Re: [PATCH 3/5] efi/x86: Make efi32_pe_entry more readable

[Ard Biesheuvel]

On Mon, 2 Mar 2020 at 18:02, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
>
> On Mon, Mar 02, 2020 at 05:57:04PM +0100, Ard Biesheuvel wrote:
> > On Mon, 2 Mar 2020 at 17:54, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
> > >
> > > On Mon, Mar 02, 2020 at 08:49:17AM +0100, Ard Biesheuvel wrote:
> > > > On Mon, 2 Mar 2020 at 00:04, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
> > > ...
> > > > >         call    1f
> > > > > -1:     pop     %ebp
> > > > > -       subl    $1b, %ebp
> > > > > +1:     pop     %ebx
> > > > > +       subl    $1b, %ebx
> > > ...
> > > > >
> > > > > +       movl    %ebx, %ebp                      // startup_32 for efi32_pe_stub_entry
> > > >
> > > > The code that follows efi32_pe_stub_entry still expects the runtime
> > > > displacement in %ebp, so we'll need to pass that in another way here.
> > > >
> > > > >         jmp     efi32_pe_stub_entry
> > >
> > > Didn't follow -- what do you mean by runtime displacement?
> > >
> > > efi32_pe_stub_entry expects the runtime address of startup_32 to be in
> > > %ebp, but with the changes for keeping the frame pointer in %ebp, I
> > > changed the runtime address to be in %ebx instead. Hence I added that
> > > movl %ebx, %ebp to put it in %ebp just before calling efi32_pe_stub_entry.
> > > That should be fine, no?
> >
> > But how does that work with:
> >
> > SYM_INNER_LABEL(efi32_pe_stub_entry, SYM_L_LOCAL)
> >     movl %ecx, efi32_boot_args(%ebp)
> >     movl %edx, efi32_boot_args+4(%ebp)
> >     movb $0, efi_is64(%ebp)
> >
> >
> > ?
>
> Why wouldn't it work? Before this change, efi32_pe_entry set %ebp to
> startup_32 (via the call/pop/sub sequence), so efi32_pe_stub_entry was
> entered with %ebp == startup_32.
>
> After this change, the call/pop/sub sequence puts startup_32 into %ebx,
> and then I copy it into %ebp just before branching to efi32_pe_stub_entry.
> So everything should continue to work the same way as before?

Ah ok, so the sequence

   call 1f
1: pop %ebp
   subl $1b, %ebp

happens to produce the value of startup_32, which is obvious now that
I think of it, but it wasn't before.