Re: [PATCH v11 13/23] x86: refactor psr: CDP: implement CPU init flow.

[Yi Sun <yi.y.sun@linux.intel.com>]

On 17-06-07 02:14:26, Jan Beulich wrote:
> >>> On 07.06.17 at 03:31, <yi.y.sun@linux.intel.com> wrote:
> > On 17-06-06 02:38:27, Jan Beulich wrote:
> >> >>> On 06.06.17 at 10:13, <yi.y.sun@linux.intel.com> wrote:
> >> > On 17-06-06 01:45:11, Jan Beulich wrote:
> >> >> >>> On 02.06.17 at 09:26, <yi.y.sun@linux.intel.com> wrote:
> >> >> > On 17-05-31 03:37:48, Jan Beulich wrote:
> >> >> >> >>> On 03.05.17 at 10:44, <yi.y.sun@linux.intel.com> wrote:
> >> >> >> > +        /* Cut half of cos_max when CDP is enabled. */
> >> >> >> > +        feat->cos_max >>= 1;
> >> >> >> > +
> >> >> >> > +        /* We only write mask1 since mask0 is always all ones by default. */
> >> >> >> 
> >> >> >> Is this, btw, just reset state or even guaranteed after offlining
> >> >> >> and re-onlining a CPU?
> >> >> >> 
> >> >> > Below MSRs are all per socket. So, we just need reset them when socket is
> >> >> > online.
> >> >> 
> >> >> Which I hope you've understood then means the comment and
> >> >> presumably also the code here need further refinement.
> >> >> 
> >> > Spec states it below. So, the mask0 is guranteed.
> >> > "the default mask in IA32_L3_MASK_0 - which is all “1”s (on reset)"
> >> 
> >> Sigh. I did ask very clearly (and this is still visible above) about
> >> the case where the CPU did _not_ undergo a reset cycle.
> >> 
> > Sorry, I think I must mis-understand your question. SW never modifies value
> > of COS ID 0 mask register. So, the value in mask0 will not be changed after
> > socket first online (mask0 is set to default by HW this time). So, we think
> > it is not necessary to restore mask0 value to default here.
> 
> Is the register read-only? If so, writing it would indeed be
> unnecessary (and likely wrong, as it might raise #GP then
> instead of being ignored). If not, firmware may have written it,
> so you'll want to write it even during boot time CPU bringup.
> 
No, it is writable. I will restore mask0 to default value here.
Furthermore, for L3/L2 CAT mask0, we may need to restore mask0
too.

> Jan

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