On 01/28/2018 01:06 PM, Josh Poimboeuf wrote:
On Mon, Jan 22, 2018 at 12:25:22PM -0800, Guenter Roeck wrote:
Hi David,
On Mon, Jan 22, 2018 at 07:34:04PM +0000, David Woodhouse wrote:
On Thu, 2018-01-18 at 11:41 -0800, Guenter Roeck wrote:
Not sure, does your gcc have retpolines? Give me your .o file and I can
diagnose it.
Yes, it does, only it is the gcc from the Google toolchain which may
generate different code than the upstream version.
I attached an affected object file. Please let me know if there is anything else
I can do to help.
Disassembly of section .text.__x86.indirect_thunk:
0000000000000000 <__x86.indirect_thunk>:
0: e8 04 00 00 00 callq 9 <__x86.indirect_thunk+0x9>
5: f3 90 pause
7: eb fc jmp 5 <__x86.indirect_thunk+0x5>
9: 48 8d 64 24 08 lea 0x8(%rsp),%rsp
e: c3 retq
That has the old-style CET-incompatible retpoline in a COMDAT section
in the .o file. What compiler options are being used for that? The
kernel should only use retpoline if GCC supports both of
-mindirect-branch=thunk-extern and -mindirect-branch-register, and this
compiler is doing *neither* of those.
It uses "-mindirect-branch=thunk -mindirect-branch-loop=pause
-fno-jump-tables", though I don't know if that even exists in
upstream gcc (it is the gcc use for Chrome OS builds). I'll pass
your feedback to our compiler team.
Either case, I think it is less than optimal that objtool crashes
with _any_ object code.
I've got a pending fix for this, so that objtool doesn't seg fault, and
instead prints out a warning:
quirks.o: warning: objtool: efi_delete_dummy_variable()+0x99: unsupported
intra-function call
quirks.o: warning: objtool: If this is a retpoline, please patch it in with
alternatives and annotate it with ANNOTATE_NOSPEC_ALTERNATIVE.
The code is here, along with a few more fixes:
https://git.kernel.org/pub/scm/linux/kernel/git/jpoimboe/linux.git/log/?h=TODO-objtool-seg-fault
Excellent. I'll give it a try tomorrow.
Thanks for looking into this!
Guenter
