STT_NOTYPE symbols with LSB set interfere with disassembly
.global foo
.type foo, %function
foo:
nop
nop
.size foo, . - fooWhen compiled and linked with -Wl,--defsym=bar=foo, disassembling gets confused by the STT_NOTYPE symbol with its LSB set, and tries to "re-align" itself with the symbol, breaking itself and ending up off by one:
Jessicas-MacBook-Pro:morello-defsym Jess% morello-llvm llvm-objdump -d foo
foo: file format elf64-littleaarch64
Disassembly of section .text:
0000000000010260 <foo>:
10260: 1f 20 03 d5 nop
0000000000010261 <bar>:
10261: 20 03 d5 1f <unknown>
10265: 20 03 d5 00 <unknown>
...
1027d: 00 <unknown>
1027e: 00 <unknown>
1027f: 00 <unknown>
Jessicas-MacBook-Pro:morello-defsym Jess% morello-llvm llvm-readelf -Ws foo
Symbol table '.dynsym' contains 1 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
Symbol table '.symtab' contains 5 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
1: 0000000000010260 0 NOTYPE LOCAL DEFAULT 6 $c.0
2: 0000000000020280 144 NOTYPE LOCAL HIDDEN 7 _DYNAMIC
3: 0000000000010261 7 FUNC GLOBAL DEFAULT 6 foo
4: 0000000000010261 0 NOTYPE GLOBAL DEFAULT 6 barThe fact that it's STT_NOTYPE (and size 0) rather than inheriting it from the symbol it's being set to an alias of is a bit sad (can't do much about the absolute address case, but in this case we could, though awkward questions arise when you ask what foo-N should put for size, and even foo+N you can't quite express what you really want in the symbol table, which is to have base foo, offset N and length the same as foo's so you'd have to approximate it by foo.len - N and hope nobody tries to access before the start) and a (CHERI-LLVM) upstream deficiency (though in practice, who actually does things like that...), but llvm-objdump should be able to cope with that input and not do clearly-wrong things like try to disassemble from an odd address.