_fpu_inst(inst, ®s)) { | ||
648 | get_user(inst, (unsigned short *)regs.pc); | ||
649 | if (!emulate_branch(inst, ®s)) | ||
650 | return; | ||
651 | /* fault in branch.*/ | ||
652 | } | ||
653 | /* not a FPU inst. */ | ||
654 | #endif | ||
655 | |||
656 | asm volatile("stc r2_bank, %0": "=r" (error_code)); | ||
657 | local_irq_enable(); | ||
658 | tsk->thread.error_code = error_code; | ||
659 | tsk->thread.trap_no = TRAP_RESERVED_INST; | ||
660 | CHK_REMOTE_DEBUG(®s); | ||
661 | force_sig(SIGILL, tsk); | ||
662 | die_if_no_fixup("illegal slot instruction", ®s, error_code); | ||
663 | } | ||
580 | 664 | ||
581 | asmlinkage void do_exception_error(unsigned long r4, unsigned long r5, | 665 | asmlinkage void do_exception_error(unsigned long r4, unsigned long r5, |
582 | unsigned long r6, unsigned long r7, | 666 | unsigned long r6, unsigned long r7, |
@@ -634,14 +718,16 @@ void __init trap_init(void) | |||
634 | exception_handling_table[TRAP_ILLEGAL_SLOT_INST] | 718 | exception_handling_table[TRAP_ILLEGAL_SLOT_INST] |
635 | = (void *)do_illegal_slot_inst; | 719 | = (void *)do_illegal_slot_inst; |
636 | 720 | ||
637 | #ifdef CONFIG_CPU_SH4 | 721 | #if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SH_FPU) || \ |
638 | if (!(cpu_data->flags & CPU_HAS_FPU)) { | 722 | defined(CONFIG_SH_FPU_EMU) |
639 | /* For SH-4 lacking an FPU, treat floating point instructions | 723 | /* |
640 | as reserved. */ | 724 | * For SH-4 lacking an FPU, treat floating point instructions as |
641 | /* entry 64 corresponds to EXPEVT=0x800 */ | 725 | * reserved. They'll be handled in the math-emu case, or faulted on |
642 | exception_handling_table[64] = (void *)do_reserved_inst; | 726 | * otherwise. |
643 | exception_handling_table[65] = (void *)do_illegal_slot_inst; | 727 | */ |
644 | } | 728 | /* entry 64 corresponds to EXPEVT=0x800 */ |
729 | exception_handling_table[64] = (void *)do_reserved_inst; | ||
730 | exception_handling_table[65] = (void *)do_illegal_slot_inst; | ||
645 | #endif | 731 | #endif |
646 | 732 | ||
647 | /* Setup VBR for boot cpu */ | 733 | /* Setup VBR for boot cpu */ |
diff --git a/arch/sh/math-emu/Makefile b/arch/sh/math-emu/Makefile new file mode 100644 index 000000000000..638b342c781a --- /dev/null +++ b/arch/sh/math-emu/Makefile | |||
@@ -0,0 +1 @@ | |||
obj-y := math.o | |||
diff --git a/arch/sh/math-emu/math.c b/arch/sh/math-emu/math.c new file mode 100644 index 000000000000..26b6046814fd --- /dev/null +++ b/arch/sh/math-emu/math.c | |||
@@ -0,0 +1,624 @@ | |||
1 | /* | ||
2 | * arch/sh/math-emu/math.c | ||
3 | * | ||
4 | * Copyright (C) 2006 Takashi YOSHII <takasi-y@ops.dti.ne.jp> | ||
5 | * | ||
6 | * This file is subject to the terms and conditions of the GNU General Public | ||
7 | * License. See the file "COPYING" in the main directory of this archive | ||
8 | * for more details. | ||
9 | */ | ||
10 | #include <linux/config.h> | ||
11 | #include <linux/kernel.h> | ||
12 | #include <linux/errno.h> | ||
13 | #include <linux/types.h> | ||
14 | #include <linux/sched.h> | ||
15 | #include <linux/signal.h> | ||
16 | |||
17 | #include <asm/system.h> | ||
18 | #include <asm/uaccess.h> | ||
19 | #include <asm/processor.h> | ||
20 | #include <asm/io.h> | ||
21 | |||
22 | #include "sfp-util.h" | ||
23 | #include <math-emu/soft-fp.h> | ||
24 | #include <math-emu/single.h> | ||
25 | #include <math-emu/double.h> | ||
26 | |||
27 | #define FPUL (fregs->fpul) | ||
28 | #define FPSCR (fregs->fpscr) | ||
29 | #define FPSCR_RM (FPSCR&3) | ||
30 | #define FPSCR_DN ((FPSCR>>18)&1) | ||
31 | #define FPSCR_PR ((FPSCR>>19)&1) | ||
32 | #define FPSCR_SZ ((FPSCR>>20)&1) | ||
33 | #define FPSCR_FR ((FPSCR>>21)&1) | ||
34 | #define FPSCR_MASK 0x003fffffUL | ||
35 | |||
36 | #define BANK(n) (n^(FPSCR_FR?16:0)) | ||
37 | #define FR ((unsigned long*)(fregs->fp_regs)) | ||
38 | #define FR0 (FR[BANK(0)]) | ||
39 | #define FRn (FR[BANK(n)]) | ||
40 | #define FRm (FR[BANK(m)]) | ||
41 | #define DR ((unsigned long long*)(fregs->fp_regs)) | ||
42 | #define DRn (DR[BANK(n)/2]) | ||
43 | #define DRm (DR[BANK(m)/2]) | ||
44 | |||
45 | #define XREG(n) (n^16) | ||
46 | #define XFn (FR[BANK(XREG(n))]) | ||
47 | #define XFm (FR[BANK(XREG(m))]) | ||
48 | #define XDn (DR[BANK(XREG(n))/2]) | ||
49 | #define XDm (DR[BANK(XREG(m))/2]) | ||
50 | |||
51 | #define R0 (regs->regs[0]) | ||
52 | #define Rn (regs->regs[n]) | ||
53 | #define Rm (regs->regs[m]) | ||
54 | |||
55 | #define WRITE(d,a) ({if(put_user(d, (typeof (d)*)a)) return -EFAULT;}) | ||
56 | #define READ(d,a) ({if(get_user(d, (typeof (d)*)a)) return -EFAULT;}) | ||
57 | |||
58 | #define PACK_S(r,f) FP_PACK_SP(&r,f) | ||
59 | #define UNPACK_S(f,r) FP_UNPACK_SP(f,&r) | ||
60 | #define PACK_D(r,f) \ | ||
61 | {u32 t[2]; FP_PACK_DP(t,f); ((u32*)&r)[0]=t[1]; ((u32*)&r)[1]=t[0];} | ||
62 | #define UNPACK_D(f,r) \ | ||
63 | {u32 t[2]; t[0]=((u32*)&r)[1]; t[1]=((u32*)&r)[0]; FP_UNPACK_DP(f,t);} | ||
64 | |||
65 | // 2 args instructions. | ||
66 | #define BOTH_PRmn(op,x) \ | ||
67 | FP_DECL_EX; if(FPSCR_PR) op(D,x,DRm,DRn); else op(S,x,FRm,FRn); | ||
68 | |||
69 | #define CMP_X(SZ,R,M,N) do{ \ | ||
70 | FP_DECL_##SZ(Fm); FP_DECL_##SZ(Fn); \ | ||
71 | UNPACK_##SZ(Fm, M); UNPACK_##SZ(Fn, N); \ | ||
72 | FP_CMP_##SZ(R, Fn, Fm, 2); }while(0) | ||
73 | #define EQ_X(SZ,R,M,N) do{ \ | ||
74 | FP_DECL_##SZ(Fm); FP_DECL_##SZ(Fn); \ | ||
75 | UNPACK_##SZ(Fm, M); UNPACK_##SZ(Fn, N); \ | ||
76 | FP_CMP_EQ_##SZ(R, Fn, Fm); }while(0) | ||
77 | #define CMP(OP) ({ int r; BOTH_PRmn(OP##_X,r); r; }) | ||
78 | |||
79 | static int | ||
80 | fcmp_gt(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
81 | { | ||
82 | if (CMP(CMP) > 0) | ||
83 | regs->sr |= 1; | ||
84 | else | ||
85 | regs->sr &= ~1; | ||
86 | |||
87 | return 0; | ||
88 | } | ||
89 | |||
90 | static int | ||
91 | fcmp_eq(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
92 | { | ||
93 | if (CMP(CMP /*EQ*/) == 0) | ||
94 | regs->sr |= 1; | ||
95 | else | ||
96 | regs->sr &= ~1; | ||
97 | return 0; | ||
98 | } | ||
99 | |||
100 | #define ARITH_X(SZ,OP,M,N) do{ \ | ||
101 | FP_DECL_##SZ(Fm); FP_DECL_##SZ(Fn); FP_DECL_##SZ(Fr); \ | ||
102 | UNPACK_##SZ(Fm, M); UNPACK_##SZ(Fn, N); \ | ||
103 | FP_##OP##_##SZ(Fr, Fn, Fm); \ | ||
104 | PACK_##SZ(N, Fr); }while(0) | ||
105 | |||
106 | static int | ||
107 | fadd(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
108 | { | ||
109 | BOTH_PRmn(ARITH_X, ADD); | ||
110 | return 0; | ||
111 | } | ||
112 | |||
113 | static int | ||
114 | fsub(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
115 | { | ||
116 | BOTH_PRmn(ARITH_X, SUB); | ||
117 | return 0; | ||
118 | } | ||
119 | |||
120 | static int | ||
121 | fmul(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
122 | { | ||
123 | BOTH_PRmn(ARITH_X, MUL); | ||
124 | return 0; | ||
125 | } | ||
126 | |||
127 | static int | ||
128 | fdiv(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
129 | { | ||
130 | BOTH_PRmn(ARITH_X, DIV); | ||
131 | return 0; | ||
132 | } | ||
133 | |||
134 | static int | ||
135 | fmac(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
136 | { | ||
137 | FP_DECL_EX; | ||
138 | FP_DECL_S(Fr); | ||
139 | FP_DECL_S(Ft); | ||
140 | FP_DECL_S(F0); | ||
141 | FP_DECL_S(Fm); | ||
142 | FP_DECL_S(Fn); | ||
143 | UNPACK_S(F0, FR0); | ||
144 | UNPACK_S(Fm, FRm); | ||
145 | UNPACK_S(Fn, FRn); | ||
146 | FP_MUL_S(Ft, Fm, F0); | ||
147 | FP_ADD_S(Fr, Fn, Ft); | ||
148 | PACK_S(FRn, Fr); | ||
149 | return 0; | ||
150 | } | ||
151 | |||
152 | // to process fmov's extention (odd n for DR access XD). | ||
153 | #define FMOV_EXT(x) if(x&1) x+=16-1 | ||
154 | |||
155 | static int | ||
156 | fmov_idx_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
157 | int n) | ||
158 | { | ||
159 | if (FPSCR_SZ) { | ||
160 | FMOV_EXT(n); | ||
161 | READ(FRn, Rm + R0 + 4); | ||
162 | n++; | ||
163 | READ(FRn, Rm + R0); | ||
164 | } else { | ||
165 | READ(FRn, Rm + R0); | ||
166 | } | ||
167 | |||
168 | return 0; | ||
169 | } | ||
170 | |||
171 | static int | ||
172 | fmov_mem_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
173 | int n) | ||
174 | { | ||
175 | if (FPSCR_SZ) { | ||
176 | FMOV_EXT(n); | ||
177 | READ(FRn, Rm + 4); | ||
178 | n++; | ||
179 | READ(FRn, Rm); | ||
180 | } else { | ||
181 | READ(FRn, Rm); | ||
182 | } | ||
183 | |||
184 | return 0; | ||
185 | } | ||
186 | |||
187 | static int | ||
188 | fmov_inc_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
189 | int n) | ||
190 | { | ||
191 | if (FPSCR_SZ) { | ||
192 | FMOV_EXT(n); | ||
193 | READ(FRn, Rm + 4); | ||
194 | n++; | ||
195 | READ(FRn, Rm); | ||
196 | Rm += 8; | ||
197 | } else { | ||
198 | READ(FRn, Rm); | ||
199 | Rm += 4; | ||
200 | } | ||
201 | |||
202 | return 0; | ||
203 | } | ||
204 | |||
205 | static int | ||
206 | fmov_reg_idx(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
207 | int n) | ||
208 | { | ||
209 | if (FPSCR_SZ) { | ||
210 | FMOV_EXT(m); | ||
211 | WRITE(FRm, Rn + R0 + 4); | ||
212 | m++; | ||
213 | WRITE(FRm, Rn + R0); | ||
214 | } else { | ||
215 | WRITE(FRm, Rn + R0); | ||
216 | } | ||
217 | |||
218 | return 0; | ||
219 | } | ||
220 | |||
221 | static int | ||
222 | fmov_reg_mem(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
223 | int n) | ||
224 | { | ||
225 | if (FPSCR_SZ) { | ||
226 | FMOV_EXT(m); | ||
227 | WRITE(FRm, Rn + 4); | ||
228 | m++; | ||
229 | WRITE(FRm, Rn); | ||
230 | } else { | ||
231 | WRITE(FRm, Rn); | ||
232 | } | ||
233 | |||
234 | return 0; | ||
235 | } | ||
236 | |||
237 | static int | ||
238 | fmov_reg_dec(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
239 | int n) | ||
240 | { | ||
241 | if (FPSCR_SZ) { | ||
242 | FMOV_EXT(m); | ||
243 | Rn -= 8; | ||
244 | WRITE(FRm, Rn + 4); | ||
245 | m++; | ||
246 | WRITE(FRm, Rn); | ||
247 | } else { | ||
248 | Rn -= 4; | ||
249 | WRITE(FRm, Rn); | ||
250 | } | ||
251 | |||
252 | return 0; | ||
253 | } | ||
254 | |||
255 | static int | ||
256 | fmov_reg_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, | ||
257 | int n) | ||
258 | { | ||
259 | if (FPSCR_SZ) { | ||
260 | FMOV_EXT(m); | ||
261 | FMOV_EXT(n); | ||
262 | DRn = DRm; | ||
263 | } else { | ||
264 | FRn = FRm; | ||
265 | } | ||
266 | |||
267 | return 0; | ||
268 | } | ||
269 | |||
270 | static int | ||
271 | fnop_mn(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n) | ||
272 | { | ||
273 | return -EINVAL; | ||
274 | } | ||
275 | |||
276 | // 1 arg instructions. | ||
277 | #define NOTYETn(i) static int i(struct sh_fpu_soft_struct *fregs, int n) \ | ||
278 | { printk( #i " not yet done.\n"); return 0; } | ||
279 | |||
280 | NOTYETn(ftrv) | ||
281 | NOTYETn(fsqrt) | ||
282 | NOTYETn(fipr) | ||
283 | NOTYETn(fsca) | ||
284 | NOTYETn(fsrra) | ||
285 | |||
286 | #define EMU_FLOAT_X(SZ,N) do { \ | ||
287 | FP_DECL_##SZ(Fn); \ | ||
288 | FP_FROM_INT_##SZ(Fn, FPUL, 32, int); \ | ||
289 | PACK_##SZ(N, Fn); }while(0) | ||
290 | static int ffloat(struct sh_fpu_soft_struct *fregs, int n) | ||
291 | { | ||
292 | FP_DECL_EX; | ||
293 | |||
294 | if (FPSCR_PR) | ||
295 | EMU_FLOAT_X(D, DRn); | ||
296 | else | ||
297 | EMU_FLOAT_X(S, FRn); | ||
298 | |||
299 | return 0; | ||
300 | } | ||
301 | |||
302 | #define EMU_FTRC_X(SZ,N) do { \ | ||
303 | FP_DECL_##SZ(Fn); \ | ||
304 | UNPACK_##SZ(Fn, N); \ | ||
305 | FP_TO_INT_##SZ(FPUL, Fn, 32, 1); }while(0) | ||
306 | static int ftrc(struct sh_fpu_soft_struct *fregs, int n) | ||
307 | { | ||
308 | FP_DECL_EX; | ||
309 | |||
310 | if (FPSCR_PR) | ||
311 | EMU_FTRC_X(D, DRn); | ||
312 | else | ||
313 | EMU_FTRC_X(S, FRn); | ||
314 | |||
315 | return 0; | ||
316 | } | ||
317 | |||
318 | static int fcnvsd(struct sh_fpu_soft_struct *fregs, int n) | ||
319 | { | ||
320 | FP_DECL_EX; | ||
321 | FP_DECL_S(Fn); | ||
322 | FP_DECL_D(Fr); | ||
323 | UNPACK_S(Fn, FPUL); | ||
324 | FP_CONV(D, S, 2, 1, Fr, Fn); | ||
325 | PACK_D(DRn, Fr); | ||
326 | return 0; | ||
327 | } | ||
328 | |||
329 | static int fcnvds(struct sh_fpu_soft_struct *fregs, int n) | ||
330 | { | ||
331 | FP_DECL_EX; | ||
332 | FP_DECL_D(Fn); | ||
333 | FP_DECL_S(Fr); | ||
334 | UNPACK_D(Fn, DRn); | ||
335 | FP_CONV(S, D, 1, 2, Fr, Fn); | ||
336 | PACK_S(FPUL, Fr); | ||
337 | return 0; | ||
338 | } | ||
339 | |||
340 | static int fxchg(struct sh_fpu_soft_struct *fregs, int flag) | ||
341 | { | ||
342 | FPSCR ^= flag; | ||
343 | return 0; | ||
344 | } | ||
345 | |||
346 | static int fsts(struct sh_fpu_soft_struct *fregs, int n) | ||
347 | { | ||
348 | FRn = FPUL; | ||
349 | return 0; | ||
350 | } | ||
351 | |||
352 | static int flds(struct sh_fpu_soft_struct *fregs, int n) | ||
353 | { | ||
354 | FPUL = FRn; | ||
355 | return 0; | ||
356 | } | ||
357 | |||
358 | static int fneg(struct sh_fpu_soft_struct *fregs, int n) | ||
359 | { | ||
360 | FRn ^= (1 << (_FP_W_TYPE_SIZE - 1)); | ||
361 | return 0; | ||
362 | } | ||
363 | |||
364 | static int fabs(struct sh_fpu_soft_struct *fregs, int n) | ||
365 | { | ||
366 | FRn &= ~(1 << (_FP_W_TYPE_SIZE - 1)); | ||
367 | return 0; | ||
368 | } | ||
369 | |||
370 | static int fld0(struct sh_fpu_soft_struct *fregs, int n) | ||
371 | { | ||
372 | FRn = 0; | ||
373 | return 0; | ||
374 | } | ||
375 | |||
376 | static int fld1(struct sh_fpu_soft_struct *fregs, int n) | ||
377 | { | ||
378 | FRn = (_FP_EXPBIAS_S << (_FP_FRACBITS_S - 1)); | ||
379 | return 0; | ||
380 | } | ||
381 | |||
382 | static int fnop_n(struct sh_fpu_soft_struct *fregs, int n) | ||
383 | { | ||
384 | return -EINVAL; | ||
385 | } | ||
386 | |||
387 | /// Instruction decoders. | ||
388 | |||
389 | static int id_fxfd(struct sh_fpu_soft_struct *, int); | ||
390 | static int id_fnxd(struct sh_fpu_soft_struct *, struct pt_regs *, int, int); | ||
391 | |||
392 | static int (*fnxd[])(struct sh_fpu_soft_struct *, int) = { | ||
393 | fsts, flds, ffloat, ftrc, fneg, fabs, fsqrt, fsrra, | ||
394 | fld0, fld1, fcnvsd, fcnvds, fnop_n, fnop_n, fipr, id_fxfd | ||
395 | }; | ||
396 | |||
397 | static int (*fnmx[])(struct sh_fpu_soft_struct *, struct pt_regs *, int, int) = { | ||
398 | fadd, fsub, fmul, fdiv, fcmp_eq, fcmp_gt, fmov_idx_reg, fmov_reg_idx, | ||
399 | fmov_mem_reg, fmov_inc_reg, fmov_reg_mem, fmov_reg_dec, | ||
400 | fmov_reg_reg, id_fnxd, fmac, fnop_mn}; | ||
401 | |||
402 | static int id_fxfd(struct sh_fpu_soft_struct *fregs, int x) | ||
403 | { | ||
404 | const int flag[] = { FPSCR_SZ, FPSCR_PR, FPSCR_FR, 0 }; | ||
405 | switch (x & 3) { | ||
406 | case 3: | ||
407 | fxchg(fregs, flag[x >> 2]); | ||
408 | break; | ||
409 | case 1: | ||
410 | ftrv(fregs, x - 1); | ||
411 | break; | ||
412 | default: | ||
413 | fsca(fregs, x); | ||
414 | } | ||
415 | return 0; | ||
416 | } | ||
417 | |||
418 | static int | ||
419 | id_fnxd(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int x, int n) | ||
420 | { | ||
421 | return (fnxd[x])(fregs, n); | ||
422 | } | ||
423 | |||
424 | static int | ||
425 | id_fnmx(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, u16 code) | ||
426 | { | ||
427 | int n = (code >> 8) & 0xf, m = (code >> 4) & 0xf, x = code & 0xf; | ||
428 | return (fnmx[x])(fregs, regs, m, n); | ||
429 | } | ||
430 | |||
431 | static int | ||
432 | id_sys(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, u16 code) | ||
433 | { | ||
434 | int n = ((code >> 8) & 0xf); | ||
435 | unsigned long *reg = (code & 0x0010) ? &FPUL : &FPSCR; | ||
436 | |||
437 | switch (code & 0xf0ff) { | ||
438 | case 0x005a: | ||
439 | case 0x006a: | ||
440 | Rn = *reg; | ||
441 | break; | ||
442 | case 0x405a: | ||
443 | case 0x406a: | ||
444 | *reg = Rn; | ||
445 | break; | ||
446 | case 0x4052: | ||
447 | case 0x4062: | ||
448 | Rn -= 4; | ||
449 | WRITE(*reg, Rn); | ||
450 | break; | ||
451 | case 0x4056: | ||
452 | case 0x4066: | ||
453 | READ(*reg, Rn); | ||
454 | Rn += 4; | ||
455 | break; | ||
456 | default: | ||
457 | return -EINVAL; | ||
458 | } | ||
459 | |||
460 | return 0; | ||
461 | } | ||
462 | |||
463 | static int fpu_emulate(u16 code, struct sh_fpu_soft_struct *fregs, struct pt_regs *regs) | ||
464 | { | ||
465 | if ((code & 0xf000) == 0xf000) | ||
466 | return id_fnmx(fregs, regs, code); | ||
467 | else | ||
468 | return id_sys(fregs, regs, code); | ||
469 | } | ||
470 | |||
471 | /** | ||
472 | * denormal_to_double - Given denormalized float number, | ||
473 | * store double float | ||
474 | * | ||
475 | * @fpu: Pointer to sh_fpu_hard structure | ||
476 | * @n: Index to FP register | ||
477 | */ | ||
478 | static void denormal_to_double(struct sh_fpu_hard_struct *fpu, int n) | ||
479 | { | ||
480 | unsigned long du, dl; | ||
481 | unsigned long x = fpu->fpul; | ||
482 | int exp = 1023 - 126; | ||
483 | |||
484 | if (x != 0 && (x & 0x7f800000) == 0) { | ||
485 | du = (x & 0x80000000); | ||
486 | while ((x & 0x00800000) == 0) { | ||
487 | x <<= 1; | ||
488 | exp--; | ||
489 | } | ||
490 | x &= 0x007fffff; | ||
491 | du |= (exp << 20) | (x >> 3); | ||
492 | dl = x << 29; | ||
493 | |||
494 | fpu->fp_regs[n] = du; | ||
495 | fpu->fp_regs[n+1] = dl; | ||
496 | } | ||
497 | } | ||
498 | |||
499 | /** | ||
500 | * ieee_fpe_handler - Handle denormalized number exception | ||
501 | * | ||
502 | * @regs: Pointer to register structure | ||
503 | * | ||
504 | * Returns 1 when it's handled (should not cause exception). | ||
505 | */ | ||
506 | static int ieee_fpe_handler(struct pt_regs *regs) | ||
507 | { | ||
508 | unsigned short insn = *(unsigned short *)regs->pc; | ||
509 | unsigned short finsn; | ||
510 | unsigned long nextpc; | ||
511 | int nib[4] = { | ||
512 | (insn >> 12) & 0xf, | ||
513 | (insn >> 8) & 0xf, | ||
514 | (insn >> 4) & 0xf, | ||
515 | insn & 0xf}; | ||
516 | |||
517 | if (nib[0] == 0xb || | ||
518 | (nib[0] == 0x4 && nib[2] == 0x0 && nib[3] == 0xb)) /* bsr & jsr */ | ||
519 | regs->pr = regs->pc + 4; | ||
520 | |||
521 | if (nib[0] == 0xa || nib[0] == 0xb) { /* bra & bsr */ | ||
522 | nextpc = regs->pc + 4 + ((short) ((insn & 0xfff) << 4) >> 3); | ||
523 | finsn = *(unsigned short *) (regs->pc + 2); | ||
524 | } else if (nib[0] == 0x8 && nib[1] == 0xd) { /* bt/s */ | ||
525 | if (regs->sr & 1) | ||
526 | nextpc = regs->pc + 4 + ((char) (insn & 0xff) << 1); | ||
527 | else | ||
528 | nextpc = regs->pc + 4; | ||
529 | finsn = *(unsigned short *) (regs->pc + 2); | ||
530 | } else if (nib[0] == 0x8 && nib[1] == 0xf) { /* bf/s */ | ||
531 | if (regs->sr & 1) | ||
532 | nextpc = regs->pc + 4; | ||
533 | else | ||
534 | nextpc = regs->pc + 4 + ((char) (insn & 0xff) << 1); | ||
535 | finsn = *(unsigned short *) (regs->pc + 2); | ||
536 | } else if (nib[0] == 0x4 && nib[3] == 0xb && | ||
537 | (nib[2] == 0x0 || nib[2] == 0x2)) { /* jmp & jsr */ | ||
538 | nextpc = regs->regs[nib[1]]; | ||
539 | finsn = *(unsigned short *) (regs->pc + 2); | ||
540 | } else if (nib[0] == 0x0 && nib[3] == 0x3 && | ||
541 | (nib[2] == 0x0 || nib[2] == 0x2)) { /* braf & bsrf */ | ||
542 | nextpc = regs->pc + 4 + regs->regs[nib[1]]; | ||
543 | finsn = *(unsigned short *) (regs->pc + 2); | ||
544 | } else if (insn == 0x000b) { /* rts */ | ||
545 | nextpc = regs->pr; | ||
546 | finsn = *(unsigned short *) (regs->pc + 2); | ||
547 | } else { | ||
548 | nextpc = regs->pc + 2; | ||
549 | finsn = insn; | ||
550 | } | ||
551 | |||
552 | if ((finsn & 0xf1ff) == 0xf0ad) { /* fcnvsd */ | ||
553 | struct task_struct *tsk = current; | ||
554 | |||
555 | if ((tsk->thread.fpu.hard.fpscr & (1 << 17))) { | ||
556 | /* FPU error */ | ||
557 | denormal_to_double (&tsk->thread.fpu.hard, | ||
558 | (finsn >> 8) & 0xf); | ||
559 | tsk->thread.fpu.hard.fpscr &= | ||
560 | ~(FPSCR_CAUSE_MASK | FPSCR_FLAG_MASK); | ||
561 | set_tsk_thread_flag(tsk, TIF_USEDFPU); | ||
562 | } else { | ||
563 | tsk->thread.trap_no = 11; | ||
564 | tsk->thread.error_code = 0; | ||
565 | force_sig(SIGFPE, tsk); | ||
566 | } | ||
567 | |||
568 | regs->pc = nextpc; | ||
569 | return 1; | ||
570 | } | ||
571 | |||
572 | return 0; | ||
573 | } | ||
574 | |||
575 | asmlinkage void do_fpu_error(unsigned long r4, unsigned long r5, | ||
576 | unsigned long r6, unsigned long r7, | ||
577 | struct pt_regs regs) | ||
578 | { | ||
579 | struct task_struct *tsk = current; | ||
580 | |||
581 | if (ieee_fpe_handler (®s)) | ||
582 | return; | ||
583 | |||
584 | regs.pc += 2; | ||
585 | tsk->thread.trap_no = 11; | ||
586 | tsk->thread.error_code = 0; | ||
587 | force_sig(SIGFPE, tsk); | ||
588 | } | ||
589 | |||
590 | /** | ||
591 | * fpu_init - Initialize FPU registers | ||
592 | * @fpu: Pointer to software emulated FPU registers. | ||
593 | */ | ||
594 | static void fpu_init(struct sh_fpu_soft_struct *fpu) | ||
595 | { | ||
596 | int i; | ||
597 | |||
598 | fpu->fpscr = FPSCR_INIT; | ||
599 | fpu->fpul = 0; | ||
600 | |||
601 | for (i = 0; i < 16; i++) { | ||
602 | fpu->fp_regs[i] = 0; | ||
603 | fpu->xfp_regs[i]= 0; | ||
604 | } | ||
605 | } | ||
606 | |||
607 | /** | ||
608 | * do_fpu_inst - Handle reserved instructions for FPU emulation | ||
609 | * @inst: instruction code. | ||
610 | * @regs: registers on stack. | ||
611 | */ | ||
612 | int do_fpu_inst(unsigned short inst, struct pt_regs *regs) | ||
613 | { | ||
614 | struct task_struct *tsk = current; | ||
615 | struct sh_fpu_soft_struct *fpu = &(tsk->thread.fpu.soft); | ||
616 | |||
617 | if (!test_tsk_thread_flag(tsk, TIF_USEDFPU)) { | ||
618 | /* initialize once. */ | ||
619 | fpu_init(fpu); | ||
620 | set_tsk_thread_flag(tsk, TIF_USEDFPU); | ||
621 | } | ||
622 | |||
623 | return fpu_emulate(inst, fpu, regs); | ||
624 | } | ||
diff --git a/arch/sh/math-emu/sfp-util.h b/arch/sh/math-emu/sfp-util.h new file mode 100644 index 000000000000..8ae1bd310ad0 --- /dev/null +++ b/arch/sh/math-emu/sfp-util.h | |||
@@ -0,0 +1,72 @@ | |||
1 | /* | ||
2 | * These are copied from glibc/stdlib/longlong.h | ||
3 | */ | ||
4 | |||
5 | #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | ||
6 | do { \ | ||
7 | UWtype __x; \ | ||
8 | __x = (al) + (bl); \ | ||
9 | (sh) = (ah) + (bh) + (__x < (al)); \ | ||
10 | (sl) = __x; \ | ||
11 | } while (0) | ||
12 | |||
13 | #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | ||
14 | do { \ | ||
15 | UWtype __x; \ | ||
16 | __x = (al) - (bl); \ | ||
17 | (sh) = (ah) - (bh) - (__x > (al)); \ | ||
18 | (sl) = __x; \ | ||
19 | } while (0) | ||
20 | |||
21 | #define umul_ppmm(w1, w0, u, v) \ | ||
22 | __asm__ ("dmulu.l %2,%3\n\tsts macl,%1\n\tsts mach,%0" \ | ||
23 | : "=r" ((u32)(w1)), "=r" ((u32)(w0)) \ | ||
24 | : "r" ((u32)(u)), "r" ((u32)(v)) \ | ||
25 | : "macl", "mach") | ||
26 | |||
27 | #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2)) | ||
28 | #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1)) | ||
29 | #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2)) | ||
30 | |||
31 | #define udiv_qrnnd(q, r, n1, n0, d) \ | ||
32 | do { \ | ||
33 | UWtype __d1, __d0, __q1, __q0; \ | ||
34 | UWtype __r1, __r0, __m; \ | ||
35 | __d1 = __ll_highpart (d); \ | ||
36 | __d0 = __ll_lowpart (d); \ | ||
37 | \ | ||
38 | __r1 = (n1) % __d1; \ | ||
39 | __q1 = (n1) / __d1; \ | ||
40 | __m = (UWtype) __q1 * __d0; \ | ||
41 | __r1 = __r1 * __ll_B | __ll_highpart (n0); \ | ||
42 | if (__r1 < __m) \ | ||
43 | { \ | ||
44 | __q1--, __r1 += (d); \ | ||
45 | if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\ | ||
46 | if (__r1 < __m) \ | ||
47 | __q1--, __r1 += (d); \ | ||
48 | } \ | ||
49 | __r1 -= __m; \ | ||
50 | \ | ||
51 | __r0 = __r1 % __d1; \ | ||
52 | __q0 = __r1 / __d1; \ | ||
53 | __m = (UWtype) __q0 * __d0; \ | ||
54 | __r0 = __r0 * __ll_B | __ll_lowpart (n0); \ | ||
55 | if (__r0 < __m) \ | ||
56 | { \ | ||
57 | __q0--, __r0 += (d); \ | ||
58 | if (__r0 >= (d)) \ | ||
59 | if (__r0 < __m) \ | ||
60 | __q0--, __r0 += (d); \ | ||
61 | } \ | ||
62 | __r0 -= __m; \ | ||
63 | \ | ||
64 | (q) = (UWtype) __q1 * __ll_B | __q0; \ | ||
65 | (r) = __r0; \ | ||
66 | } while (0) | ||
67 | |||
68 | #define abort() return 0 | ||
69 | |||
70 | #define __BYTE_ORDER __LITTLE_ENDIAN | ||
71 | |||
72 | |||