From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Date: Mon, 5 May 2025 15:22:54 +0000 (-0300)
Subject: math: Fix UB in flt-32 and dbl-64 significand
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=b91a1355001faa6b42c6a0361e420ba70a3d0412;p=thirdparty%2Fglibc.git

math: Fix UB in flt-32 and dbl-64 significand

On x86_64:

$ math/test-float-significand
testing float (without inline functions)
UBSAN: Undefined behaviour in ./s_significand_template.c:31:45 negation of 2147483648 cannot be represented in type 'int'
---

diff --git a/math/s_significand_template.c b/math/s_significand_template.c
index d00b0525e4..6a0e3faa70 100644
--- a/math/s_significand_template.c
+++ b/math/s_significand_template.c
@@ -28,7 +28,8 @@
 FLOAT
 M_DECL_FUNC (__significand) (FLOAT x)
 {
-  return M_SUF (__ieee754_scalb) (x,(FLOAT) - M_SUF (__ilogb) (x));
+  int ex = - (unsigned int) M_SUF (__ilogb) (x);
+  return M_SUF (__ieee754_scalb) (x,(FLOAT) ex);
 }
 
 /* Don't provide _FloatN aliases for significand which was originally