toolchain/musl/patches/030-fix-x87-stack-imbalance-in-corner-cases-of-i386-math.patch

   1 From f3ed8bfe8a82af1870ddc8696ed4cc1d5aa6b441 Mon Sep 17 00:00:00 2001
   2 From: Rich Felker <dalias@aerifal.cx>
   3 Date: Mon, 5 Aug 2019 18:41:47 -0400
   4 Subject: fix x87 stack imbalance in corner cases of i386 math asm
   5
   6 commit 31c5fb80b9eae86f801be4f46025bc6532a554c5 introduced underflow
   7 code paths for the i386 math asm, along with checks on the fpu status
   8 word to skip the underflow-generation instructions if the underflow
   9 flag was already raised. unfortunately, at least one such path, in
  10 log1p, returned with 2 items on the x87 stack rather than just 1 item
  11 for the return value. this is a violation of the ABI's calling
  12 convention, and could cause subsequent floating point code to produce
  13 NANs due to x87 stack overflow. if floating point results are used in
  14 flow control, this can lead to runaway wrong code execution.
  15
  16 rather than reviewing each "underflow already raised" code path for
  17 correctness, remove them all. they're likely slower than just
  18 performing the underflow code unconditionally, and significantly more
  19 complex.
  20
  21 all of this code should be ripped out and replaced by C source files
  22 with inline asm. doing so would preclude this kind of error by having
  23 the compiler perform all x87 stack register allocation and stack
  24 manipulation, and would produce comparable or better code. however
  25 such a change is a much larger project.
  26 ---
  27  src/math/i386/asin.s   | 10 ++--------
  28  src/math/i386/atan.s   |  7 ++-----
  29  src/math/i386/atan2.s  |  5 +----
  30  src/math/i386/atan2f.s |  5 +----
  31  src/math/i386/atanf.s  |  7 ++-----
  32  src/math/i386/exp.s    | 10 ++--------
  33  src/math/i386/log1p.s  |  7 ++-----
  34  src/math/i386/log1pf.s |  7 ++-----
  35  8 files changed, 14 insertions(+), 44 deletions(-)
  36
  37 --- a/src/math/i386/asin.s
  38 +++ b/src/math/i386/asin.s
  39 @@ -7,13 +7,10 @@ asinf:
  40         cmp $0x01000000,%eax
  41         jae 1f
  42                 # subnormal x, return x with underflow
  43 -       fnstsw %ax
  44 -       and $16,%ax
  45 -       jnz 2f
  46         fld %st(0)
  47         fmul %st(1)
  48         fstps 4(%esp)
  49 -2:     ret
  50 +       ret
  51
  52  .global asinl
  53  .type asinl,@function
  54 @@ -30,11 +27,8 @@ asin:
  55         cmp $0x00200000,%eax
  56         jae 1f
  57                 # subnormal x, return x with underflow
  58 -       fnstsw %ax
  59 -       and $16,%ax
  60 -       jnz 2f
  61         fsts 4(%esp)
  62 -2:     ret
  63 +       ret
  64  1:     fld %st(0)
  65         fld1
  66         fsub %st(0),%st(1)
  67 --- a/src/math/i386/atan.s
  68 +++ b/src/math/i386/atan.s
  69 @@ -10,8 +10,5 @@ atan:
  70         fpatan
  71         ret
  72                 # subnormal x, return x with underflow
  73 -1:     fnstsw %ax
  74 -       and $16,%ax
  75 -       jnz 2f
  76 -       fsts 4(%esp)
  77 -2:     ret
  78 +1:     fsts 4(%esp)
  79 +       ret
  80 --- a/src/math/i386/atan2.s
  81 +++ b/src/math/i386/atan2.s
  82 @@ -10,8 +10,5 @@ atan2:
  83         cmp $0x00200000,%eax
  84         jae 1f
  85                 # subnormal x, return x with underflow
  86 -       fnstsw %ax
  87 -       and $16,%ax
  88 -       jnz 1f
  89         fsts 4(%esp)
  90 -1:     ret
  91 +       ret
  92 --- a/src/math/i386/atan2f.s
  93 +++ b/src/math/i386/atan2f.s
  94 @@ -10,10 +10,7 @@ atan2f:
  95         cmp $0x01000000,%eax
  96         jae 1f
  97                 # subnormal x, return x with underflow
  98 -       fnstsw %ax
  99 -       and $16,%ax
 100 -       jnz 1f
 101         fld %st(0)
 102         fmul %st(1)
 103         fstps 4(%esp)
 104 -1:     ret
 105 +       ret
 106 --- a/src/math/i386/atanf.s
 107 +++ b/src/math/i386/atanf.s
 108 @@ -10,10 +10,7 @@ atanf:
 109         fpatan
 110         ret
 111                 # subnormal x, return x with underflow
 112 -1:     fnstsw %ax
 113 -       and $16,%ax
 114 -       jnz 2f
 115 -       fld %st(0)
 116 +1:     fld %st(0)
 117         fmul %st(1)
 118         fstps 4(%esp)
 119 -2:     ret
 120 +       ret
 121 --- a/src/math/i386/exp.s
 122 +++ b/src/math/i386/exp.s
 123 @@ -7,13 +7,10 @@ expm1f:
 124         cmp $0x01000000,%eax
 125         jae 1f
 126                 # subnormal x, return x with underflow
 127 -       fnstsw %ax
 128 -       and $16,%ax
 129 -       jnz 2f
 130         fld %st(0)
 131         fmul %st(1)
 132         fstps 4(%esp)
 133 -2:     ret
 134 +       ret
 135
 136  .global expm1l
 137  .type expm1l,@function
 138 @@ -30,11 +27,8 @@ expm1:
 139         cmp $0x00200000,%eax
 140         jae 1f
 141                 # subnormal x, return x with underflow
 142 -       fnstsw %ax
 143 -       and $16,%ax
 144 -       jnz 2f
 145         fsts 4(%esp)
 146 -2:     ret
 147 +       ret
 148  1:     fldl2e
 149         fmulp
 150         mov $0xc2820000,%eax
 151 --- a/src/math/i386/log1p.s
 152 +++ b/src/math/i386/log1p.s
 153 @@ -16,9 +16,6 @@ log1p:
 154         fyl2x
 155         ret
 156                 # subnormal x, return x with underflow
 157 -2:     fnstsw %ax
 158 -       and $16,%ax
 159 -       jnz 1f
 160 -       fsts 4(%esp)
 161 +2:     fsts 4(%esp)
 162         fstp %st(1)
 163 -1:     ret
 164 +       ret
 165 --- a/src/math/i386/log1pf.s
 166 +++ b/src/math/i386/log1pf.s
 167 @@ -16,10 +16,7 @@ log1pf:
 168         fyl2x
 169         ret
 170                 # subnormal x, return x with underflow
 171 -2:     fnstsw %ax
 172 -       and $16,%ax
 173 -       jnz 1f
 174 -       fxch
 175 +2:     fxch
 176         fmul %st(1)
 177         fstps 4(%esp)
 178 -1:     ret
 179 +       ret