Re: [PATCH v3 9/9] x86/xsaves: Re-enable XSAVES

[Yu-cheng Yu <yu-cheng.yu@intel.com>]

On Tue, Mar 01, 2016 at 03:56:12PM -0800, Dave Hansen wrote:
> On 02/29/2016 09:42 AM, Yu-cheng Yu wrote:
> >  	/*
> > -	 * Quirk: we don't yet handle the XSAVES* instructions
> > -	 * correctly, as we don't correctly convert between
> > -	 * standard and compacted format when interfacing
> > -	 * with user-space - so disable it for now.
> > -	 *
> > -	 * The difference is small: with recent CPUs the
> > -	 * compacted format is only marginally smaller than
> > -	 * the standard FPU state format.
> > -	 *
> > -	 * ( This is easy to backport while we are fixing
> > -	 *   XSAVES* support. )
> > +	 * Most recent CPUs supporting XSAVES can run 64-bit mode.
> > +	 * Enable XSAVES for 64-bit.
> >  	 */
> > -	setup_clear_cpu_cap(X86_FEATURE_XSAVES);
> > +	if (!config_enabled(CONFIG_X86_64))
> > +		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
> >  }
> 
> I think we need a much better explanation of this for posterity.  Why
> are we not supporting this now, and what would someone have to do in the
> future in order to enable it?
> 
If anyone is using this newer feature, then that user is most likely using
a 64-bit capable processor and a 64-bit kernel. The intention here is to
take out the complexity and any potential of error. If the user removes 
the restriction and builds a private kernel, it should work but we have
not checked all possible combinations. I will put these in the comments.

> >  /*
> > diff --git a/arch/x86/kernel/fpu/xstate.c b/arch/x86/kernel/fpu/xstate.c
> > index 2e80d6f..cb2a484 100644
> > --- a/arch/x86/kernel/fpu/xstate.c
> > +++ b/arch/x86/kernel/fpu/xstate.c
> > @@ -204,6 +204,14 @@ void fpu__init_cpu_xstate(void)
> >  	if (!cpu_has_xsave || !xfeatures_mask)
> >  		return;
> >  
> > +	/*
> > +	 * Make it clear that XSAVES supervisor states are not yet
> > +	 * implemented should anyone expect it to work by changing
> > +	 * bits in XFEATURE_MASK_* macros and XCR0.
> > +	 */
> > +	WARN_ONCE((xfeatures_mask & XFEATURE_MASK_SUPERVISOR),
> > +		"x86/fpu: XSAVES supervisor states are not yet implemented.\n");
> > +
> >  	cr4_set_bits(X86_CR4_OSXSAVE);
> >  	xsetbv(XCR_XFEATURE_ENABLED_MASK, xfeatures_mask);
> >  }
> 
> Let's also do a:
> 
> 	xfeatures_mask &= ~XFEATURE_MASK_SUPERVISOR;
> 
> Otherwise, we have a broken system at the moment.
> 
Currently, if anyone sets any supervisor state in xfeatures_mask, the
kernel prints out the warning then goes into a protection fault.
That is a very strong indication to the user. Do we want to mute it? 

Yu-cheng