http://sourceware.org/bugzilla/show_bug.cgi?id=15836
Bug ID: 15836
Summary: Undefined behavior in gas (causes segfault when built
with clang)
Product: binutils
Version: 2.24 (HEAD)
Status: NEW
Severity: normal
Priority: P2
Component: gas
Assignee: unassigned at sourceware dot org
Reporter: sourceware at wdtz dot org
Created attachment 7144
--> http://sourceware.org/bugzilla/attachment.cgi?id=7144&action=edit
valgrind log from step 5
Description
-----------
The error in question can be seen here:
http://sourceware.org/git/?p=binutils.git;a=blob;f=gas/config/tc-i386-intel.c;h=3f6b057613451839c796ca8a9cdbef2fe6532ec6;hb=HEAD#l432
Where the code assumes unsigned integer wrapping semantics for pointer
arithmetic on the variable 'scale' in a number of places. In particular,
the check:
432 if (ret && scale && (scale + 1))
Gets optimized to "if (ret && scale)" because it is impossible for "scale + 1"
to evaluate to NULL without invoking undefined behavior. Note that the earlier
decrement from NULL is also invalid, and possibly other constructs in related
code.
This is is a problem as it results in the conditional being taken when scale is
"(int*)-sizeof(int)", which leads to an invalid pointer being dereferenced in
resolve_expression().
Steps to reproduce
------------------
1) Obtain and unpack binutils 2.22 or latest via git (tested with 0b0b7b5).
2) Obtain clang 3.3 or latest trunk (from your package manager or build) and
modify PATH as appropriate.
3) Configure similar to the following:
$ CC=clang CXX=clang++ ./configure --disable-werror --enable=ld=no
4) Build.
$ make -j
5) Run the just-built 'as' using the following program from the testsuite:
$ valgrind gas/as-new --32 gas/testsuite/gas/i386/intelbad.s
6) Observe segfault, see attached 'valgrind.log' for the output of the above
command.
Impact
-------
Presently prevents building a functional binutils with recent versions of
clang, and is a time-bomb for breaking future builds. Compilers (including gcc
and clang) are known to increasingly take advantage of undefined behavior in
newer versions and so this may be an issue in the future even with
compilers/platforms that safely build this today.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
bug-binutils mailing list
bug-binutils@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-binutils
