Re: [PATCH v21 06/26] x86/cet: Add control-protection fault handler

[Yu, Yu-cheng]

On 2/24/2021 11:42 AM, Borislav Petkov wrote:
> On Wed, Feb 24, 2021 at 11:30:34AM -0800, Andy Lutomirski wrote:
> > On Wed, Feb 24, 2021 at 11:20 AM Borislav Petkov <bp@xxxxxxxxx> wrote:
> > > On Wed, Feb 24, 2021 at 09:56:13AM -0800, Yu, Yu-cheng wrote:
> > > > No.  Maybe I am doing too much.  The GP fault sets si_addr to zero, for
> > > > example.  So maybe do the same here?
> > >
> > > No, you're looking at this from the wrong angle. This is going to be
> > > user-visible and the moment it gets upstream, it is cast in stone.
> > >
> > > So the whole use case of what luserspace needs to do or is going to do
> > > or wants to do on a SEGV_CPERR, needs to be described, agreed upon by
> > > people etc before it goes out. And thus clarified whether the address
> > > gets copied out or not.
> >
> > I vote 0.  The address is in ucontext->gregs[REG_RIP] [0] regardless.
> > Why do we need to stick a copy somewhere else?
> >
> > [0] or however it's spelled.  i can never remember.
>
> Fine with me. Let's have this documented in the manpage and then we can
> move forward with this.
>
> Thx.

The man page at https://man7.org/linux/man-pages/man2/sigaction.2.html says,

SIGILL, SIGFPE, SIGSEGV, SIGBUS, and SIGTRAP fill in si_addr with the 
address of the fault.

But it is not entirely true.

I will send a patch to update it, and another patch for the si_code.

--
Yu-cheng