Re: [PATCH 2/3] x86/entry: Make #PF/NMI/INT0x82 more amenable to livepatching

["Roger Pau Monné" <roger.pau@citrix.com>]

On Tue, Jan 23, 2024 at 10:22:10AM +0100, Jan Beulich wrote:
> On 22.01.2024 19:17, Andrew Cooper wrote:
> > It is bad form to have inter-function fallthrough.  It only functions right
> > now because alignment padding bytes are NOPs.
> 
> But that's a requirement anyway in executable sections.

Really?  I was under the impression we wanted to replace the padding
nops with rets maybe, or even poison the padding with int3 or ud2.

> > --- a/xen/arch/x86/x86_64/compat/entry.S
> > +++ b/xen/arch/x86/x86_64/compat/entry.S
> > @@ -29,6 +29,7 @@ FUNC(entry_int82)
> >  
> >          mov   %rsp, %rdi
> >          call  do_entry_int82
> > +        jmp   compat_test_all_events
> >  END(entry_int82)
> >  
> >  /* %rbx: struct vcpu */
> > diff --git a/xen/arch/x86/x86_64/entry.S b/xen/arch/x86/x86_64/entry.S
> > index c3f6b667a72a..fc64ef1fd460 100644
> > --- a/xen/arch/x86/x86_64/entry.S
> > +++ b/xen/arch/x86/x86_64/entry.S
> > @@ -723,7 +723,9 @@ END(common_interrupt)
> >  FUNC(entry_PF)
> >          ENDBR64
> >          movl  $X86_EXC_PF, 4(%rsp)
> > +        jmp   handle_exception
> >  END(entry_PF)
> > +
> >  /* No special register assumptions. */
> >  FUNC(handle_exception, 0)
> >          ALTERNATIVE "", clac, X86_FEATURE_XEN_SMAP
> > @@ -1023,6 +1025,7 @@ FUNC(entry_NMI)
> >          ENDBR64
> >          pushq $0
> >          movl  $X86_EXC_NMI, 4(%rsp)
> > +        jmp   handle_ist_exception
> >  END(entry_NMI)
> >  
> >  FUNC(handle_ist_exception)
> 
> Hmm, so here you (partly) do what I was meaning to do in the one patch
> left from the entry point annotations series, "common: honor
> CONFIG_CC_SPLIT_SECTIONS also for assembly functions". However, I'm
> wrapping the JMPs there in #ifdef CONFIG_CC_SPLIT_SECTIONS. Thoughts?
> I view the JMPs as pretty useless otherwise, even if there is a
> small risk of a future code change not respecting the ordering
> requirements. Yet such would be noticed pretty quickly, I suppose.

I think it's clearer with the jumps.

Thanks, Roger.