Fix nwfpe so GDB can debug user space floating point again.

Patch 960/1 (Peter Teichmann):
   NWFPE patch to be more compliant to IEEE-754

1. The RND/URD instruction was handled as int_to_float(float_to_int
   (number)) which is wrong because it only works for floating point
   numbers that fit in an integer.

2. The FLT instruction was setting the rounding precision for
   extended precision calculations, which is not necessary
   (probably a historic relict) but has undesirable side effects
   on all extended precision calculations.

arch/arm/nwfpe/ChangeLog

+2002-01-19  Russell King <rmk@arm.linux.org.uk>
+
+	* fpa11.h - Add documentation
+	          - remove userRegisters pointer from this structure.
+	          - add new method to obtain integer register values.
+	* softfloat.c - Remove float128
+	* softfloat.h - Remove float128
+	* softfloat-specialize - Remove float128
+
+	* The FPA11 structure is not a kernel-specific data structure.
+	  It is used by users of ptrace to examine the values of the
+	  floating point registers.  Therefore, any changes to the
+	  FPA11 structure (size or position of elements contained
+	  within) have to be well thought out.
+
+	* Since 128-bit float requires the FPA11 structure to change
+	  size, it has been removed.  128-bit float is currently unused,
+	  and needs various things to be re-worked so that we won't
+	  overflow the available space in the task structure.
+
+	* The changes are designed to break any patch that goes on top
+	  of this code, so that the authors properly review their changes.
+
 1999-08-19  Scott Bambrough  <scottb@netwinder.org>
 
 	* fpmodule.c - Changed version number to 0.95

arch/arm/nwfpe/double_cpdo.c

@@ -19,9 +19,9 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
 #include "softfloat.h"
 #include "fpopcode.h"
-#include "fpa11.h"
 
 float64 float64_exp(float64 Fm);
 float64 float64_ln(float64 Fm);
@@ -165,8 +165,7 @@ unsigned int DoubleCPDO(const unsigned int opcode)
 
       case RND_CODE:
       case URD_CODE:
-         fpa11->fpreg[Fd].fDouble = 
-             int32_to_float64(float64_to_int32(rFm));
+         fpa11->fpreg[Fd].fDouble = float64_round_to_int(rFm);
       break;
 
       case SQT_CODE:

arch/arm/nwfpe/entry.S

@@ -52,8 +52,6 @@ This routine does three things:
 
 1) The kernel has created a struct pt_regs on the stack and saved the
 user registers into it.  See /usr/include/asm/proc/ptrace.h for details.
-The emulator code uses userRegisters as the base of an array of words
-from which the contents of the registers can be extracted.
 
 2) It calls EmulateAll to emulate a floating point instruction.
 EmulateAll returns 1 if the emulation was successful, or 0 if not.
@@ -72,14 +70,9 @@ several floating point instructions.  */
 
 	.globl	nwfpe_enter
 nwfpe_enter:
-	/* ?? Could put userRegisters and fpa11 into fixed regs during
-	   emulation.  This would reduce load/store overhead at the expense
-	   of stealing two regs from the register allocator.  Not sure if
-	   it's worth it.  */
-	str sp, [r10]                   @ Store the user registers pointer in the fpa11 structure.
         mov r4, lr			@ save the failure-return addresses
+	mov sl, sp
 
-	mov r0, r10
 	bl FPA11_CheckInit		@ check to see if we are initialised
 
         ldr r5, [sp, #60]	 	@ get contents of PC;

arch/arm/nwfpe/entry26.S

@@ -65,18 +65,10 @@ several floating point instructions.  */
 
 	.globl	nwfpe_enter
 nwfpe_enter:
-	ldr	r4, =userRegisters
-	str	sp, [r4]		@ save pointer to user regs
+	mov	sl, sp
+	bl	FPA11_CheckInit		@ check to see if we are initialised
 
-	mov	r10, sp, lsr #13	@ find workspace
-	mov	r10, r10, lsl #13
-	add	r10, r10, #TSS_FPESAVE
-
-	ldr	r4, =fpa11
-	str	r10, [r4]		@ store pointer to our state
-	mov	r4, sp			@ use r4 for local pointer
-
-	ldr	r5, [r4, #60]		@ get contents of PC
+	ldr	r5, [sp, #60]		@ get contents of PC
 	bic	r5, r5, #0xfc000003
 	ldr	r0, [r5, #-4]		@ get actual instruction into r0
 	bl	EmulateAll		@ emulate the instruction
@@ -93,10 +85,10 @@ next:
 	teqne	r2, #0x0E000000
 	bne	ret_from_exception	@ return ok if not a fp insn
 
-	ldr	r9, [r4, #60]		@ get new condition codes
+	ldr	r9, [sp, #60]		@ get new condition codes
 	and	r9, r9, #0xfc000003
 	orr	r7, r5, r9
-	str	r7, [r4, #60]		@ update PC copy in regs
+	str	r7, [sp, #60]		@ update PC copy in regs
 
 	mov	r0, r6			@ save a copy
 	mov	r1, r9			@ fetch the condition codes

arch/arm/nwfpe/extended_cpdo.c

@@ -19,9 +19,9 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
 #include "softfloat.h"
 #include "fpopcode.h"
-#include "fpa11.h"
 
 floatx80 floatx80_exp(floatx80 Fm);
 floatx80 floatx80_ln(floatx80 Fm);
@@ -157,8 +157,7 @@ unsigned int ExtendedCPDO(const unsigned int opcode)
 
       case RND_CODE:
       case URD_CODE:
-         fpa11->fpreg[Fd].fExtended = 
-             int32_to_floatx80(floatx80_to_int32(rFm));
+         fpa11->fpreg[Fd].fExtended = floatx80_round_to_int(rFm);
       break;
 
       case SQT_CODE:

arch/arm/nwfpe/fpa11.c

@@ -18,8 +18,6 @@
     along with this program; if not, write to the Free Software
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
-#include <linux/compiler.h>
-#include <asm/system.h>
 
 #include "fpa11.h"
 #include "fpopcode.h"
@@ -27,6 +25,9 @@
 #include "fpmodule.h"
 #include "fpmodule.inl"
 
+#include <linux/compiler.h>
+#include <asm/system.h>
+
 /* forward declarations */
 unsigned int EmulateCPDO(const unsigned int);
 unsigned int EmulateCPDT(const unsigned int);
@@ -56,6 +57,7 @@ void resetFPA11(void)
 void SetRoundingMode(const unsigned int opcode)
 {
 #if MAINTAIN_FPCR
+   FPA11 *fpa11 = GET_FPA11();
    fpa11->fpcr &= ~MASK_ROUNDING_MODE;
 #endif   
    switch (opcode & MASK_ROUNDING_MODE)
@@ -94,6 +96,7 @@ void SetRoundingMode(const unsigned int opcode)
 void SetRoundingPrecision(const unsigned int opcode)
 {
 #if MAINTAIN_FPCR
+   FPA11 *fpa11 = GET_FPA11();
    fpa11->fpcr &= ~MASK_ROUNDING_PRECISION;
 #endif   
    switch (opcode & MASK_ROUNDING_PRECISION)
@@ -123,8 +126,9 @@ void SetRoundingPrecision(const unsigned int opcode)
   }
 }
 
-void FPA11_CheckInit(FPA11 *fpa11)
+void FPA11_CheckInit(void)
 {
+  FPA11 *fpa11 = GET_FPA11();
   if (unlikely(fpa11->initflag == 0))
   {
     resetFPA11();

arch/arm/nwfpe/fpa11.h

@@ -22,45 +22,66 @@
 #ifndef __FPA11_H__
 #define __FPA11_H__
 
+#define GET_FPA11() ((FPA11 *)(&current_thread_info()->fpstate))
+
+/*
+ * The processes registers are always at the very top of the 8K
+ * stack+task struct.  Use the same method as 'current' uses to
+ * reach them.
+ */
+register unsigned int *user_registers asm("sl");
+
+#define GET_USERREG() (user_registers)
+
+#include <linux/thread_info.h>
+
 /* includes */
 #include "fpsr.h"		/* FP control and status register definitions */
 #include "softfloat.h"
 
-/* Need task_struct */
-#include <linux/sched.h>
-
 #define		typeNone		0x00
 #define		typeSingle		0x01
 #define		typeDouble		0x02
 #define		typeExtended		0x03
 
+/*
+ * This must be no more and no less than 12 bytes.
+ */
 typedef union tagFPREG {
-   float32  fSingle;
-   float64  fDouble;
    floatx80 fExtended;
+   float64  fDouble;
+   float32  fSingle;
 } FPREG;
 
-/* FPA11 device model */
+/*
+ * FPA11 device model.
+ *
+ * This structure is exported to user space.  Do not re-order.
+ * Only add new stuff to the end, and do not change the size of
+ * any element.  Elements of this structure are used by user
+ * space, and must match struct user_fp in include/asm-arm/user.h.
+ * We include the byte offsets below for documentation purposes.
+ *
+ * The size of this structure and FPREG are checked by fpmodule.c
+ * on initialisation.  If the rules have been broken, NWFPE will
+ * not initialise.
+ */
 typedef struct tagFPA11 {
-  unsigned int *userRegisters;
-  FPREG fpreg[8];		/* 8 floating point registers */
-  FPSR fpsr;			/* floating point status register */
-  FPCR fpcr;			/* floating point control register */
-  unsigned char fType[8];	/* type of floating point value held in
-				   floating point registers.  One of none
-				   single, double or extended. */
-  int initflag;			/* this is special.  The kernel guarantees
-				   to set it to 0 when a thread is launched,
-				   so we can use it to detect whether this
-				   instance of the emulator needs to be
-				   initialised. */
+/*   0 */  FPREG fpreg[8];		/* 8 floating point registers */
+/*  96 */  FPSR fpsr;			/* floating point status register */
+/* 100 */  FPCR fpcr;			/* floating point control register */
+/* 104 */  unsigned char fType[8];	/* type of floating point value held in
+					   floating point registers.  One of none
+					   single, double or extended. */
+/* 112 */  int initflag;		/* this is special.  The kernel guarantees
+					   to set it to 0 when a thread is launched,
+					   so we can use it to detect whether this
+					   instance of the emulator needs to be
+					   initialised. */
 } FPA11;
 
 extern void resetFPA11(void);
 extern void SetRoundingMode(const unsigned int);
 extern void SetRoundingPrecision(const unsigned int);
 
-#define GET_FPA11() ((FPA11 *)(&current->thread.fpstate))
-#define GET_USERREG() (GET_FPA11()->userRegisters)
-
 #endif

arch/arm/nwfpe/fpa11_cpdt.c

@@ -20,9 +20,9 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
 #include "softfloat.h"
 #include "fpopcode.h"
-#include "fpa11.h"
 #include "fpmodule.h"
 #include "fpmodule.inl"
 

arch/arm/nwfpe/fpa11_cprt.c

@@ -20,10 +20,10 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
 #include "milieu.h"
 #include "softfloat.h"
 #include "fpopcode.h"
-#include "fpa11.h"
 #include "fpa11.inl"
 #include "fpmodule.h"
 #include "fpmodule.inl"
@@ -82,8 +82,7 @@ unsigned int PerformFLT(const unsigned int opcode)
    
    unsigned int nRc = 1;
    SetRoundingMode(opcode);
-   SetRoundingPrecision(opcode);
-   
+
    switch (opcode & MASK_ROUNDING_PRECISION)
    {
       case ROUND_SINGLE:

arch/arm/nwfpe/fpmodule.c

@@ -21,6 +21,8 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
+
 #include <linux/module.h>
 #include <linux/version.h>
 #include <linux/config.h>
@@ -36,7 +38,6 @@
 #include "softfloat.h"
 #include "fpopcode.h"
 #include "fpmodule.h"
-#include "fpa11.h"
 #include "fpa11.inl"
 
 /* kernel symbols required for signal handling */
@@ -86,6 +87,11 @@ static int __init fpe_init(void)
     return -EINVAL;
   }
 
+  if (sizeof(FPREG) != 12) {
+    printk(KERN_ERR "nwfpe: bad register size\n");
+    return -EINVAL;
+  }
+
 #ifdef MODULE
   if (!mod_member_present(&__this_module, can_unload))
     return -EINVAL;

arch/arm/nwfpe/fpopcode.c

@@ -19,10 +19,10 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
 #include "softfloat.h"
 #include "fpopcode.h"
 #include "fpsr.h"
-#include "fpa11.h"
 #include "fpmodule.h"
 #include "fpmodule.inl"
 

arch/arm/nwfpe/single_cpdo.c

@@ -19,9 +19,9 @@
     Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#include "fpa11.h"
 #include "softfloat.h"
 #include "fpopcode.h"
-#include "fpa11.h"
 
 float32 float32_exp(float32 Fm);
 float32 float32_ln(float32 Fm);
@@ -139,8 +139,7 @@ unsigned int SingleCPDO(const unsigned int opcode)
 
       case RND_CODE:
       case URD_CODE:
-         fpa11->fpreg[Fd].fSingle = 
-             int32_to_float32(float32_to_int32(rFm));
+         fpa11->fpreg[Fd].fSingle = float32_round_to_int(rFm);
       break;
 
       case SQT_CODE:

arch/arm/nwfpe/softfloat-specialize

@@ -364,108 +364,3 @@ static floatx80 propagateFloatx80NaN( floatx80 a, floatx80 b )
 }
 
 #endif
-
-#ifdef FLOAT128
-
-/*
--------------------------------------------------------------------------------
-The pattern for a default generated quadruple-precision NaN.  The `high' and
-`low' values hold the most- and least-significant bits, respectively.
--------------------------------------------------------------------------------
-*/
-#define float128_default_nan_high LIT64( 0xFFFFFFFFFFFFFFFF )
-#define float128_default_nan_low  LIT64( 0xFFFFFFFFFFFFFFFF )
-
-/*
--------------------------------------------------------------------------------
-Returns 1 if the quadruple-precision floating-point value `a' is a NaN;
-otherwise returns 0.
--------------------------------------------------------------------------------
-*/
-flag float128_is_nan( float128 a )
-{
-
-    return
-           ( LIT64( 0xFFFE000000000000 ) <= (bits64) ( a.high<<1 ) )
-        && ( a.low || ( a.high & LIT64( 0x0000FFFFFFFFFFFF ) ) );
-
-}
-
-/*
--------------------------------------------------------------------------------
-Returns 1 if the quadruple-precision floating-point value `a' is a
-signaling NaN; otherwise returns 0.
--------------------------------------------------------------------------------
-*/
-flag float128_is_signaling_nan( float128 a )
-{
-
-    return
-           ( ( ( a.high>>47 ) & 0xFFFF ) == 0xFFFE )
-        && ( a.low || ( a.high & LIT64( 0x00007FFFFFFFFFFF ) ) );
-
-}
-
-/*
--------------------------------------------------------------------------------
-Returns the result of converting the quadruple-precision floating-point NaN
-`a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
-exception is raised.
--------------------------------------------------------------------------------
-*/
-static commonNaNT float128ToCommonNaN( float128 a )
-{
-    commonNaNT z;
-
-    if ( float128_is_signaling_nan( a ) ) float_raise( float_flag_invalid );
-    z.sign = a.high>>63;
-    shortShift128Left( a.high, a.low, 16, &z.high, &z.low );
-    return z;
-
-}
-
-/*
--------------------------------------------------------------------------------
-Returns the result of converting the canonical NaN `a' to the quadruple-
-precision floating-point format.
--------------------------------------------------------------------------------
-*/
-static float128 commonNaNToFloat128( commonNaNT a )
-{
-    float128 z;
-
-    shift128Right( a.high, a.low, 16, &z.high, &z.low );
-    z.high |= ( ( (bits64) a.sign )<<63 ) | LIT64( 0x7FFF800000000000 );
-    return z;
-
-}
-
-/*
--------------------------------------------------------------------------------
-Takes two quadruple-precision floating-point values `a' and `b', one of
-which is a NaN, and returns the appropriate NaN result.  If either `a' or
-`b' is a signaling NaN, the invalid exception is raised.
--------------------------------------------------------------------------------
-*/
-static float128 propagateFloat128NaN( float128 a, float128 b )
-{
-    flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN;
-
-    aIsNaN = float128_is_nan( a );
-    aIsSignalingNaN = float128_is_signaling_nan( a );
-    bIsNaN = float128_is_nan( b );
-    bIsSignalingNaN = float128_is_signaling_nan( b );
-    a.high |= LIT64( 0x0000800000000000 );
-    b.high |= LIT64( 0x0000800000000000 );
-    if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid );
-    if ( aIsNaN ) {
-        return ( aIsSignalingNaN & bIsNaN ) ? b : a;
-    }
-    else {
-        return b;
-    }
-
-}
-
-#endif
-

arch/arm/nwfpe/softfloat.h

@@ -37,12 +37,10 @@ this code that are retained.
 The macro `FLOATX80' must be defined to enable the extended double-precision
 floating-point format `floatx80'.  If this macro is not defined, the
 `floatx80' type will not be defined, and none of the functions that either
-input or output the `floatx80' type will be defined.  The same applies to
-the `FLOAT128' macro and the quadruple-precision format `float128'.
+input or output the `floatx80' type will be defined.
 -------------------------------------------------------------------------------
 */
 #define FLOATX80
-/* #define FLOAT128 */
 
 /*
 -------------------------------------------------------------------------------
@@ -51,17 +49,10 @@ Software IEC/IEEE floating-point types.
 */
 typedef unsigned long int float32;
 typedef unsigned long long float64;
-#ifdef FLOATX80
 typedef struct {
     unsigned short high;
     unsigned long long low;
 } floatx80;
-#endif
-#ifdef FLOAT128
-typedef struct {
-    unsigned long long high, low;
-} float128;
-#endif
 
 /*
 -------------------------------------------------------------------------------
@@ -131,9 +122,6 @@ float64 int32_to_float64( signed int );
 #ifdef FLOATX80
 floatx80 int32_to_floatx80( signed int );
 #endif
-#ifdef FLOAT128
-float128 int32_to_float128( signed int );
-#endif
 
 /*
 -------------------------------------------------------------------------------
@@ -146,9 +134,6 @@ float64 float32_to_float64( float32 );
 #ifdef FLOATX80
 floatx80 float32_to_floatx80( float32 );
 #endif
-#ifdef FLOAT128
-float128 float32_to_float128( float32 );
-#endif
 
 /*
 -------------------------------------------------------------------------------
@@ -181,9 +166,6 @@ float32 float64_to_float32( float64 );
 #ifdef FLOATX80
 floatx80 float64_to_floatx80( float64 );
 #endif
-#ifdef FLOAT128
-float128 float64_to_float128( float64 );
-#endif
 
 /*
 -------------------------------------------------------------------------------
@@ -216,9 +198,6 @@ signed int floatx80_to_int32( floatx80 );
 signed int floatx80_to_int32_round_to_zero( floatx80 );
 float32 floatx80_to_float32( floatx80 );
 float64 floatx80_to_float64( floatx80 );
-#ifdef FLOAT128
-float128 floatx80_to_float128( floatx80 );
-#endif
 
 /*
 -------------------------------------------------------------------------------
@@ -250,41 +229,4 @@ char floatx80_is_signaling_nan( floatx80 );
 
 #endif
 
-#ifdef FLOAT128
-
-/*
--------------------------------------------------------------------------------
-Software IEC/IEEE quadruple-precision conversion routines.
--------------------------------------------------------------------------------
-*/
-signed int float128_to_int32( float128 );
-signed int float128_to_int32_round_to_zero( float128 );
-float32 float128_to_float32( float128 );
-float64 float128_to_float64( float128 );
-#ifdef FLOATX80
-floatx80 float128_to_floatx80( float128 );
-#endif
-
-/*
--------------------------------------------------------------------------------
-Software IEC/IEEE quadruple-precision operations.
--------------------------------------------------------------------------------
-*/
-float128 float128_round_to_int( float128 );
-float128 float128_add( float128, float128 );
-float128 float128_sub( float128, float128 );
-float128 float128_mul( float128, float128 );
-float128 float128_div( float128, float128 );
-float128 float128_rem( float128, float128 );
-float128 float128_sqrt( float128 );
-char float128_eq( float128, float128 );
-char float128_le( float128, float128 );
-char float128_lt( float128, float128 );
-char float128_eq_signaling( float128, float128 );
-char float128_le_quiet( float128, float128 );
-char float128_lt_quiet( float128, float128 );
-char float128_is_signaling_nan( float128 );
-
-#endif
-
 #endif