Re: [PATCH V6 1/2] x86/xsaves: calculate the comp_offsets base on xcomp_bv

[Shuai Ruan <shuai.ruan@linux.intel.com>]

On Tue, Mar 29, 2016 at 08:35:34AM -0600, Jan Beulich wrote:
> >>> On 24.03.16 at 09:29, <shuai.ruan@linux.intel.com> wrote:
> >  static void *get_xsave_addr(struct xsave_struct *xsave,
> > -        unsigned int xfeature_idx)
> > +                            const uint16_t *comp_offsets,
> > +                            unsigned int xfeature_idx)
> >  {
> >      if ( !((1ul << xfeature_idx) & xsave->xsave_hdr.xstate_bv) )
> >          return NULL;
> 
> While not a problem in this patch, I think this being at least
> questionable now becomes more obvious: With XSAVEC and
> XSAVES this field may have clear bits just because the
> component was in init state. Yet I don't think this should be
> fixed here - instead the consumer(s) should better deal with
> NULL coming back here (i.e. expand_xsave_states() should
> probably memset() the area in an "else" to the if() leading to
> memcpy(), whereas compress_xsave_states() is fine as is).
> 
> That's a 3rd patch, though.
> 
I will add code only in expand_xsave_states() to deal with NULL 
returned by get_xsave_addr().
> > +    return (void *)xsave + (xsave_area_compressed(xsave) ?
> > +                            comp_offsets[xfeature_idx] :
> > +                            xstate_offsets[xfeature_idx]);
> 
> This function doesn't ever get called when
> !xsave_area_compressed(xsave), so I don't really see why this
> conditional expression ever got added. Yet another follow-up
> patch I would say.
I will drop this.
> 
> The patch here looks okay to me now.
> 
Thanks
> Jan
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel