Re: [PATCH 4/8] arm64: Basic Branch Target Identification support

[Dave Martin]

On Thu, Jun 06, 2019 at 10:34:22AM -0700, Yu-cheng Yu wrote:
> On Thu, 2019-06-06 at 18:23 +0100, Dave Martin wrote:
> > On Thu, Jun 06, 2019 at 06:11:56PM +0100, Catalin Marinas wrote:
> > > On Fri, May 24, 2019 at 03:53:06PM +0100, Dave P Martin wrote:
> > > > On Fri, May 24, 2019 at 02:02:17PM +0100, Mark Rutland wrote:
> > > > > On Fri, May 24, 2019 at 11:25:29AM +0100, Dave Martin wrote:
> > > > > >  #endif /* _UAPI__ASM_HWCAP_H */
> > > > > > diff --git a/arch/arm64/include/uapi/asm/mman.h
> > > > > > b/arch/arm64/include/uapi/asm/mman.h
> > > > > > new file mode 100644
> > > > > > index 0000000..4776b43
> > > > > > --- /dev/null
> > > > > > +++ b/arch/arm64/include/uapi/asm/mman.h
> > > > > > @@ -0,0 +1,9 @@
> > > > > > +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> > > > > > +#ifndef _UAPI__ASM_MMAN_H
> > > > > > +#define _UAPI__ASM_MMAN_H
> > > > > > +
> > > > > > +#include <asm-generic/mman.h>
> > > > > > +
> > > > > > +#define PROT_BTI_GUARDED	0x10		/* BTI guarded
> > > > > > page */
> > > > > 
> > > > > From prior discussions, I thought this would be PROT_BTI, without the
> > > > > _GUARDED suffix. Do we really need that?
> > > > > 
> > > > > AFAICT, all other PROT_* definitions only have a single underscore, and
> > > > > the existing arch-specific flags are PROT_ADI on sparc, and PROT_SAO on
> > > > > powerpc.
> > > > 
> > > > No strong opinon.  I was trying to make the name less obscure, but I'm
> > > > equally happy with PROT_BTI if people prefer that.
> > > 
> > > I prefer PROT_BTI as well. We are going to add a PROT_MTE at some point
> > > (and a VM_ARM64_MTE in the high VMA flag bits).
> > 
> > Ack.
> > 
> > Some things need attention, so I need to respin this series anyway.
> > 
> > skip_faulting_instruction() and kprobes/uprobes may need looking at,
> > plus I want to simply the ELF parsing (at least to skip some cost for
> > arm64).
> 
> Can we add a case in the 'consistency checks for the interpreter' (right above
> where you add arch_parse_property()) for PT_NOTE?  That way you can still use
> part of the same parser.

I think for arm64 that we can skip searching all the notes by checking
for a PT_GNU_PROPERTY entry; once that's found, the actual
NT_GNU_PROPERTY_TYPE_0 parsing should be common.  If there's no
PT_GNU_PROPERTY entry, we can immediately give up.

For x86, would it makes sense to use PT_GNU_PROPERTY if it's there,
and fall back to scanning all the notes otherwise?  Ideally we
wouldn't need the fallback, but if there are binaries in the wild with
NT_GNU_PROPERTY_TYPE_0 that lack a PT_GNU_PROPERTY entry, we may be
stuck with that.

Thoughts?

Cheers
---Dave