Sathvika Vasireddy wrote:
On 29/11/22 20:58, Christophe Leroy wrote:
Le 29/11/2022 à 16:13, Sathvika Vasireddy a écrit :
Hi all,
On 25/11/22 09:00, Stephen Rothwell wrote:
Hi all,
After merging the powerpc-objtool tree, today's linux-next build (powerpc
pseries_le_defconfig) produced these warnings:
arch/powerpc/kernel/head_64.o: warning: objtool: end_first_256B():
can't find starting instruction
arch/powerpc/kernel/optprobes_head.o: warning: objtool:
optprobe_template_end(): can't find starting instruction
I have no idea what started this (they may have been there yesterday).
I was able to recreate the above mentioned warnings with
pseries_le_defconfig and powernv_defconfig. The regression report also
mentions a warning
(https://lore.kernel.org/oe-kbuild-all/202211282102.QUr7HHrW-lkp@xxxxxxxxx/) seen with arch/powerpc/kernel/kvm_emul.S assembly file.
[1] arch/powerpc/kernel/optprobes_head.o: warning: objtool:
optprobe_template_end(): can't find starting instruction
[2] arch/powerpc/kernel/kvm_emul.o: warning: objtool:
kvm_template_end(): can't find starting instruction
[3] arch/powerpc/kernel/head_64.o: warning: objtool: end_first_256B():
can't find starting instruction
The warnings [1] and [2] go away after adding 'nop' instruction. Below
diff fixes it for me:
You have to add NOPs just because those labels are at the end of the
files. That's a bit odd.
I think either we are missing some kind of flagging for the symbols, or
objtool has a bug. In both cases, I'm not sure adding an artificial
'nop' is the solution. At least there should be a big hammer warning
explaining why.
The problem looks to be that commit dbcdbdfdf137b4 ("objtool: Rework
instruction -> symbol mapping"), which was referenced by Sathvika below,
changes how STT_NOTYPE symbols are handled. In the files throwing that
warning, there are labels either at the very end of the file, or at the
end of a section with no subsequent instruction. Before that commit, we
didn't used to expect an instruction for STT_NOTYPE symbols.
I don't see these warnings with powerpc/topic/objtool branch. However,
they are seen with linux-next master branch.
Commit dbcdbdfdf137b49144204571f1a5e5dc01b8aaad objtool: Rework
instruction -> symbol mapping in linux-next is resulting in objtool
can't find starting instruction warnings on powerpc.
Reverting this particular hunk (pasted below), resolves it and we don't
see the problem anymore.
@@ -427,7 +427,10 @@ static int decode_instructions(struct objtool_file
*file)
}
list_for_each_entry(func, &sec->symbol_list, list) {
- if (func->type != STT_FUNC || func->alias != func)
+ if (func->type != STT_NOTYPE && func->type !=
STT_FUNC)
+ continue;
+
+ if (func->return_thunk || func->alias != func)
continue;
if (!find_insn(file, sec, func->offset)) {
We are currently bailing out if find_insn() there fails. Should we
instead just continue by not setting insn->sym?
@@ -430,11 +430,8 @@ static int decode_instructions(struct objtool_file *file)
if (func->return_thunk || func->alias != func)
continue;
- if (!find_insn(file, sec, func->offset)) {
- WARN("%s(): can't find starting instruction",
- func->name);
- return -1;
- }
+ if (!find_insn(file, sec, func->offset))
+ continue;
sym_for_each_insn(file, func, insn) {
insn->sym = func;
- Naveen