Re: gcc-8 objtool warnings

[Josh Poimboeuf]

On Tue, Sep 19, 2017 at 10:43:31PM +0200, Arnd Bergmann wrote:
> On Tue, Sep 19, 2017 at 9:52 PM, Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> > On Mon, Sep 11, 2017 at 04:34:52PM +0200, Arnd Bergmann wrote:
> >> On Thu, Aug 24, 2017 at 9:19 PM, Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> >> 1. one configuration causing tons of warnings, on most compiler
> >>   versions (4.9 and newer), and even a couple of "unreachable
> >>   instruction warnings on gcc-4.3.
> >>
> >> arch/x86/mm/pageattr.o: warning: objtool: set_memory_x()+0x3a: call
> >> without frame pointer save/setup
> >> security/keys/keyring.o: warning: objtool: keyring_read()+0x70: leave
> >> instruction with modified stack frame
> >> arch/x86/events/intel/pt.o: warning: objtool:
> >> pt_event_addr_filters_sync uses BP as a scratch register
> >
> > I downloaded the tarball, and I see configs/logs for the other problems,
> > but not for this one.  Did you forget to attach it, or did I miss it?
> 
> I think what I ended up doing was to take the smallest files for that
> configuration and packed them up, but the warnings I listed in the
> mail don't match the ones for those files. However, it seems the
> 0xEBFDB964_defconfig/log file got corrupted.
> 
> I've uploaded a new copy to http://paste.ubuntu.com/25574959/ now.
> 
> I've also experimented with other compiler versions
> in the meantime and ran into similar problems on those version but did
> not bother to take detailed notes about them since they are likely just
> variants of the ones I already reported. If you can work around the
> warnings I reported here, I can rerun with all compilers I have to
> see if anything else shows up.

Here's the fix for #1 and #2.  I still need to work on the GCC 8
warnings.

-----
From: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
Subject: [PATCH] objtool: Support unoptimized frame pointer setup

Arnd Bergmann reported a bunch of warnings like:

  crypto/jitterentropy.o: warning: objtool: jent_fold_time()+0x3b: call without frame pointer save/setup
  crypto/jitterentropy.o: warning: objtool: jent_stuck()+0x1d: call without frame pointer save/setup
  crypto/jitterentropy.o: warning: objtool: jent_unbiased_bit()+0x15: call without frame pointer save/setup
  crypto/jitterentropy.o: warning: objtool: jent_read_entropy()+0x32: call without frame pointer save/setup
  crypto/jitterentropy.o: warning: objtool: jent_entropy_collector_free()+0x19: call without frame pointer save/setup

and

  arch/x86/events/core.o: warning: objtool: collect_events uses BP as a scratch register
  arch/x86/events/core.o: warning: objtool: events_ht_sysfs_show()+0x22: call without frame pointer save/setup

With certain rare configurations, GCC sometimes sets up the frame
pointer with:

  lea    (%rsp),%rbp

instead of:

  mov    %rsp,%rbp

The instructions are equivalent, so treat the former like the latter.

Reported-by: Arnd Bergmann <arnd@xxxxxxxx>
Signed-off-by: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
---
 tools/objtool/arch/x86/decode.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/tools/objtool/arch/x86/decode.c b/tools/objtool/arch/x86/decode.c
index 0e8c8ec4fd4e..97a1fdb3db63 100644
--- a/tools/objtool/arch/x86/decode.c
+++ b/tools/objtool/arch/x86/decode.c
@@ -284,11 +284,16 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
 	case 0x8d:
 		if (sib == 0x24 && rex_w && !rex_b && !rex_x) {
 
-			/* lea disp(%rsp), reg */
 			*type = INSN_STACK;
-			op->src.type = OP_SRC_ADD;
+			if (!insn.displacement.value) {
+				/* lea (%rsp), reg */
+				op->src.type = OP_SRC_REG;
+			} else {
+				/* lea disp(%rsp), reg */
+				op->src.type = OP_SRC_ADD;
+				op->src.offset = insn.displacement.value;
+			}
 			op->src.reg = CFI_SP;
-			op->src.offset = insn.displacement.value;
 			op->dest.type = OP_DEST_REG;
 			op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
 
-- 
2.13.5