arch/sh/kernel/cpu/sh4/fpu.c

   1 /* $Id: fpu.c,v 1.4 2004/01/13 05:52:11 kkojima Exp $
   2  *
   3  * linux/arch/sh/kernel/fpu.c
   4  *
   5  * Save/restore floating point context for signal handlers.
   6  *
   7  * This file is subject to the terms and conditions of the GNU General Public
   8  * License.  See the file "COPYING" in the main directory of this archive
   9  * for more details.
  10  *
  11  * Copyright (C) 1999, 2000  Kaz Kojima & Niibe Yutaka
  12  *
  13  * FIXME! These routines can be optimized in big endian case.
  14  */
  15
  16 #include <linux/sched.h>
  17 #include <linux/signal.h>
  18 #include <asm/processor.h>
  19 #include <asm/io.h>
  20
  21 /* The PR (precision) bit in the FP Status Register must be clear when
  22  * an frchg instruction is executed, otherwise the instruction is undefined.
  23  * Executing frchg with PR set causes a trap on some SH4 implementations.
  24  */
  25
  26 #define FPSCR_RCHG 0x00000000
  27
  28
  29 /*
  30  * Save FPU registers onto task structure.
  31  * Assume called with FPU enabled (SR.FD=0).
  32  */
  33 void
  34 save_fpu(struct task_struct *tsk, struct pt_regs *regs)
  35 {
  36         unsigned long dummy;
  37
  38         clear_tsk_thread_flag(tsk, TIF_USEDFPU);
  39         enable_fpu();
  40         asm volatile("sts.l     fpul, @-%0\n\t"
  41                      "sts.l     fpscr, @-%0\n\t"
  42                      "lds       %2, fpscr\n\t"
  43                      "frchg\n\t"
  44                      "fmov.s    fr15, @-%0\n\t"
  45                      "fmov.s    fr14, @-%0\n\t"
  46                      "fmov.s    fr13, @-%0\n\t"
  47                      "fmov.s    fr12, @-%0\n\t"
  48                      "fmov.s    fr11, @-%0\n\t"
  49                      "fmov.s    fr10, @-%0\n\t"
  50                      "fmov.s    fr9, @-%0\n\t"
  51                      "fmov.s    fr8, @-%0\n\t"
  52                      "fmov.s    fr7, @-%0\n\t"
  53                      "fmov.s    fr6, @-%0\n\t"
  54                      "fmov.s    fr5, @-%0\n\t"
  55                      "fmov.s    fr4, @-%0\n\t"
  56                      "fmov.s    fr3, @-%0\n\t"
  57                      "fmov.s    fr2, @-%0\n\t"
  58                      "fmov.s    fr1, @-%0\n\t"
  59                      "fmov.s    fr0, @-%0\n\t"
  60                      "frchg\n\t"
  61                      "fmov.s    fr15, @-%0\n\t"
  62                      "fmov.s    fr14, @-%0\n\t"
  63                      "fmov.s    fr13, @-%0\n\t"
  64                      "fmov.s    fr12, @-%0\n\t"
  65                      "fmov.s    fr11, @-%0\n\t"
  66                      "fmov.s    fr10, @-%0\n\t"
  67                      "fmov.s    fr9, @-%0\n\t"
  68                      "fmov.s    fr8, @-%0\n\t"
  69                      "fmov.s    fr7, @-%0\n\t"
  70                      "fmov.s    fr6, @-%0\n\t"
  71                      "fmov.s    fr5, @-%0\n\t"
  72                      "fmov.s    fr4, @-%0\n\t"
  73                      "fmov.s    fr3, @-%0\n\t"
  74                      "fmov.s    fr2, @-%0\n\t"
  75                      "fmov.s    fr1, @-%0\n\t"
  76                      "fmov.s    fr0, @-%0\n\t"
  77                      "lds       %3, fpscr\n\t"
  78                      : "=r" (dummy)
  79                      : "0" ((char *)(&tsk->thread.fpu.hard.status)),
  80                        "r" (FPSCR_RCHG),
  81                        "r" (FPSCR_INIT)
  82                      : "memory");
  83
  84         disable_fpu();
  85         release_fpu(regs);
  86 }
  87
  88 static void
  89 restore_fpu(struct task_struct *tsk)
  90 {
  91         unsigned long dummy;
  92
  93         enable_fpu();
  94         asm volatile("lds       %2, fpscr\n\t"
  95                      "fmov.s    @%0+, fr0\n\t"
  96                      "fmov.s    @%0+, fr1\n\t"
  97                      "fmov.s    @%0+, fr2\n\t"
  98                      "fmov.s    @%0+, fr3\n\t"
  99                      "fmov.s    @%0+, fr4\n\t"
 100                      "fmov.s    @%0+, fr5\n\t"
 101                      "fmov.s    @%0+, fr6\n\t"
 102                      "fmov.s    @%0+, fr7\n\t"
 103                      "fmov.s    @%0+, fr8\n\t"
 104                      "fmov.s    @%0+, fr9\n\t"
 105                      "fmov.s    @%0+, fr10\n\t"
 106                      "fmov.s    @%0+, fr11\n\t"
 107                      "fmov.s    @%0+, fr12\n\t"
 108                      "fmov.s    @%0+, fr13\n\t"
 109                      "fmov.s    @%0+, fr14\n\t"
 110                      "fmov.s    @%0+, fr15\n\t"
 111                      "frchg\n\t"
 112                      "fmov.s    @%0+, fr0\n\t"
 113                      "fmov.s    @%0+, fr1\n\t"
 114                      "fmov.s    @%0+, fr2\n\t"
 115                      "fmov.s    @%0+, fr3\n\t"
 116                      "fmov.s    @%0+, fr4\n\t"
 117                      "fmov.s    @%0+, fr5\n\t"
 118                      "fmov.s    @%0+, fr6\n\t"
 119                      "fmov.s    @%0+, fr7\n\t"
 120                      "fmov.s    @%0+, fr8\n\t"
 121                      "fmov.s    @%0+, fr9\n\t"
 122                      "fmov.s    @%0+, fr10\n\t"
 123                      "fmov.s    @%0+, fr11\n\t"
 124                      "fmov.s    @%0+, fr12\n\t"
 125                      "fmov.s    @%0+, fr13\n\t"
 126                      "fmov.s    @%0+, fr14\n\t"
 127                      "fmov.s    @%0+, fr15\n\t"
 128                      "frchg\n\t"
 129                      "lds.l     @%0+, fpscr\n\t"
 130                      "lds.l     @%0+, fpul\n\t"
 131                      : "=r" (dummy)
 132                      : "0" (&tsk->thread.fpu), "r" (FPSCR_RCHG)
 133                      : "memory");
 134         disable_fpu();
 135 }
 136
 137 /*
 138  * Load the FPU with signalling NANS.  This bit pattern we're using
 139  * has the property that no matter wether considered as single or as
 140  * double precission represents signaling NANS.
 141  */
 142
 143 static void
 144 fpu_init(void)
 145 {
 146         enable_fpu();
 147         asm volatile("lds       %0, fpul\n\t"
 148                      "lds       %1, fpscr\n\t"
 149                      "fsts      fpul, fr0\n\t"
 150                      "fsts      fpul, fr1\n\t"
 151                      "fsts      fpul, fr2\n\t"
 152                      "fsts      fpul, fr3\n\t"
 153                      "fsts      fpul, fr4\n\t"
 154                      "fsts      fpul, fr5\n\t"
 155                      "fsts      fpul, fr6\n\t"
 156                      "fsts      fpul, fr7\n\t"
 157                      "fsts      fpul, fr8\n\t"
 158                      "fsts      fpul, fr9\n\t"
 159                      "fsts      fpul, fr10\n\t"
 160                      "fsts      fpul, fr11\n\t"
 161                      "fsts      fpul, fr12\n\t"
 162                      "fsts      fpul, fr13\n\t"
 163                      "fsts      fpul, fr14\n\t"
 164                      "fsts      fpul, fr15\n\t"
 165                      "frchg\n\t"
 166                      "fsts      fpul, fr0\n\t"
 167                      "fsts      fpul, fr1\n\t"
 168                      "fsts      fpul, fr2\n\t"
 169                      "fsts      fpul, fr3\n\t"
 170                      "fsts      fpul, fr4\n\t"
 171                      "fsts      fpul, fr5\n\t"
 172                      "fsts      fpul, fr6\n\t"
 173                      "fsts      fpul, fr7\n\t"
 174                      "fsts      fpul, fr8\n\t"
 175                      "fsts      fpul, fr9\n\t"
 176                      "fsts      fpul, fr10\n\t"
 177                      "fsts      fpul, fr11\n\t"
 178                      "fsts      fpul, fr12\n\t"
 179                      "fsts      fpul, fr13\n\t"
 180                      "fsts      fpul, fr14\n\t"
 181                      "fsts      fpul, fr15\n\t"
 182                      "frchg\n\t"
 183                      "lds       %2, fpscr\n\t"
 184                      : /* no output */
 185                      : "r" (0), "r" (FPSCR_RCHG), "r" (FPSCR_INIT));
 186         disable_fpu();
 187 }
 188
 189 /**
 190  *      denormal_to_double - Given denormalized float number,
 191  *                           store double float
 192  *
 193  *      @fpu: Pointer to sh_fpu_hard structure
 194  *      @n: Index to FP register
 195  */
 196 static void
 197 denormal_to_double (struct sh_fpu_hard_struct *fpu, int n)
 198 {
 199         unsigned long du, dl;
 200         unsigned long x = fpu->fpul;
 201         int exp = 1023 - 126;
 202
 203         if (x != 0 && (x & 0x7f800000) == 0) {
 204                 du = (x & 0x80000000);
 205                 while ((x & 0x00800000) == 0) {
 206                         x <<= 1;
 207                         exp--;
 208                 }
 209                 x &= 0x007fffff;
 210                 du |= (exp << 20) | (x >> 3);
 211                 dl = x << 29;
 212
 213                 fpu->fp_regs[n] = du;
 214                 fpu->fp_regs[n+1] = dl;
 215         }
 216 }
 217
 218 /**
 219  *      ieee_fpe_handler - Handle denormalized number exception
 220  *
 221  *      @regs: Pointer to register structure
 222  *
 223  *      Returns 1 when it's handled (should not cause exception).
 224  */
 225 static int
 226 ieee_fpe_handler (struct pt_regs *regs)
 227 {
 228         unsigned short insn = *(unsigned short *) regs->pc;
 229         unsigned short finsn;
 230         unsigned long nextpc;
 231         int nib[4] = {
 232                 (insn >> 12) & 0xf,
 233                 (insn >> 8) & 0xf,
 234                 (insn >> 4) & 0xf,
 235                 insn & 0xf};
 236
 237         if (nib[0] == 0xb ||
 238             (nib[0] == 0x4 && nib[2] == 0x0 && nib[3] == 0xb)) /* bsr & jsr */
 239                 regs->pr = regs->pc + 4;
 240
 241         if (nib[0] == 0xa || nib[0] == 0xb) { /* bra & bsr */
 242                 nextpc = regs->pc + 4 + ((short) ((insn & 0xfff) << 4) >> 3);
 243                 finsn = *(unsigned short *) (regs->pc + 2);
 244         } else if (nib[0] == 0x8 && nib[1] == 0xd) { /* bt/s */
 245                 if (regs->sr & 1)
 246                         nextpc = regs->pc + 4 + ((char) (insn & 0xff) << 1);
 247                 else
 248                         nextpc = regs->pc + 4;
 249                 finsn = *(unsigned short *) (regs->pc + 2);
 250         } else if (nib[0] == 0x8 && nib[1] == 0xf) { /* bf/s */
 251                 if (regs->sr & 1)
 252                         nextpc = regs->pc + 4;
 253                 else
 254                         nextpc = regs->pc + 4 + ((char) (insn & 0xff) << 1);
 255                 finsn = *(unsigned short *) (regs->pc + 2);
 256         } else if (nib[0] == 0x4 && nib[3] == 0xb &&
 257                  (nib[2] == 0x0 || nib[2] == 0x2)) { /* jmp & jsr */
 258                 nextpc = regs->regs[nib[1]];
 259                 finsn = *(unsigned short *) (regs->pc + 2);
 260         } else if (nib[0] == 0x0 && nib[3] == 0x3 &&
 261                  (nib[2] == 0x0 || nib[2] == 0x2)) { /* braf & bsrf */
 262                 nextpc = regs->pc + 4 + regs->regs[nib[1]];
 263                 finsn = *(unsigned short *) (regs->pc + 2);
 264         } else if (insn == 0x000b) { /* rts */
 265                 nextpc = regs->pr;
 266                 finsn = *(unsigned short *) (regs->pc + 2);
 267         } else {
 268                 nextpc = regs->pc + 2;
 269                 finsn = insn;
 270         }
 271
 272         if ((finsn & 0xf1ff) == 0xf0ad) { /* fcnvsd */
 273                 struct task_struct *tsk = current;
 274
 275                 save_fpu(tsk, regs);
 276                 if ((tsk->thread.fpu.hard.fpscr & (1 << 17))) {
 277                         /* FPU error */
 278                         denormal_to_double (&tsk->thread.fpu.hard,
 279                                             (finsn >> 8) & 0xf);
 280                         tsk->thread.fpu.hard.fpscr &=
 281                                 ~(FPSCR_CAUSE_MASK | FPSCR_FLAG_MASK);
 282                         grab_fpu(regs);
 283                         restore_fpu(tsk);
 284                         set_tsk_thread_flag(tsk, TIF_USEDFPU);
 285                 } else {
 286                         tsk->thread.trap_no = 11;
 287                         tsk->thread.error_code = 0;
 288                         force_sig(SIGFPE, tsk);
 289                 }
 290
 291                 regs->pc = nextpc;
 292                 return 1;
 293         }
 294
 295         return 0;
 296 }
 297
 298 asmlinkage void
 299 do_fpu_error(unsigned long r4, unsigned long r5, unsigned long r6, unsigned long r7,
 300              struct pt_regs regs)
 301 {
 302         struct task_struct *tsk = current;
 303
 304         if (ieee_fpe_handler (&regs))
 305                 return;
 306
 307         regs.pc += 2;
 308         save_fpu(tsk, &regs);
 309         tsk->thread.trap_no = 11;
 310         tsk->thread.error_code = 0;
 311         force_sig(SIGFPE, tsk);
 312 }
 313
 314 asmlinkage void
 315 do_fpu_state_restore(unsigned long r4, unsigned long r5, unsigned long r6,
 316                      unsigned long r7, struct pt_regs regs)
 317 {
 318         struct task_struct *tsk = current;
 319
 320         grab_fpu(&regs);
 321         if (!user_mode(&regs)) {
 322                 printk(KERN_ERR "BUG: FPU is used in kernel mode.\n");
 323                 return;
 324         }
 325
 326         if (used_math()) {
 327                 /* Using the FPU again.  */
 328                 restore_fpu(tsk);
 329         } else  {
 330                 /* First time FPU user.  */
 331                 fpu_init();
 332                 set_used_math();
 333         }
 334         set_tsk_thread_flag(tsk, TIF_USEDFPU);
 335 }