sh: math-emu support
authorTakashi YOSHII <takasi-y@ops.dti.ne.jp>
Wed, 27 Sep 2006 08:15:32 +0000 (17:15 +0900)
committerPaul Mundt <lethal@linux-sh.org>
Wed, 27 Sep 2006 08:15:32 +0000 (17:15 +0900)
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>
arch/sh/Kconfig
arch/sh/Makefile
arch/sh/kernel/traps.c
arch/sh/math-emu/Makefile [new file with mode: 0644]
arch/sh/math-emu/math.c [new file with mode: 0644]
arch/sh/math-emu/sfp-util.h [new file with mode: 0644]
include/asm-sh/sfp-machine.h [new file with mode: 0644]

index d1d7244..07d6d69 100644 (file)
@@ -339,6 +339,15 @@ config SH_FPU
 
          This option must be set in order to enable the FPU.
 
+config SH_FPU_EMU
+       bool "FPU emulation support"
+       depends on !SH_FPU && EXPERIMENTAL
+       default n
+       help
+         Selecting this option will enable support for software FPU emulation.
+         Most SH-3 users will want to say Y here, whereas most SH-4 users will
+         want to say N.
+
 config SH_DSP
        bool "DSP support"
        depends on !CPU_SH4
index 859285d..13b688b 100644 (file)
@@ -79,6 +79,7 @@ head-y := arch/sh/kernel/head.o arch/sh/kernel/init_task.o
 LIBGCC := $(shell $(CC) $(CFLAGS) -print-libgcc-file-name)
 
 core-y                         += arch/sh/kernel/ arch/sh/mm/
+core-$(CONFIG_SH_FPU_EMU)      += arch/sh/math-emu/
 
 # Boards
 machdir-$(CONFIG_SH_SOLUTION_ENGINE)           := se/770x
index 21bef1b..95c810b 100644 (file)
 
 #ifdef CONFIG_SH_KGDB
 #include <asm/kgdb.h>
-#define CHK_REMOTE_DEBUG(regs)                                               \
-{                                                                            \
-  if ((kgdb_debug_hook != (kgdb_debug_hook_t *) NULL) && (!user_mode(regs))) \
-  {                                                                          \
-    (*kgdb_debug_hook)(regs);                                                \
-  }                                                                          \
+#define CHK_REMOTE_DEBUG(regs)                         \
+{                                              \
+       if (kgdb_debug_hook && !user_mode(regs))\
+               (*kgdb_debug_hook)(regs);       \
 }
 #else
 #define CHK_REMOTE_DEBUG(regs)
 #endif
 
-#define DO_ERROR(trapnr, signr, str, name, tsk)                                \
-asmlinkage void do_##name(unsigned long r4, unsigned long r5,          \
-                         unsigned long r6, unsigned long r7,           \
-                         struct pt_regs regs)                          \
-{                                                                      \
-       unsigned long error_code;                                       \
-                                                                       \
-       /* Check if it's a DSP instruction */                           \
-       if (is_dsp_inst(&regs)) {                                       \
-               /* Enable DSP mode, and restart instruction. */         \
-               regs.sr |= SR_DSP;                                      \
-               return;                                                 \
-       }                                                               \
-                                                                       \
-       asm volatile("stc       r2_bank, %0": "=r" (error_code));       \
-       local_irq_enable();                                             \
-       tsk->thread.error_code = error_code;                            \
-       tsk->thread.trap_no = trapnr;                                   \
-        CHK_REMOTE_DEBUG(&regs);                                       \
-       force_sig(signr, tsk);                                          \
-       die_if_no_fixup(str,&regs,error_code);                          \
-}
-
 #ifdef CONFIG_CPU_SH2
 #define TRAP_RESERVED_INST     4
 #define TRAP_ILLEGAL_SLOT_INST 6
@@ -575,8 +550,117 @@ int is_dsp_inst(struct pt_regs *regs)
 #define is_dsp_inst(regs)      (0)
 #endif /* CONFIG_SH_DSP */
 
-DO_ERROR(TRAP_RESERVED_INST, SIGILL, "reserved instruction", reserved_inst, current)
-DO_ERROR(TRAP_ILLEGAL_SLOT_INST, SIGILL, "illegal slot instruction", illegal_slot_inst, current)
+extern int do_fpu_inst(unsigned short, struct pt_regs*);
+
+asmlinkage void do_reserved_inst(unsigned long r4, unsigned long r5,
+                               unsigned long r6, unsigned long r7,
+                               struct pt_regs regs)
+{
+       unsigned long error_code;
+       struct task_struct *tsk = current;
+
+#ifdef CONFIG_SH_FPU_EMU
+       unsigned short inst;
+       int err;
+
+       get_user(inst, (unsigned short*)regs.pc);
+
+       err = do_fpu_inst(inst, &regs);
+       if (!err) {
+               regs.pc += 2;
+               return;
+       }
+       /* not a FPU inst. */
+#endif
+
+#ifdef CONFIG_SH_DSP
+       /* Check if it's a DSP instruction */
+       if (is_dsp_inst(&regs)) {
+               /* Enable DSP mode, and restart instruction. */
+               regs.sr |= SR_DSP;
+               return;
+       }
+#endif
+
+       asm volatile("stc       r2_bank, %0": "=r" (error_code));
+       local_irq_enable();
+       tsk->thread.error_code = error_code;
+       tsk->thread.trap_no = TRAP_RESERVED_INST;
+       CHK_REMOTE_DEBUG(&regs);
+       force_sig(SIGILL, tsk);
+       die_if_no_fixup("reserved instruction", &regs, error_code);
+}
+
+#ifdef CONFIG_SH_FPU_EMU
+static int emulate_branch(unsigned short inst, struct pt_regs* regs)
+{
+       /*
+        * bfs: 8fxx: PC+=d*2+4;
+        * bts: 8dxx: PC+=d*2+4;
+        * bra: axxx: PC+=D*2+4;
+        * bsr: bxxx: PC+=D*2+4  after PR=PC+4;
+        * braf:0x23: PC+=Rn*2+4;
+        * bsrf:0x03: PC+=Rn*2+4 after PR=PC+4;
+        * jmp: 4x2b: PC=Rn;
+        * jsr: 4x0b: PC=Rn      after PR=PC+4;
+        * rts: 000b: PC=PR;
+        */
+       if ((inst & 0xfd00) == 0x8d00) {
+               regs->pc += SH_PC_8BIT_OFFSET(inst);
+               return 0;
+       }
+
+       if ((inst & 0xe000) == 0xa000) {
+               regs->pc += SH_PC_12BIT_OFFSET(inst);
+               return 0;
+       }
+
+       if ((inst & 0xf0df) == 0x0003) {
+               regs->pc += regs->regs[(inst & 0x0f00) >> 8] + 4;
+               return 0;
+       }
+
+       if ((inst & 0xf0df) == 0x400b) {
+               regs->pc = regs->regs[(inst & 0x0f00) >> 8];
+               return 0;
+       }
+
+       if ((inst & 0xffff) == 0x000b) {
+               regs->pc = regs->pr;
+               return 0;
+       }
+
+       return 1;
+}
+#endif
+
+asmlinkage void do_illegal_slot_inst(unsigned long r4, unsigned long r5,
+                               unsigned long r6, unsigned long r7,
+                               struct pt_regs regs)
+{
+       unsigned long error_code;
+       struct task_struct *tsk = current;
+#ifdef CONFIG_SH_FPU_EMU
+       unsigned short inst;
+
+       get_user(inst, (unsigned short *)regs.pc + 1);
+       if (!do_fpu_inst(inst, &regs)) {
+               get_user(inst, (unsigned short *)regs.pc);
+               if (!emulate_branch(inst, &regs))
+                       return;
+               /* fault in branch.*/
+       }
+       /* not a FPU inst. */
+#endif
+
+       asm volatile("stc       r2_bank, %0": "=r" (error_code));
+       local_irq_enable();
+       tsk->thread.error_code = error_code;
+       tsk->thread.trap_no = TRAP_RESERVED_INST;
+       CHK_REMOTE_DEBUG(&regs);
+       force_sig(SIGILL, tsk);
+       die_if_no_fixup("illegal slot instruction", &regs, error_code);
+}
 
 asmlinkage void do_exception_error(unsigned long r4, unsigned long r5,
                                   unsigned long r6, unsigned long r7,
@@ -634,14 +718,16 @@ void __init trap_init(void)
        exception_handling_table[TRAP_ILLEGAL_SLOT_INST]
                = (void *)do_illegal_slot_inst;
 
-#ifdef CONFIG_CPU_SH4
-       if (!(cpu_data->flags & CPU_HAS_FPU)) {
-               /* For SH-4 lacking an FPU, treat floating point instructions
-                  as reserved. */
-               /* entry 64 corresponds to EXPEVT=0x800 */
-               exception_handling_table[64] = (void *)do_reserved_inst;
-               exception_handling_table[65] = (void *)do_illegal_slot_inst;
-       }
+#if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SH_FPU) || \
+    defined(CONFIG_SH_FPU_EMU)
+       /*
+        * For SH-4 lacking an FPU, treat floating point instructions as
+        * reserved. They'll be handled in the math-emu case, or faulted on
+        * otherwise.
+        */
+       /* entry 64 corresponds to EXPEVT=0x800 */
+       exception_handling_table[64] = (void *)do_reserved_inst;
+       exception_handling_table[65] = (void *)do_illegal_slot_inst;
 #endif
                
        /* Setup VBR for boot cpu */
diff --git a/arch/sh/math-emu/Makefile b/arch/sh/math-emu/Makefile
new file mode 100644 (file)
index 0000000..638b342
--- /dev/null
@@ -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 (file)
index 0000000..26b6046
--- /dev/null
@@ -0,0 +1,624 @@
+/*
+ * arch/sh/math-emu/math.c
+ *
+ * Copyright (C) 2006 Takashi YOSHII <takasi-y@ops.dti.ne.jp>
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/types.h>
+#include <linux/sched.h>
+#include <linux/signal.h>
+
+#include <asm/system.h>
+#include <asm/uaccess.h>
+#include <asm/processor.h>
+#include <asm/io.h>
+
+#include "sfp-util.h"
+#include <math-emu/soft-fp.h>
+#include <math-emu/single.h>
+#include <math-emu/double.h>
+
+#define        FPUL            (fregs->fpul)
+#define FPSCR          (fregs->fpscr)
+#define FPSCR_RM       (FPSCR&3)
+#define FPSCR_DN       ((FPSCR>>18)&1)
+#define FPSCR_PR       ((FPSCR>>19)&1)
+#define FPSCR_SZ       ((FPSCR>>20)&1)
+#define FPSCR_FR       ((FPSCR>>21)&1)
+#define FPSCR_MASK     0x003fffffUL
+
+#define BANK(n)        (n^(FPSCR_FR?16:0))
+#define FR     ((unsigned long*)(fregs->fp_regs))
+#define FR0    (FR[BANK(0)])
+#define FRn    (FR[BANK(n)])
+#define FRm    (FR[BANK(m)])
+#define DR     ((unsigned long long*)(fregs->fp_regs))
+#define DRn    (DR[BANK(n)/2])
+#define DRm    (DR[BANK(m)/2])
+
+#define XREG(n)        (n^16)
+#define XFn    (FR[BANK(XREG(n))])
+#define XFm    (FR[BANK(XREG(m))])
+#define XDn    (DR[BANK(XREG(n))/2])
+#define XDm    (DR[BANK(XREG(m))/2])
+
+#define R0     (regs->regs[0])
+#define Rn     (regs->regs[n])
+#define Rm     (regs->regs[m])
+
+#define WRITE(d,a)     ({if(put_user(d, (typeof (d)*)a)) return -EFAULT;})
+#define READ(d,a)      ({if(get_user(d, (typeof (d)*)a)) return -EFAULT;})
+
+#define PACK_S(r,f)    FP_PACK_SP(&r,f)
+#define UNPACK_S(f,r)  FP_UNPACK_SP(f,&r)
+#define PACK_D(r,f) \
+       {u32 t[2]; FP_PACK_DP(t,f); ((u32*)&r)[0]=t[1]; ((u32*)&r)[1]=t[0];}
+#define UNPACK_D(f,r) \
+       {u32 t[2]; t[0]=((u32*)&r)[1]; t[1]=((u32*)&r)[0]; FP_UNPACK_DP(f,t);}
+
+// 2 args instructions.
+#define BOTH_PRmn(op,x) \
+       FP_DECL_EX; if(FPSCR_PR) op(D,x,DRm,DRn); else op(S,x,FRm,FRn);
+
+#define CMP_X(SZ,R,M,N) do{ \
+       FP_DECL_##SZ(Fm); FP_DECL_##SZ(Fn); \
+       UNPACK_##SZ(Fm, M); UNPACK_##SZ(Fn, N); \
+       FP_CMP_##SZ(R, Fn, Fm, 2); }while(0)
+#define EQ_X(SZ,R,M,N) do{ \
+       FP_DECL_##SZ(Fm); FP_DECL_##SZ(Fn); \
+       UNPACK_##SZ(Fm, M); UNPACK_##SZ(Fn, N); \
+       FP_CMP_EQ_##SZ(R, Fn, Fm); }while(0)
+#define CMP(OP) ({ int r; BOTH_PRmn(OP##_X,r); r; })
+
+static int
+fcmp_gt(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       if (CMP(CMP) > 0)
+               regs->sr |= 1;
+       else
+               regs->sr &= ~1;
+
+       return 0;
+}
+
+static int
+fcmp_eq(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       if (CMP(CMP /*EQ*/) == 0)
+               regs->sr |= 1;
+       else
+               regs->sr &= ~1;
+       return 0;
+}
+
+#define ARITH_X(SZ,OP,M,N) do{ \
+       FP_DECL_##SZ(Fm); FP_DECL_##SZ(Fn); FP_DECL_##SZ(Fr); \
+       UNPACK_##SZ(Fm, M); UNPACK_##SZ(Fn, N); \
+       FP_##OP##_##SZ(Fr, Fn, Fm); \
+       PACK_##SZ(N, Fr); }while(0)
+
+static int
+fadd(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       BOTH_PRmn(ARITH_X, ADD);
+       return 0;
+}
+
+static int
+fsub(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       BOTH_PRmn(ARITH_X, SUB);
+       return 0;
+}
+
+static int
+fmul(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       BOTH_PRmn(ARITH_X, MUL);
+       return 0;
+}
+
+static int
+fdiv(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       BOTH_PRmn(ARITH_X, DIV);
+       return 0;
+}
+
+static int
+fmac(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       FP_DECL_EX;
+       FP_DECL_S(Fr);
+       FP_DECL_S(Ft);
+       FP_DECL_S(F0);
+       FP_DECL_S(Fm);
+       FP_DECL_S(Fn);
+       UNPACK_S(F0, FR0);
+       UNPACK_S(Fm, FRm);
+       UNPACK_S(Fn, FRn);
+       FP_MUL_S(Ft, Fm, F0);
+       FP_ADD_S(Fr, Fn, Ft);
+       PACK_S(FRn, Fr);
+       return 0;
+}
+
+// to process fmov's extention (odd n for DR access XD).
+#define FMOV_EXT(x) if(x&1) x+=16-1
+
+static int
+fmov_idx_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(n);
+               READ(FRn, Rm + R0 + 4);
+               n++;
+               READ(FRn, Rm + R0);
+       } else {
+               READ(FRn, Rm + R0);
+       }
+
+       return 0;
+}
+
+static int
+fmov_mem_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(n);
+               READ(FRn, Rm + 4);
+               n++;
+               READ(FRn, Rm);
+       } else {
+               READ(FRn, Rm);
+       }
+
+       return 0;
+}
+
+static int
+fmov_inc_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(n);
+               READ(FRn, Rm + 4);
+               n++;
+               READ(FRn, Rm);
+               Rm += 8;
+       } else {
+               READ(FRn, Rm);
+               Rm += 4;
+       }
+
+       return 0;
+}
+
+static int
+fmov_reg_idx(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(m);
+               WRITE(FRm, Rn + R0 + 4);
+               m++;
+               WRITE(FRm, Rn + R0);
+       } else {
+               WRITE(FRm, Rn + R0);
+       }
+
+       return 0;
+}
+
+static int
+fmov_reg_mem(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(m);
+               WRITE(FRm, Rn + 4);
+               m++;
+               WRITE(FRm, Rn);
+       } else {
+               WRITE(FRm, Rn);
+       }
+
+       return 0;
+}
+
+static int
+fmov_reg_dec(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(m);
+               Rn -= 8;
+               WRITE(FRm, Rn + 4);
+               m++;
+               WRITE(FRm, Rn);
+       } else {
+               Rn -= 4;
+               WRITE(FRm, Rn);
+       }
+
+       return 0;
+}
+
+static int
+fmov_reg_reg(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m,
+            int n)
+{
+       if (FPSCR_SZ) {
+               FMOV_EXT(m);
+               FMOV_EXT(n);
+               DRn = DRm;
+       } else {
+               FRn = FRm;
+       }
+
+       return 0;
+}
+
+static int
+fnop_mn(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int m, int n)
+{
+       return -EINVAL;
+}
+
+// 1 arg instructions.
+#define NOTYETn(i) static int i(struct sh_fpu_soft_struct *fregs, int n) \
+       { printk( #i " not yet done.\n"); return 0; }
+
+NOTYETn(ftrv)
+NOTYETn(fsqrt)
+NOTYETn(fipr)
+NOTYETn(fsca)
+NOTYETn(fsrra)
+
+#define EMU_FLOAT_X(SZ,N) do { \
+       FP_DECL_##SZ(Fn); \
+       FP_FROM_INT_##SZ(Fn, FPUL, 32, int); \
+       PACK_##SZ(N, Fn); }while(0)
+static int ffloat(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FP_DECL_EX;
+
+       if (FPSCR_PR)
+               EMU_FLOAT_X(D, DRn);
+       else
+               EMU_FLOAT_X(S, FRn);
+
+       return 0;
+}
+
+#define EMU_FTRC_X(SZ,N) do { \
+       FP_DECL_##SZ(Fn); \
+       UNPACK_##SZ(Fn, N); \
+       FP_TO_INT_##SZ(FPUL, Fn, 32, 1); }while(0)
+static int ftrc(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FP_DECL_EX;
+
+       if (FPSCR_PR)
+               EMU_FTRC_X(D, DRn);
+       else
+               EMU_FTRC_X(S, FRn);
+
+       return 0;
+}
+
+static int fcnvsd(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FP_DECL_EX;
+       FP_DECL_S(Fn);
+       FP_DECL_D(Fr);
+       UNPACK_S(Fn, FPUL);
+       FP_CONV(D, S, 2, 1, Fr, Fn);
+       PACK_D(DRn, Fr);
+       return 0;
+}
+
+static int fcnvds(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FP_DECL_EX;
+       FP_DECL_D(Fn);
+       FP_DECL_S(Fr);
+       UNPACK_D(Fn, DRn);
+       FP_CONV(S, D, 1, 2, Fr, Fn);
+       PACK_S(FPUL, Fr);
+       return 0;
+}
+
+static int fxchg(struct sh_fpu_soft_struct *fregs, int flag)
+{
+       FPSCR ^= flag;
+       return 0;
+}
+
+static int fsts(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FRn = FPUL;
+       return 0;
+}
+
+static int flds(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FPUL = FRn;
+       return 0;
+}
+
+static int fneg(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FRn ^= (1 << (_FP_W_TYPE_SIZE - 1));
+       return 0;
+}
+
+static int fabs(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FRn &= ~(1 << (_FP_W_TYPE_SIZE - 1));
+       return 0;
+}
+
+static int fld0(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FRn = 0;
+       return 0;
+}
+
+static int fld1(struct sh_fpu_soft_struct *fregs, int n)
+{
+       FRn = (_FP_EXPBIAS_S << (_FP_FRACBITS_S - 1));
+       return 0;
+}
+
+static int fnop_n(struct sh_fpu_soft_struct *fregs, int n)
+{
+       return -EINVAL;
+}
+
+/// Instruction decoders.
+
+static int id_fxfd(struct sh_fpu_soft_struct *, int);
+static int id_fnxd(struct sh_fpu_soft_struct *, struct pt_regs *, int, int);
+
+static int (*fnxd[])(struct sh_fpu_soft_struct *, int) = {
+       fsts, flds, ffloat, ftrc, fneg, fabs, fsqrt, fsrra,
+       fld0, fld1, fcnvsd, fcnvds, fnop_n, fnop_n, fipr, id_fxfd
+};
+
+static int (*fnmx[])(struct sh_fpu_soft_struct *, struct pt_regs *, int, int) = {
+       fadd, fsub, fmul, fdiv, fcmp_eq, fcmp_gt, fmov_idx_reg, fmov_reg_idx,
+       fmov_mem_reg, fmov_inc_reg, fmov_reg_mem, fmov_reg_dec,
+       fmov_reg_reg, id_fnxd, fmac, fnop_mn};
+
+static int id_fxfd(struct sh_fpu_soft_struct *fregs, int x)
+{
+       const int flag[] = { FPSCR_SZ, FPSCR_PR, FPSCR_FR, 0 };
+       switch (x & 3) {
+       case 3:
+               fxchg(fregs, flag[x >> 2]);
+               break;
+       case 1:
+               ftrv(fregs, x - 1);
+               break;
+       default:
+               fsca(fregs, x);
+       }
+       return 0;
+}
+
+static int
+id_fnxd(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, int x, int n)
+{
+       return (fnxd[x])(fregs, n);
+}
+
+static int
+id_fnmx(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, u16 code)
+{
+       int n = (code >> 8) & 0xf, m = (code >> 4) & 0xf, x = code & 0xf;
+       return (fnmx[x])(fregs, regs, m, n);
+}
+
+static int
+id_sys(struct sh_fpu_soft_struct *fregs, struct pt_regs *regs, u16 code)
+{
+       int n = ((code >> 8) & 0xf);
+       unsigned long *reg = (code & 0x0010) ? &FPUL : &FPSCR;
+
+       switch (code & 0xf0ff) {
+       case 0x005a:
+       case 0x006a:
+               Rn = *reg;
+               break;
+       case 0x405a:
+       case 0x406a:
+               *reg = Rn;
+               break;
+       case 0x4052:
+       case 0x4062:
+               Rn -= 4;
+               WRITE(*reg, Rn);
+               break;
+       case 0x4056:
+       case 0x4066:
+               READ(*reg, Rn);
+               Rn += 4;
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int fpu_emulate(u16 code, struct sh_fpu_soft_struct *fregs, struct pt_regs *regs)
+{
+       if ((code & 0xf000) == 0xf000)
+               return id_fnmx(fregs, regs, code);
+       else
+               return id_sys(fregs, regs, code);
+}
+
+/**
+ *     denormal_to_double - Given denormalized float number,
+ *                          store double float
+ *
+ *     @fpu: Pointer to sh_fpu_hard structure
+ *     @n: Index to FP register
+ */
+static void denormal_to_double(struct sh_fpu_hard_struct *fpu, int n)
+{
+       unsigned long du, dl;
+       unsigned long x = fpu->fpul;
+       int exp = 1023 - 126;
+
+       if (x != 0 && (x & 0x7f800000) == 0) {
+               du = (x & 0x80000000);
+               while ((x & 0x00800000) == 0) {
+                       x <<= 1;
+                       exp--;
+               }
+               x &= 0x007fffff;
+               du |= (exp << 20) | (x >> 3);
+               dl = x << 29;
+
+               fpu->fp_regs[n] = du;
+               fpu->fp_regs[n+1] = dl;
+       }
+}
+
+/**
+ *     ieee_fpe_handler - Handle denormalized number exception
+ *
+ *     @regs: Pointer to register structure
+ *
+ *     Returns 1 when it's handled (should not cause exception).
+ */
+static int ieee_fpe_handler(struct pt_regs *regs)
+{
+       unsigned short insn = *(unsigned short *)regs->pc;
+       unsigned short finsn;
+       unsigned long nextpc;
+       int nib[4] = {
+               (insn >> 12) & 0xf,
+               (insn >> 8) & 0xf,
+               (insn >> 4) & 0xf,
+               insn & 0xf};
+
+       if (nib[0] == 0xb ||
+           (nib[0] == 0x4 && nib[2] == 0x0 && nib[3] == 0xb)) /* bsr & jsr */
+               regs->pr = regs->pc + 4;
+
+       if (nib[0] == 0xa || nib[0] == 0xb) { /* bra & bsr */
+               nextpc = regs->pc + 4 + ((short) ((insn & 0xfff) << 4) >> 3);
+               finsn = *(unsigned short *) (regs->pc + 2);
+       } else if (nib[0] == 0x8 && nib[1] == 0xd) { /* bt/s */
+               if (regs->sr & 1)
+                       nextpc = regs->pc + 4 + ((char) (insn & 0xff) << 1);
+               else
+                       nextpc = regs->pc + 4;
+               finsn = *(unsigned short *) (regs->pc + 2);
+       } else if (nib[0] == 0x8 && nib[1] == 0xf) { /* bf/s */
+               if (regs->sr & 1)
+                       nextpc = regs->pc + 4;
+               else
+                       nextpc = regs->pc + 4 + ((char) (insn & 0xff) << 1);
+               finsn = *(unsigned short *) (regs->pc + 2);
+       } else if (nib[0] == 0x4 && nib[3] == 0xb &&
+                (nib[2] == 0x0 || nib[2] == 0x2)) { /* jmp & jsr */
+               nextpc = regs->regs[nib[1]];
+               finsn = *(unsigned short *) (regs->pc + 2);
+       } else if (nib[0] == 0x0 && nib[3] == 0x3 &&
+                (nib[2] == 0x0 || nib[2] == 0x2)) { /* braf & bsrf */
+               nextpc = regs->pc + 4 + regs->regs[nib[1]];
+               finsn = *(unsigned short *) (regs->pc + 2);
+       } else if (insn == 0x000b) { /* rts */
+               nextpc = regs->pr;
+               finsn = *(unsigned short *) (regs->pc + 2);
+       } else {
+               nextpc = regs->pc + 2;
+               finsn = insn;
+       }
+
+       if ((finsn & 0xf1ff) == 0xf0ad) { /* fcnvsd */
+               struct task_struct *tsk = current;
+
+               if ((tsk->thread.fpu.hard.fpscr & (1 << 17))) {
+                       /* FPU error */
+                       denormal_to_double (&tsk->thread.fpu.hard,
+                                           (finsn >> 8) & 0xf);
+                       tsk->thread.fpu.hard.fpscr &=
+                               ~(FPSCR_CAUSE_MASK | FPSCR_FLAG_MASK);
+                       set_tsk_thread_flag(tsk, TIF_USEDFPU);
+               } else {
+                       tsk->thread.trap_no = 11;
+                       tsk->thread.error_code = 0;
+                       force_sig(SIGFPE, tsk);
+               }
+
+               regs->pc = nextpc;
+               return 1;
+       }
+
+       return 0;
+}
+
+asmlinkage void do_fpu_error(unsigned long r4, unsigned long r5,
+                            unsigned long r6, unsigned long r7,
+                            struct pt_regs regs)
+{
+       struct task_struct *tsk = current;
+
+       if (ieee_fpe_handler (&regs))
+               return;
+
+       regs.pc += 2;
+       tsk->thread.trap_no = 11;
+       tsk->thread.error_code = 0;
+       force_sig(SIGFPE, tsk);
+}
+
+/**
+ * fpu_init - Initialize FPU registers
+ * @fpu: Pointer to software emulated FPU registers.
+ */
+static void fpu_init(struct sh_fpu_soft_struct *fpu)
+{
+       int i;
+
+       fpu->fpscr = FPSCR_INIT;
+       fpu->fpul = 0;
+
+       for (i = 0; i < 16; i++) {
+               fpu->fp_regs[i] = 0;
+               fpu->xfp_regs[i]= 0;
+       }
+}
+
+/**
+ * do_fpu_inst - Handle reserved instructions for FPU emulation
+ * @inst: instruction code.
+ * @regs: registers on stack.
+ */
+int do_fpu_inst(unsigned short inst, struct pt_regs *regs)
+{
+       struct task_struct *tsk = current;
+       struct sh_fpu_soft_struct *fpu = &(tsk->thread.fpu.soft);
+
+       if (!test_tsk_thread_flag(tsk, TIF_USEDFPU)) {
+               /* initialize once. */
+               fpu_init(fpu);
+               set_tsk_thread_flag(tsk, TIF_USEDFPU);
+       }
+
+       return fpu_emulate(inst, fpu, regs);
+}
diff --git a/arch/sh/math-emu/sfp-util.h b/arch/sh/math-emu/sfp-util.h
new file mode 100644 (file)
index 0000000..8ae1bd3
--- /dev/null
@@ -0,0 +1,72 @@
+/*
+ * These are copied from glibc/stdlib/longlong.h
+ */
+
+#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
+  do {                                                                  \
+    UWtype __x;                                                         \
+    __x = (al) + (bl);                                                  \
+    (sh) = (ah) + (bh) + (__x < (al));                                  \
+    (sl) = __x;                                                         \
+  } while (0)
+
+#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
+  do {                                                                  \
+    UWtype __x;                                                         \
+    __x = (al) - (bl);                                                  \
+    (sh) = (ah) - (bh) - (__x > (al));                                  \
+    (sl) = __x;                                                         \
+  } while (0)
+
+#define umul_ppmm(w1, w0, u, v) \
+  __asm__ ("dmulu.l %2,%3\n\tsts    macl,%1\n\tsts  mach,%0"   \
+       : "=r" ((u32)(w1)), "=r" ((u32)(w0))    \
+       :  "r" ((u32)(u)),   "r" ((u32)(v))     \
+       : "macl", "mach")
+
+#define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
+#define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
+#define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
+
+#define udiv_qrnnd(q, r, n1, n0, d) \
+  do {                                                                 \
+    UWtype __d1, __d0, __q1, __q0;                                     \
+    UWtype __r1, __r0, __m;                                            \
+    __d1 = __ll_highpart (d);                                          \
+    __d0 = __ll_lowpart (d);                                           \
+                                                                       \
+    __r1 = (n1) % __d1;                                                        \
+    __q1 = (n1) / __d1;                                                        \
+    __m = (UWtype) __q1 * __d0;                                                \
+    __r1 = __r1 * __ll_B | __ll_highpart (n0);                         \
+    if (__r1 < __m)                                                    \
+      {                                                                        \
+       __q1--, __r1 += (d);                                            \
+       if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
+         if (__r1 < __m)                                               \
+           __q1--, __r1 += (d);                                        \
+      }                                                                        \
+    __r1 -= __m;                                                       \
+                                                                       \
+    __r0 = __r1 % __d1;                                                        \
+    __q0 = __r1 / __d1;                                                        \
+    __m = (UWtype) __q0 * __d0;                                                \
+    __r0 = __r0 * __ll_B | __ll_lowpart (n0);                          \
+    if (__r0 < __m)                                                    \
+      {                                                                        \
+       __q0--, __r0 += (d);                                            \
+       if (__r0 >= (d))                                                \
+         if (__r0 < __m)                                               \
+           __q0--, __r0 += (d);                                        \
+      }                                                                        \
+    __r0 -= __m;                                                       \
+                                                                       \
+    (q) = (UWtype) __q1 * __ll_B | __q0;                               \
+    (r) = __r0;                                                                \
+  } while (0)
+
+#define abort()        return 0
+
+#define __BYTE_ORDER __LITTLE_ENDIAN
+
+
diff --git a/include/asm-sh/sfp-machine.h b/include/asm-sh/sfp-machine.h
new file mode 100644 (file)
index 0000000..8a6399a
--- /dev/null
@@ -0,0 +1,86 @@
+/* Machine-dependent software floating-point definitions.
+   SuperH kernel version.
+   Copyright (C) 1997,1998,1999 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Richard Henderson (rth@cygnus.com),
+                 Jakub Jelinek (jj@ultra.linux.cz),
+                 David S. Miller (davem@redhat.com) and
+                 Peter Maydell (pmaydell@chiark.greenend.org.uk).
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library 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
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with the GNU C Library; see the file COPYING.LIB.  If
+   not, write to the Free Software Foundation, Inc.,
+   59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.  */
+
+#ifndef _SFP_MACHINE_H
+#define _SFP_MACHINE_H
+
+#include <linux/config.h>
+
+#define _FP_W_TYPE_SIZE                32
+#define _FP_W_TYPE             unsigned long
+#define _FP_WS_TYPE            signed long
+#define _FP_I_TYPE             long
+
+#define _FP_MUL_MEAT_S(R,X,Y)                                  \
+  _FP_MUL_MEAT_1_wide(_FP_WFRACBITS_S,R,X,Y,umul_ppmm)
+#define _FP_MUL_MEAT_D(R,X,Y)                                  \
+  _FP_MUL_MEAT_2_wide(_FP_WFRACBITS_D,R,X,Y,umul_ppmm)
+#define _FP_MUL_MEAT_Q(R,X,Y)                                  \
+  _FP_MUL_MEAT_4_wide(_FP_WFRACBITS_Q,R,X,Y,umul_ppmm)
+
+#define _FP_DIV_MEAT_S(R,X,Y)  _FP_DIV_MEAT_1_udiv(S,R,X,Y)
+#define _FP_DIV_MEAT_D(R,X,Y)  _FP_DIV_MEAT_2_udiv(D,R,X,Y)
+#define _FP_DIV_MEAT_Q(R,X,Y)  _FP_DIV_MEAT_4_udiv(Q,R,X,Y)
+
+#define _FP_NANFRAC_S          ((_FP_QNANBIT_S << 1) - 1)
+#define _FP_NANFRAC_D          ((_FP_QNANBIT_D << 1) - 1), -1
+#define _FP_NANFRAC_Q          ((_FP_QNANBIT_Q << 1) - 1), -1, -1, -1
+#define _FP_NANSIGN_S          0
+#define _FP_NANSIGN_D          0
+#define _FP_NANSIGN_Q          0
+
+#define _FP_KEEPNANFRACP 1
+
+/*
+ * If one NaN is signaling and the other is not,
+ * we choose that one, otherwise we choose X.
+ */
+#define _FP_CHOOSENAN(fs, wc, R, X, Y, OP)                      \
+  do {                                                          \
+    if ((_FP_FRAC_HIGH_RAW_##fs(X) & _FP_QNANBIT_##fs)          \
+        && !(_FP_FRAC_HIGH_RAW_##fs(Y) & _FP_QNANBIT_##fs))     \
+      {                                                         \
+        R##_s = Y##_s;                                          \
+        _FP_FRAC_COPY_##wc(R,Y);                                \
+      }                                                         \
+    else                                                        \
+      {                                                         \
+        R##_s = X##_s;                                          \
+        _FP_FRAC_COPY_##wc(R,X);                                \
+      }                                                         \
+    R##_c = FP_CLS_NAN;                                         \
+  } while (0)
+
+//#define FP_ROUNDMODE         FPSCR_RM
+#define FP_DENORM_ZERO         1/*FPSCR_DN*/
+
+/* Exception flags. */
+#define FP_EX_INVALID          (1<<4)
+#define FP_EX_DIVZERO          (1<<3)
+#define FP_EX_OVERFLOW         (1<<2)
+#define FP_EX_UNDERFLOW                (1<<1)
+#define FP_EX_INEXACT          (1<<0)
+
+#endif
+