Re: Remove __end_entry_SYSENTER_compat?

From: Borislav Petkov
Date: Wed Jul 12 2017 - 05:25:30 EST

On Wed, Jul 12, 2017 at 10:48:03AM +0200, Ingo Molnar wrote:
> Hm, I'd argue that the old code is much clearer: we need both the start and the
> end of a function and have the properly named symbols for that.
> That entry_SYSCALL_compat() happens to start just where
> __end_entry_SYSENTER_compat is an accident of placement.

Yeah, probably not worth the effort of actually making it less reliable
this way...

> Is it even true - doesn't ENTRY() imply an .align, in which case it might be that
> __end_entry_SYSENTER_compat != entry_SYSCALL_compat?

Yes, ENTRY does .p2align:

.globl __end_entry_SYSENTER_compat; __end_entry_SYSENTER_compat:
.type entry_SYSENTER_compat, @function ; .size entry_SYSENTER_compat, .-entry_SYSENTER_compat
# 184 "arch/x86/entry/entry_64_compat.S"
.globl entry_SYSCALL_compat ; .p2align 4, 0x90 ; entry_SYSCALL_compat:
Pads to an alignment of 4 with NOPs.

> In fact that appears to be the case for my defconfig:
> ffffffff81942f90 T entry_SYSENTER_compat
> ffffffff81942feb T __end_entry_SYSENTER_compat
> ffffffff81942ff0 T entry_SYSCALL_compat
> So unless there's some disadvantage beyond having one more symbol, I'd favor the
> old approach.

Yeah, my only concern was getting rid of the global symbol but it
doesn't work in this case.



ECO tip #101: Trim your mails when you reply.