<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - MIPS: Thread local variable not flagged as thread local in object file"
href="https://bugs.llvm.org/show_bug.cgi?id=35140">35140</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>MIPS: Thread local variable not flagged as thread local in object file
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Windows NT
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Backend: MIPS
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>alex@crichton.co
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>We've been getting weird linkage errors in rust-lang/rust [1] which I believe
we've minimized to this IR:
@foo = external hidden thread_local global i32
define i32 @bar() {
%a = load i32, i32* @foo
ret i32 %a
}
I don't personally know a lot about MIPS unfortunately, but I think this may
show the problem:
$ llc foo.ll -filetype=obj -o foo.o -mtriple=mipsel-unknown-linux-gnu
-relocation-model=pic
$ readelf -Ws foo.o
Symbol table '.symtab' contains 7 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 00000000 0 NOTYPE LOCAL DEFAULT UND
1: 00000000 0 FILE LOCAL DEFAULT ABS foo.ll
2: 00000000 0 SECTION LOCAL DEFAULT 2
3: 00000000 0 NOTYPE GLOBAL DEFAULT UND __tls_get_addr
4: 00000000 0 NOTYPE GLOBAL DEFAULT UND _gp_disp
5: 00000000 56 FUNC GLOBAL DEFAULT 2 bar
6: 00000000 0 NOTYPE GLOBAL HIDDEN UND foo
Note that the import of `foo` under "Type" says "NOTYPE", whereas normally TLS
variables have "TLS" mentioned there. Also note that if `hidden` is removed
from the IR or the `-relocation-model=pic` argument is dropped then the "Type"
listed is indeed "TLS".
Is this perhaps expected behavior and I'm barking up the wrong tree? I tried
compiling the equivalent C code with gcc to see what it output:
$ cat foo.c
extern __thread int foo __attribute__((visibility("hidden")));
int bar() {
return foo;
}
$ mipsel-linux-gnu-gcc foo.c -c && mipsel-linux-gnu-readelf -Ws foo.o
Symbol table '.symtab' contains 14 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 00000000 0 NOTYPE LOCAL DEFAULT UND
1: 00000000 0 FILE LOCAL DEFAULT ABS foo.c
2: 00000000 0 SECTION LOCAL DEFAULT 1
3: 00000000 0 SECTION LOCAL DEFAULT 3
4: 00000000 0 SECTION LOCAL DEFAULT 4
5: 00000000 0 SECTION LOCAL DEFAULT 9
6: 00000000 0 SECTION LOCAL DEFAULT 5
7: 00000000 0 SECTION LOCAL DEFAULT 6
8: 00000000 0 SECTION LOCAL DEFAULT 7
9: 00000000 0 SECTION LOCAL DEFAULT 10
10: 00000000 0 SECTION LOCAL DEFAULT 11
11: 00000000 60 FUNC GLOBAL DEFAULT 1 bar
12: 00000000 0 NOTYPE GLOBAL DEFAULT UND __gnu_local_gp
13: 00000000 0 TLS GLOBAL HIDDEN UND foo
Note that the reference to `foo` is flagged as TLS and HIDDEN, so that's what
caused me to think that this may be an accidental bug in LLVM's codegen
perhaps?
[1]: <a href="https://travis-ci.org/rust-lang/rust/jobs/293501043">https://travis-ci.org/rust-lang/rust/jobs/293501043</a></pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>