Re: [RFC PATCH v2 20/21] x86: Add support for CONFIG_CFI_CLANG

[Peter Zijlstra]

On Tue, May 17, 2022 at 08:48:41AM +0000, David Laight wrote:
> From: Peter Zijlstra
> > Sent: 17 May 2022 09:41
> ...
> > > If we use %eax instead of %r10d for the hash transfer (as per Joao), and
> > > use int3 instead of ud2, then we can shrink the fineibt sequence to:
> > >
> > > __cfi_\func:
> > > 	endbr				# 4
> > > 	xorl	$0x12345678, %eax	# 5
> > > 	jz	1f			# 2
> > > 	int3				# 1
> > > \func:
> > > 	...
> > >
> > > Which is 12 bytes, and needs a larger preamble (up from 9 in the current
> > > proposal).
> > 
> > On all that; perhaps it would be good to have a compiler option to
> > specify the preamble size. It can enforce the minimum at 7 to have at
> > least the required:
> > 
> > 	movl $0x12345678, %eax
> > 	int3
> > 	int3
> > 
> > but any larger number will just increase the preamble with int3 padding
> > at the top.
> > 
> > That can go right along with the option to supress endbr when preamble
> > :-)
> 
> You also need a compiler option to specify the register.
> While (I think) %eax is usable in kernel, it isn't in userspace.
> It is used in varargs calls to pass (IIRC) the number of fp
> args that are passed in registers.

You're mistaken, the compiler doesn't emit the FineIBT code *at*all*.
That's all patched in later. For kCFI the mov is never executed and is
only there to make it a valid instruction.