aboutsummaryrefslogblamecommitdiffstats
path: root/erts/emulator/sys/unix/sys_float.c
blob: d6d06d00364ae24b756b6726037ed8fc0f645993 (plain) (tree)
1
2
3
4
5
6
7
8
9
10
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812

                   


                                                        




                                                                      
  



                                                                         
  














































































                                                                                           

                                             










































































































































































































































































































































































































































































































































































































































































































































                                                                                                                                                                                                                                                                                                                                                                                     


                                                                     


             
/*
 * %CopyrightBegin%
 *
 * Copyright Ericsson AB 2001-2010. All Rights Reserved.
 *
 * The contents of this file are subject to the Erlang Public License,
 * Version 1.1, (the "License"); you may not use this file except in
 * compliance with the License. You should have received a copy of the
 * Erlang Public License along with this software. If not, it can be
 * retrieved online at http://www.erlang.org/.
 *
 * Software distributed under the License is distributed on an "AS IS"
 * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
 * the License for the specific language governing rights and limitations
 * under the License.
 *
 * %CopyrightEnd%
 */

#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

#include "sys.h"
#include "global.h"
#include "erl_process.h"


#ifdef NO_FPE_SIGNALS

void
erts_sys_init_float(void)
{
# ifdef SIGFPE
    sys_sigset(SIGFPE, SIG_IGN); /* Ignore so we can test for NaN and Inf */
# endif
}

static ERTS_INLINE void set_current_fp_exception(unsigned long pc)
{
    /* nothing to do */
}

#else  /* !NO_FPE_SIGNALS */

#ifdef ERTS_SMP
static erts_tsd_key_t fpe_key;

/* once-only initialisation early in the main thread (via erts_sys_init_float()) */
static void erts_init_fp_exception(void)
{
    /* XXX: the wrappers prevent using a pthread destructor to
       deallocate the key's value; so when/where do we do that? */
    erts_tsd_key_create(&fpe_key);
}

void erts_thread_init_fp_exception(void)
{
    unsigned long *fpe = erts_alloc(ERTS_ALC_T_FP_EXCEPTION, sizeof(*fpe));
    *fpe = 0L;
    erts_tsd_set(fpe_key, fpe);
}

static ERTS_INLINE volatile unsigned long *erts_thread_get_fp_exception(void)
{
    return (volatile unsigned long*)erts_tsd_get(fpe_key);
}
#else /* !SMP */
#define erts_init_fp_exception()	/*empty*/
static volatile unsigned long fp_exception;
#define erts_thread_get_fp_exception()	(&fp_exception)
#endif /* SMP */

volatile unsigned long *erts_get_current_fp_exception(void)
{
    Process *c_p;

    c_p = erts_get_current_process();
    if (c_p)
	return &c_p->fp_exception;
    return erts_thread_get_fp_exception();
}

static void set_current_fp_exception(unsigned long pc)
{
    volatile unsigned long *fpexnp = erts_get_current_fp_exception();
    ASSERT(fpexnp != NULL);
    *fpexnp = pc;
}

void erts_fp_check_init_error(volatile unsigned long *fpexnp)
{
    char buf[64];
    snprintf(buf, sizeof buf, "ERTS_FP_CHECK_INIT at %p: detected unhandled FPE at %p\r\n",
	     __builtin_return_address(0), (void*)*fpexnp);
    if (write(2, buf, strlen(buf)) <= 0)
	erl_exit(ERTS_ABORT_EXIT, "%s", buf);
    *fpexnp = 0;
#if defined(__i386__) || defined(__x86_64__)
    erts_restore_fpu();
#endif
}

/* Is there no standard identifier for Darwin/MacOSX ? */
#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
#define __DARWIN__ 1
#endif

#if (defined(__i386__) || defined(__x86_64__)) && defined(__GNUC__)

static void unmask_x87(void)
{
    unsigned short cw;

    __asm__ __volatile__("fstcw %0" : "=m"(cw));
    cw &= ~(0x01|0x04|0x08);   /* unmask IM, ZM, OM */
    __asm__ __volatile__("fldcw %0" : : "m"(cw));
}

/* mask x87 FPE, return true if the previous state was unmasked */
static int mask_x87(void)
{
    unsigned short cw;
    int unmasked;

    __asm__ __volatile__("fstcw %0" : "=m"(cw));
    unmasked = (cw & (0x01|0x04|0x08)) == 0;
    /* or just set cw = 0x37f */
    cw |= (0x01|0x04|0x08); /* mask IM, ZM, OM */
    __asm__ __volatile__("fldcw %0" : : "m"(cw));
    return unmasked;
}

static void unmask_sse2(void)
{
    unsigned int mxcsr;

    __asm__ __volatile__("stmxcsr %0" : "=m"(mxcsr));
    mxcsr &= ~(0x003F|0x0680); /* clear exn flags, unmask OM, ZM, IM (not PM, UM, DM) */
    __asm__ __volatile__("ldmxcsr %0" : : "m"(mxcsr));
}

/* mask SSE2 FPE, return true if the previous state was unmasked */
static int mask_sse2(void)
{
    unsigned int mxcsr;
    int unmasked;

    __asm__ __volatile__("stmxcsr %0" : "=m"(mxcsr));
    unmasked = (mxcsr & 0x0680) == 0;
    /* or just set mxcsr = 0x1f80 */
    mxcsr &= ~0x003F; /* clear exn flags */
    mxcsr |=  0x0680; /* mask OM, ZM, IM (not PM, UM, DM) */
    __asm__ __volatile__("ldmxcsr %0" : : "m"(mxcsr));
    return unmasked;
}

#if defined(__x86_64__)

static inline int cpu_has_sse2(void) { return 1; }

#else /* !__x86_64__ */

/*
 * Check if an x86-32 processor has SSE2.
 */
static unsigned int xor_eflags(unsigned int mask)
{
    unsigned int eax, edx;

    eax = mask;			/* eax = mask */
    __asm__("pushfl\n\t"
	    "popl %0\n\t"	/* edx = original EFLAGS */
	    "xorl %0, %1\n\t"	/* eax = mask ^ EFLAGS */
	    "pushl %1\n\t"
	    "popfl\n\t"		/* new EFLAGS = mask ^ original EFLAGS */
	    "pushfl\n\t"
	    "popl %1\n\t"	/* eax = new EFLAGS */
	    "xorl %0, %1\n\t"	/* eax = new EFLAGS ^ old EFLAGS */
	    "pushl %0\n\t"
	    "popfl"		/* restore original EFLAGS */
	    : "=d"(edx), "=a"(eax)
	    : "1"(eax));
    return eax;
}

static __inline__ unsigned int cpuid_eax(unsigned int op)
{
    unsigned int eax, save_ebx;

    /* In PIC mode i386 reserves EBX. So we must save
       and restore it ourselves to not upset gcc. */
    __asm__(
	"movl %%ebx, %1\n\t"
	"cpuid\n\t"
	"movl %1, %%ebx"
	: "=a"(eax), "=m"(save_ebx)
	: "0"(op)
	: "cx", "dx");
    return eax;
}

static __inline__ unsigned int cpuid_edx(unsigned int op)
{
    unsigned int eax, edx, save_ebx;
 
    /* In PIC mode i386 reserves EBX. So we must save
       and restore it ourselves to not upset gcc. */
    __asm__(
	"movl %%ebx, %2\n\t"
	"cpuid\n\t"
	"movl %2, %%ebx"
	: "=a"(eax), "=d"(edx), "=m"(save_ebx)
	: "0"(op)
	: "cx");
    return edx;
}

/* The AC bit, bit #18, is a new bit introduced in the EFLAGS
 * register on the Intel486 processor to generate alignment
 * faults. This bit cannot be set on the Intel386 processor.
 */
static __inline__ int is_386(void)
{
    return ((xor_eflags(1<<18) >> 18) & 1) == 0;
}

/* Newer x86 processors have a CPUID instruction, as indicated by
 * the ID bit (#21) in EFLAGS being modifiable.
 */
static __inline__ int has_CPUID(void)
{
    return (xor_eflags(1<<21) >> 21) & 1;
}

static int cpu_has_sse2(void)
{
    unsigned int maxlev, features;
    static int has_sse2 = -1;

    if (has_sse2 >= 0)
	return has_sse2;
    has_sse2 = 0;

    if (is_386())
	return 0;
    if (!has_CPUID())
	return 0;
    maxlev = cpuid_eax(0);
    /* Intel A-step Pentium had a preliminary version of CPUID.
       It also didn't have SSE2. */
    if ((maxlev & 0xFFFFFF00) == 0x0500)
	return 0;
    /* If max level is zero then CPUID cannot report any features. */
    if (maxlev == 0)
	return 0;
    features = cpuid_edx(1);
    has_sse2 = (features & (1 << 26)) != 0;

    return has_sse2;
}
#endif /* !__x86_64__ */

static void unmask_fpe(void)
{
    __asm__ __volatile__("fnclex");
    unmask_x87();
    if (cpu_has_sse2())
	unmask_sse2();
}

static void unmask_fpe_conditional(int unmasked)
{
    if (unmasked)
	unmask_fpe();
}

/* mask x86 FPE, return true if the previous state was unmasked */
static int mask_fpe(void)
{
    int unmasked;

    unmasked = mask_x87();
    if (cpu_has_sse2())
	unmasked |= mask_sse2();
    return unmasked;
}

void erts_restore_fpu(void)
{
    __asm__ __volatile__("fninit");
    unmask_x87();
    if (cpu_has_sse2())
	unmask_sse2();
}

#elif defined(__sparc__) && defined(__linux__)

#if defined(__arch64__)
#define LDX "ldx"
#define STX "stx"
#else
#define LDX "ld"
#define STX "st"
#endif

static void unmask_fpe(void)
{
    unsigned long fsr;

    __asm__(STX " %%fsr, %0" : "=m"(fsr));
    fsr &= ~(0x1FUL << 23);	/* clear FSR[TEM] field */
    fsr |= (0x1AUL << 23);	/* enable NV, OF, DZ exceptions */
    __asm__ __volatile__(LDX " %0, %%fsr" : : "m"(fsr));
}

static void unmask_fpe_conditional(int unmasked)
{
    if (unmasked)
	unmask_fpe();
}

/* mask SPARC FPE, return true if the previous state was unmasked */
static int mask_fpe(void)
{
    unsigned long fsr;
    int unmasked;

    __asm__(STX " %%fsr, %0" : "=m"(fsr));
    unmasked = ((fsr >> 23) & 0x1A) == 0x1A;
    fsr &= ~(0x1FUL << 23);	/* clear FSR[TEM] field */
    __asm__ __volatile__(LDX " %0, %%fsr" : : "m"(fsr));
    return unmasked;
}

#elif (defined(__powerpc__) && defined(__linux__)) || (defined(__ppc__) && defined(__DARWIN__))

#if defined(__linux__)
#include <sys/prctl.h>

static void set_fpexc_precise(void)
{
    if (prctl(PR_SET_FPEXC, PR_FP_EXC_PRECISE) < 0) {
	perror("PR_SET_FPEXC");
	exit(1);
    }
}

#elif defined(__DARWIN__)

#include <mach/mach.h>
#include <pthread.h>

/*
 * FE0 FE1	MSR bits
 *  0   0	floating-point exceptions disabled
 *  0   1	floating-point imprecise nonrecoverable
 *  1   0	floating-point imprecise recoverable
 *  1   1	floating-point precise mode
 *
 * Apparently:
 * - Darwin 5.5 (MacOS X <= 10.1) starts with FE0 == FE1 == 0,
 *   and resets FE0 and FE1 to 0 after each SIGFPE.
 * - Darwin 6.0 (MacOS X 10.2) starts with FE0 == FE1 == 1,
 *   and does not reset FE0 or FE1 after a SIGFPE.
 */
#define FE0_MASK	(1<<11)
#define FE1_MASK	(1<<8)

/* a thread cannot get or set its own MSR bits */
static void *fpu_fpe_enable(void *arg)
{
    thread_t t = *(thread_t*)arg;
    struct ppc_thread_state state;
    unsigned int state_size = PPC_THREAD_STATE_COUNT;

    if (thread_get_state(t, PPC_THREAD_STATE, (natural_t*)&state, &state_size) != KERN_SUCCESS) {
	perror("thread_get_state");
	exit(1);
    }
    if ((state.srr1 & (FE1_MASK|FE0_MASK)) != (FE1_MASK|FE0_MASK)) {
#if 1
	/* This would also have to be performed in the SIGFPE handler
	   to work around the MSR reset older Darwin releases do. */
	state.srr1 |= (FE1_MASK|FE0_MASK);
	thread_set_state(t, PPC_THREAD_STATE, (natural_t*)&state, state_size);
#else
	fprintf(stderr, "srr1 == 0x%08x, your Darwin is too old\n", state.srr1);
	exit(1);
#endif
    }
    return NULL; /* Ok, we appear to be on Darwin 6.0 or later */
}

static void set_fpexc_precise(void)
{
    thread_t self = mach_thread_self();
    pthread_t enabler;

    if (pthread_create(&enabler, NULL, fpu_fpe_enable, &self)) {
	perror("pthread_create");
    } else if (pthread_join(enabler, NULL)) {
	perror("pthread_join");
    }
}

#endif

static void set_fpscr(unsigned int fpscr)
{
    union {
	double d;
	unsigned int fpscr[2];
    } u;

    u.fpscr[0] = 0xFFF80000;
    u.fpscr[1] = fpscr;
    __asm__ __volatile__("mtfsf 255,%0" : : "f"(u.d));
}

static unsigned int get_fpscr(void)
{
    union {
	double d;
	unsigned int fpscr[2];
    } u;

    __asm__("mffs %0" : "=f"(u.d));
    return u.fpscr[1];
}

static void unmask_fpe(void)
{
    set_fpexc_precise();
    set_fpscr(0x80|0x40|0x10);	/* VE, OE, ZE; not UE or XE */
}

static void unmask_fpe_conditional(int unmasked)
{
    if (unmasked)
	unmask_fpe();
}

/* mask PowerPC FPE, return true if the previous state was unmasked */
static int mask_fpe(void)
{
    int unmasked;

    unmasked = (get_fpscr() & (0x80|0x40|0x10)) == (0x80|0x40|0x10);
    set_fpscr(0x00);
    return unmasked;
}

#else

static void unmask_fpe(void)
{
    fpsetmask(FP_X_INV | FP_X_OFL | FP_X_DZ);
}

static void unmask_fpe_conditional(int unmasked)
{
    if (unmasked)
	unmask_fpe();
}

/* mask IEEE FPE, return true if previous state was unmasked */
static int mask_fpe(void)
{
    const fp_except unmasked_mask = FP_X_INV | FP_X_OFL | FP_X_DZ;
    fp_except old_mask;

    old_mask = fpsetmask(0);
    return (old_mask & unmasked_mask) == unmasked_mask;
}

#endif

#if (defined(__linux__) && (defined(__i386__) || defined(__x86_64__) || defined(__sparc__) || defined(__powerpc__))) || (defined(__DARWIN__) && (defined(__i386__) || defined(__x86_64__) || defined(__ppc__))) || (defined(__FreeBSD__) && (defined(__x86_64__) || defined(__i386__))) || (defined(__OpenBSD__) && defined(__x86_64__)) || (defined(__sun__) && defined(__x86_64__))

#if defined(__linux__) && defined(__i386__)
#if !defined(X86_FXSR_MAGIC)
#define X86_FXSR_MAGIC 0x0000
#endif
#elif defined(__FreeBSD__) && defined(__x86_64__)
#include <sys/types.h>
#include <machine/fpu.h>
#elif defined(__FreeBSD__) && defined(__i386__)
#include <sys/types.h>
#include <machine/npx.h>
#elif defined(__DARWIN__)
#include <machine/signal.h>
#elif defined(__OpenBSD__) && defined(__x86_64__)
#include <sys/types.h>
#include <machine/fpu.h>
#endif
#if !(defined(__OpenBSD__) && defined(__x86_64__))
#include <ucontext.h>
#endif
#include <string.h>

#if defined(__linux__) && defined(__x86_64__)
#define mc_pc(mc)	((mc)->gregs[REG_RIP])
#elif defined(__linux__) && defined(__i386__)
#define mc_pc(mc)	((mc)->gregs[REG_EIP])
#elif defined(__DARWIN__) && defined(__i386__)
#ifdef DARWIN_MODERN_MCONTEXT
#define mc_pc(mc)	((mc)->__ss.__eip)
#else
#define mc_pc(mc)	((mc)->ss.eip)
#endif
#elif defined(__DARWIN__) && defined(__x86_64__)
#ifdef DARWIN_MODERN_MCONTEXT
#define mc_pc(mc)	((mc)->__ss.__rip)
#else
#define mc_pc(mc)	((mc)->ss.rip)
#endif
#elif defined(__FreeBSD__) && defined(__x86_64__)
#define mc_pc(mc)	((mc)->mc_rip)
#elif defined(__FreeBSD__) && defined(__i386__)
#define mc_pc(mc)	((mc)->mc_eip)
#elif defined(__OpenBSD__) && defined(__x86_64__)
#define mc_pc(mc)	((mc)->sc_rip)
#elif defined(__sun__) && defined(__x86_64__)
#define mc_pc(mc)	((mc)->gregs[REG_RIP])
#endif

static void fpe_sig_action(int sig, siginfo_t *si, void *puc)
{
    ucontext_t *uc = puc;
    unsigned long pc;

#if defined(__linux__)
#if defined(__x86_64__)
    mcontext_t *mc = &uc->uc_mcontext;
    fpregset_t fpstate = mc->fpregs;
    pc = mc_pc(mc);
    /* A failed SSE2 instruction will restart. To avoid
       looping we mask SSE2 exceptions now and unmask them
       again later in erts_check_fpe()/erts_restore_fpu().
       On RISCs we update PC to skip the failed instruction,
       but the ever increasing complexity of the x86 instruction
       set encoding makes that a poor solution here. */
    fpstate->mxcsr = 0x1F80;
    fpstate->swd &= ~0xFF;
#elif defined(__i386__)
    mcontext_t *mc = &uc->uc_mcontext;
    fpregset_t fpstate = mc->fpregs;
    pc = mc_pc(mc);
    if ((fpstate->status >> 16) == X86_FXSR_MAGIC)
	((struct _fpstate*)fpstate)->mxcsr = 0x1F80;
    fpstate->sw &= ~0xFF;
#elif defined(__sparc__) && defined(__arch64__)
    /* on SPARC the 3rd parameter points to a sigcontext not a ucontext */
    struct sigcontext *sc = (struct sigcontext*)puc;
    pc = sc->sigc_regs.tpc;
    sc->sigc_regs.tpc = sc->sigc_regs.tnpc;
    sc->sigc_regs.tnpc += 4;
#elif defined(__sparc__)
    /* on SPARC the 3rd parameter points to a sigcontext not a ucontext */
    struct sigcontext *sc = (struct sigcontext*)puc;
    pc = sc->si_regs.pc;
    sc->si_regs.pc = sc->si_regs.npc;
    sc->si_regs.npc = (unsigned long)sc->si_regs.npc + 4;
#elif defined(__powerpc__)
#if defined(__powerpc64__)
    mcontext_t *mc = &uc->uc_mcontext;
    unsigned long *regs = &mc->gp_regs[0];
#else
    mcontext_t *mc = uc->uc_mcontext.uc_regs;
    unsigned long *regs = &mc->gregs[0];
#endif
    pc = regs[PT_NIP];
    regs[PT_NIP] += 4;
    regs[PT_FPSCR] = 0x80|0x40|0x10;	/* VE, OE, ZE; not UE or XE */
#endif
#elif defined(__DARWIN__) && (defined(__i386__) || defined(__x86_64__))
#ifdef DARWIN_MODERN_MCONTEXT
    mcontext_t mc = uc->uc_mcontext;
    pc = mc_pc(mc);
    mc->__fs.__fpu_mxcsr = 0x1F80;
    *(unsigned short *)&mc->__fs.__fpu_fsw &= ~0xFF;
#else
    mcontext_t mc = uc->uc_mcontext;
    pc = mc_pc(mc);
    mc->fs.fpu_mxcsr = 0x1F80;
    *(unsigned short *)&mc->fs.fpu_fsw &= ~0xFF;
#endif /* DARWIN_MODERN_MCONTEXT */
#elif defined(__DARWIN__) && defined(__ppc__)
    mcontext_t mc = uc->uc_mcontext;
    pc = mc->ss.srr0;
    mc->ss.srr0 += 4;
    mc->fs.fpscr = 0x80|0x40|0x10;
#elif defined(__FreeBSD__) && defined(__x86_64__)
    mcontext_t *mc = &uc->uc_mcontext;
    struct savefpu *savefpu = (struct savefpu*)&mc->mc_fpstate;
    struct envxmm *envxmm = &savefpu->sv_env;
    pc = mc_pc(mc);
    envxmm->en_mxcsr = 0x1F80;
    envxmm->en_sw &= ~0xFF;
#elif defined(__FreeBSD__) && defined(__i386__)
    mcontext_t *mc = &uc->uc_mcontext;
    union savefpu *savefpu = (union savefpu*)&mc->mc_fpstate;
    pc = mc_pc(mc);
    if (mc->mc_fpformat == _MC_FPFMT_XMM) {
	struct envxmm *envxmm = &savefpu->sv_xmm.sv_env;
	envxmm->en_mxcsr = 0x1F80;
	envxmm->en_sw &= ~0xFF;
    } else {
	struct env87 *env87 = &savefpu->sv_87.sv_env;
	env87->en_sw &= ~0xFF;
    }
#elif defined(__OpenBSD__) && defined(__x86_64__)
    struct fxsave64 *fxsave = uc->sc_fpstate;
    pc = mc_pc(uc);
    fxsave->fx_mxcsr = 0x1F80;
    fxsave->fx_fsw &= ~0xFF;
#elif defined(__sun__) && defined(__x86_64__)
    mcontext_t *mc = &uc->uc_mcontext;
    struct fpchip_state *fpstate = &mc->fpregs.fp_reg_set.fpchip_state;
    pc = mc_pc(mc);
    fpstate->mxcsr = 0x1F80;
    fpstate->sw &= ~0xFF;
#endif
#if 0
    {
	char buf[64];
	snprintf(buf, sizeof buf, "%s: FPE at %p\r\n", __FUNCTION__, (void*)pc);
	write(2, buf, strlen(buf));
    }
#endif
    set_current_fp_exception(pc);
}

static void erts_thread_catch_fp_exceptions(void)
{
    struct sigaction act;
    memset(&act, 0, sizeof act);
    act.sa_sigaction = fpe_sig_action;
    act.sa_flags = SA_SIGINFO;
    sigaction(SIGFPE, &act, NULL);
    unmask_fpe();
}

#else  /* !((__linux__ && (__i386__ || __x86_64__ || __powerpc__)) || (__DARWIN__ && (__i386__ || __x86_64__ || __ppc__))) */

static void fpe_sig_handler(int sig)
{
    set_current_fp_exception(1); /* XXX: convert to sigaction so we can get the trap PC */
}

static void erts_thread_catch_fp_exceptions(void)
{
    sys_sigset(SIGFPE, fpe_sig_handler);
    unmask_fpe();
}

#endif /* (__linux__ && (__i386__ || __x86_64__ || __powerpc__)) || (__DARWIN__ && (__i386__ || __x86_64__ || __ppc__))) */

/* once-only initialisation early in the main thread */
void erts_sys_init_float(void)
{
    erts_init_fp_exception();
    erts_thread_catch_fp_exceptions();
    erts_printf_block_fpe = erts_sys_block_fpe;
    erts_printf_unblock_fpe = erts_sys_unblock_fpe;
}

#endif /* NO_FPE_SIGNALS */

void erts_thread_init_float(void)
{
#ifdef ERTS_SMP
    /* This allows Erlang schedulers to leave Erlang-process context
       and still have working FP exceptions. XXX: is this needed? */
    erts_thread_init_fp_exception();
#endif

#ifndef NO_FPE_SIGNALS
    /* NOTE:
     *  erts_thread_disable_fpe() is called in all threads at
     *  creation. We at least need to call unmask_fpe()
     */
#if defined(__DARWIN__) || defined(__FreeBSD__)
    /* Darwin (7.9.0) does not appear to propagate FP exception settings
       to a new thread from its parent. So if we want FP exceptions, we
       must manually re-enable them in each new thread.
       FreeBSD 6.1 appears to suffer from a similar issue. */
    erts_thread_catch_fp_exceptions();
#else
    unmask_fpe();
#endif

#endif
}

void erts_thread_disable_fpe(void)
{
#if !defined(NO_FPE_SIGNALS)
    (void)mask_fpe();
#endif
}

#if !defined(NO_FPE_SIGNALS)
int erts_sys_block_fpe(void)
{
    return mask_fpe();
}

void erts_sys_unblock_fpe(int unmasked)
{
    unmask_fpe_conditional(unmasked);
}
#endif

/* The following check is incorporated from the Vee machine */
    
#define ISDIGIT(d) ((d) >= '0' && (d) <= '9')

/* 
 ** Convert a double to ascii format 0.dddde[+|-]ddd
 ** return number of characters converted
 **
 ** These two functions should maybe use localeconv() to pick up
 ** the current radix character, but since it is uncertain how
 ** expensive such a system call is, and since no-one has heard
 ** of other radix characters than '.' and ',' an ad-hoc 
 ** low execution time solution is used instead.
 */

int
sys_double_to_chars(double fp, char *buf)
{
    char *s = buf;
    
    (void) sprintf(buf, "%.20e", fp);
    /* Search upto decimal point */
    if (*s == '+' || *s == '-') s++;
    while (ISDIGIT(*s)) s++;
    if (*s == ',') *s++ = '.'; /* Replace ',' with '.' */
    /* Scan to end of string */
    while (*s) s++;
    return s-buf; /* i.e strlen(buf) */
}

/* Float conversion */

int
sys_chars_to_double(char* buf, double* fp)
{
#ifndef NO_FPE_SIGNALS
    volatile unsigned long *fpexnp = erts_get_current_fp_exception();
#endif
    char *s = buf, *t, *dp;

    /* Robert says that something like this is what he really wanted:
     * (The [.,] radix test is NOT what Robert wanted - it was added later)
     *
     * 7 == sscanf(Tbuf, "%[+-]%[0-9][.,]%[0-9]%[eE]%[+-]%[0-9]%s", ....);
     * if (*s2 == 0 || *s3 == 0 || *s4 == 0 || *s6 == 0 || *s7)
     *   break;
     */

    /* Scan string to check syntax. */
    if (*s == '+' || *s == '-') s++;
    if (!ISDIGIT(*s))		/* Leading digits. */
      return -1;
    while (ISDIGIT(*s)) s++;
    if (*s != '.' && *s != ',')	/* Decimal part. */
      return -1;
    dp = s++;			/* Remember decimal point pos just in case */
    if (!ISDIGIT(*s))
      return -1;
    while (ISDIGIT(*s)) s++;
    if (*s == 'e' || *s == 'E') {
	/* There is an exponent. */
	s++;
	if (*s == '+' || *s == '-') s++;
	if (!ISDIGIT(*s))
	  return -1;
	while (ISDIGIT(*s)) s++;
    }
    if (*s)			/* That should be it */
      return -1;

#ifdef NO_FPE_SIGNALS
    errno = 0;
#endif
    __ERTS_FP_CHECK_INIT(fpexnp);
    *fp = strtod(buf, &t);
    __ERTS_FP_ERROR_THOROUGH(fpexnp, *fp, return -1);
    if (t != s) {		/* Whole string not scanned */
	/* Try again with other radix char */
	*dp = (*dp == '.') ? ',' : '.';
	errno = 0;
	__ERTS_FP_CHECK_INIT(fpexnp);
	*fp = strtod(buf, &t);
	__ERTS_FP_ERROR_THOROUGH(fpexnp, *fp, return -1);
    }

#ifdef NO_FPE_SIGNALS
    if (errno == ERANGE && (*fp == 0.0 || *fp == HUGE_VAL || *fp == -HUGE_VAL)) {
	return -1;
    }
#endif
    return 0;
}

int
matherr(struct exception *exc)
{
#if !defined(NO_FPE_SIGNALS)
    volatile unsigned long *fpexnp = erts_get_current_fp_exception();
    if (fpexnp != NULL)
	*fpexnp = (unsigned long)__builtin_return_address(0);
#endif
    return 1;
}