[LLVMbugs] [Bug 18881] New: [AArch64] Produce STLXR with unpredictable behavior.

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Tue Feb 18 02:15:09 PST 2014


http://llvm.org/bugs/show_bug.cgi?id=18881

            Bug ID: 18881
           Summary: [AArch64] Produce STLXR with unpredictable behavior.
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: kevinqindev at gmail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 12081
  --> http://llvm.org/bugs/attachment.cgi?id=12081&action=edit
test file to reproduce

With the following command line

$clang --target=aarch64-linux-gnuabi -mfpu=neon -D_GNU_SOURCE
-D__STDC_LIMIT_MACROS -DNDEBUG  -O3 -m64 -fomit-frame-pointer AtomicOps.c

We would see failure as below on foundation model,

$./a.out
[ 8592.507973] a.out[1447]: undefined instruction: pc=00000000004005d0
[ 8592.508058] Code: f94007e0 885ffc01 b94007e2 0b020023 (8803fc03) 
Illegal instruction

The assembly code in 00000000004005d0 is

  4005d0:    8803fc03     stlxr    w3, w3, [x0]

Which result is UNPREDICTABLE.
Compiler shouldn't produce unpredictable instructions, so Ws and Wt must be
different registers.

Below is quoted from ARM reference manual:

STLXR <Ws>, <Wt>, [<Xn|SP>{,#0}]

if memop == MemOp_STORE && excl then 
    if s == t || (pair && s == t2) then 
        Constraint c = ConstrainUnpredictable(Unpredictable_DATAOVERLAP); 
        assert c IN {Constraint_UNKNOWN, Constraint_NONE, Constraint_UNDEF,
Constraint_NOP}; 
        case c of 
            when Constraint_UNKNOWN rt_unknown = TRUE; // store UNKNOWN value 
            when Constraint_NONE rt_unknown = FALSE;   // store original value 
            when Constraint_UNDEF UnallocatedEncoding(); 
            when Constraint_NOP EndOfInstruction();

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20140218/a23a358b/attachment.html>


More information about the llvm-bugs mailing list