Re: [PATCH] [RFC] x86: avoid -mtune=atom for objtool warnings

From: Josh Poimboeuf
Date: Mon Oct 10 2016 - 16:30:08 EST


On Mon, Oct 10, 2016 at 02:56:56PM +0200, Arnd Bergmann wrote:
> I have no idea what is actually going on here, but building an x86 kernel
> with CONFIG_MATOM results in countless warnings from objtool, such as
>
> arch/x86/events/intel/ds.o: warning: objtool: intel_pmu_pebs_del()+0x43: call without frame pointer save/setup
> security/keys/keyring.o: warning: objtool: keyring_read()+0x59: call without frame pointer save/setup
> kernel/signal.o: warning: objtool: __dequeue_signal()+0xd8: call without frame pointer save/setup
> kernel/signal.o: warning: objtool: kill_pid()+0x15: call without frame pointer save/setup
> kernel/signal.o: warning: objtool: SyS_signal()+0x27: call without frame pointer save/setup
> mm/page_alloc.o: warning: objtool: zone_watermark_ok_safe()+0x27: call without frame pointer save/setup
> fs/exec.o: warning: objtool: read_code()+0x18: call without frame pointer save/setup
> mm/swap.o: warning: objtool: get_kernel_page()+0x24: call without frame pointer save/setup
> mm/swap.o: warning: objtool: pagevec_move_tail.constprop.25()+0x26: call without frame pointer save/setup
> block/bio.o: warning: objtool: bio_map_kern()+0x47: call without frame pointer save/setup
> arch/x86/crypto/poly1305_glue.o: warning: objtool: poly1305_simd_mult()+0x2d: call without frame pointer save/setup
> crypto/skcipher.o: warning: objtool: skcipher_encrypt_ablkcipher()+0x58: call without frame pointer save/setup
> crypto/skcipher.o: warning: objtool: skcipher_decrypt_ablkcipher()+0x58: call without frame pointer save/setup
> fs/inode.o: warning: objtool: ilookup()+0x5d: call without frame pointer save/setup
> fs/inode.o: warning: objtool: proc_nr_inodes()+0x3e: call without frame pointer save/setup
> fs/namei.o: warning: objtool: lookup_one_len_unlocked()+0x21: call without frame pointer save/setup
> block/elevator.o: warning: objtool: elv_rb_add()+0x5b: call without frame pointer save/setup
> crypto/shash.o: warning: objtool: shash_async_init()+0x1e: call without frame pointer save/setup
> crypto/shash.o: warning: objtool: shash_async_import()+0x1e: call without frame pointer save/setup
> mm/vmscan.o: warning: objtool: pfmemalloc_watermark_ok()+0xb9: call without frame pointer save/setup
>
> I have not looked at whether this is a bug in gcc or in objtool, however
> I found that not using -mtune=atom reliably avoids the problem. I could
> reproduce the problem with gcc versions 4.7 through 6.1.

Thanks for reporting it. It looks like 'mtune=atom' sometimes makes a
slight change to one of the stack frame setup instructions. Instead of:

move rsp, rbp

It sometimes does:

lea (%rsp),%rbp

They're two different instructions, but they have the same result. It's
an easy fix for objtool. I'll post a patch soon.

--
Josh