Re: [PATCH v9 15/29] x86/insn-eval: Add utility functions to get segment descriptor base address and limit

[Ricardo Neri <ricardo.neri-calderon@linux.intel.com>]

On Wed, 2017-10-11 at 22:16 +0200, Borislav Petkov wrote:
> On Wed, Oct 11, 2017 at 12:57:01PM -0700, Ricardo Neri wrote:
> > 
> > This is meant to be an error case. In long mode,
> > only INAT_SEG_REG_IGNORE/FS/GS
> > are valid. All other indices are invalid.
> > 
> > Perhaps we could return -EINVAL instead?
> So, my question is, when are you ever going to have that case? What
> constellation of events would ever hit this else branch for long mode?
> Because it looks impossible to me. What I can imagine only is something
> like this:
> 
>                 else if (seg_reg != INAT_SEG_REG_IGNORE)
> 			WARN_ONCE(1, "This should never happen!\n");
> 
> assertion.

To clarify, I think you mean seg_reg_idx.

Yes, it would be impossible to hit this else branch provided that callers don't
attempt to use an invalid seg_reg_idx while in long mode. Probably this is not
critical as this is a static function and as such we control who can call it and
make sure seg_reg_idx is always valid (i.e., INAT_SEG_REG_IGNORE/FS/GS in long
mode).



> But you don't really need that - you can simply ignore seg_reg in that
> case:
> 
>         if (user_64bit_mode(regs)) {
>                 /*
>                  * Only FS or GS will have a base address, the rest of
>                  * the segments' bases are forced to 0.
>                  */
>                 unsigned long base;
> 
>                 if (seg_reg == INAT_SEG_REG_FS)
>                         rdmsrl(MSR_FS_BASE, base);
>                 else if (seg_reg == INAT_SEG_REG_GS)
>                         /*
>                          * swapgs was called at the kernel entry point. Thus,
>                          * MSR_KERNEL_GS_BASE will have the user-space GS
> base.
>                          */
>                         rdmsrl(MSR_KERNEL_GS_BASE, base);
>                 else
>                         base = 0;
> 
>                 return base;
>         }
> 
> Or am I missing something?

My intention is to let the caller know about the invalid seg_reg_idx instead of
silently correcting the caller's input by ignoring seg_reg_idx.

On the other hand, in long mode, hardware ignore all segment registers except FS
and GS.

Hence, I guess I can remove the check in question.

Thanks and BR,
Ricardo