On Wed, Oct 04, 2017 at 02:30:42PM -0700, Linus Torvalds wrote:
> On Wed, Oct 4, 2017 at 2:06 PM, Josh Poimboeuf <jpoimboe@redhat.com> wrote:
> >
> > I compiled the same kernel with a similar version of GCC.  It turns out
> > that GCC *does* create unaligned stacks with frame pointers enabled:
> 
> Christ. What a piece of crap.
> 
> It doesn't even seem to make any sense. Spill room for the "u16
> item_count" and "u8 move_type"?

I didn't have the patience to try to figure out what it thought it was
trying to do.  But it doesn't even use %esp in the function, so it's
obviously just pointless code.  Pasting the full function below.

> That function is disgusting anyway (the switch really should be
> outside the loop, not inside it), but whatever. No excuse for that
> kind of garbage code generation.
> 
> > This was a leaf function.  For no apparent reason, GCC 4.8 decided to
> > subtract 3 from the stack pointer in the prologue.
> 
> Can you make objtool warn about unaligned stack pointer additions like that?

That should be easy to add, and a good idea to check for.  I don't look
forward to the continual 0-day bot complaints, so hopefully we can
figure out a way to fix them (or deprecate GCC 4!), or at least disable
the checking for older versions of GCC with the known issue.

> Maybe it only happens in very limited cases, and we can find a pattern
> to why gcc generates garbage code like that? And perhaps even how to
> just avoid it?

Just from grepping the objdump I can tell it happened in several
different functions.  For example all the following functions have 'sub
$0x1,%esp':

  match_wildcard()
  pci_vpd_find_tag()
  strspn()

sub $0x2,%esp:

  rgb_foreground()
  scsi_extd_sense_format()

sub $0x5,%esp:

  iot2040_rs485_config()

sub $0x6,%esp:

  x86_match_cpu()
  clear_buffer_attributes()

The only thing I can see they have in common is that they're leaf
functions.  In all cases the stack pointer is otherwise unused.

The problem seemed to go away when I built the same kernel with GCC 5.3.

FYI, here's the full function (4.14-rc3 compiled with RHEL GCC 4.8.3-9).
Config is attached.

c124a388 <acpi_rs_move_data>:
c124a388:       55                      push   %ebp
c124a389:       89 e5                   mov    %esp,%ebp
c124a38b:       57                      push   %edi
c124a38c:       56                      push   %esi
c124a38d:       89 d6                   mov    %edx,%esi
c124a38f:       53                      push   %ebx
c124a390:       31 db                   xor    %ebx,%ebx
c124a392:       83 ec 03                sub    $0x3,%esp
c124a395:       8a 55 08                mov    0x8(%ebp),%dl
c124a398:       66 89 4d f2             mov    %cx,-0xe(%ebp)
c124a39c:       83 ea 15                sub    $0x15,%edx
c124a39f:       88 55 f1                mov    %dl,-0xf(%ebp)
c124a3a2:       0f b6 fa                movzbl %dl,%edi
c124a3a5:       0f b7 4d f2             movzwl -0xe(%ebp),%ecx
c124a3a9:       39 cb                   cmp    %ecx,%ebx
c124a3ab:       73 36                   jae    c124a3e3 <acpi_rs_move_data+0x5b>
c124a3ad:       80 7d f1 07             cmpb   $0x7,-0xf(%ebp)
c124a3b1:       77 30                   ja     c124a3e3 <acpi_rs_move_data+0x5b>
c124a3b3:       ff 24 bd cc e9 36 c1    jmp    *-0x3ec91634(,%edi,4)
                        c124a3b6: R_386_32      .rodata
c124a3ba:       89 c7                   mov    %eax,%edi
c124a3bc:       f3 a4                   rep movsb %ds:(%esi),%es:(%edi)
c124a3be:       eb 23                   jmp    c124a3e3 <acpi_rs_move_data+0x5b>
c124a3c0:       66 8b 0c 5e             mov    (%esi,%ebx,2),%cx
c124a3c4:       66 89 0c 58             mov    %cx,(%eax,%ebx,2)
c124a3c8:       eb 16                   jmp    c124a3e0 <acpi_rs_move_data+0x58>
c124a3ca:       8b 0c 9e                mov    (%esi,%ebx,4),%ecx
c124a3cd:       89 0c 98                mov    %ecx,(%eax,%ebx,4)
c124a3d0:       eb 0e                   jmp    c124a3e0 <acpi_rs_move_data+0x58>
c124a3d2:       8b 14 de                mov    (%esi,%ebx,8),%edx
c124a3d5:       8b 4c de 04             mov    0x4(%esi,%ebx,8),%ecx
c124a3d9:       89 14 d8                mov    %edx,(%eax,%ebx,8)
c124a3dc:       89 4c d8 04             mov    %ecx,0x4(%eax,%ebx,8)
c124a3e0:       43                      inc    %ebx
c124a3e1:       eb c2                   jmp    c124a3a5 <acpi_rs_move_data+0x1d>
c124a3e3:       83 c4 03                add    $0x3,%esp
c124a3e6:       5b                      pop    %ebx
c124a3e7:       5e                      pop    %esi
c124a3e8:       5f                      pop    %edi
c124a3e9:       5d                      pop    %ebp
c124a3ea:       c3                      ret