On Thu, Mar 27, 2025 at 10:54:23AM +0100, Jan Beulich wrote:
> mtrr_set_all() has quite a bit of overhead, which is entirely useless
> when set_mtrr_state() really does nothing. Furthermore, with
> mtrr_state.def_type never initialized from hardware, post_set()'s
> unconditional writing of the MSR means would leave us running in UC
> mode after the sync.
>
> Signed-off-by: Jan Beulich <jbeul...@suse.com>
>
> --- a/xen/arch/x86/cpu/mtrr/main.c
> +++ b/xen/arch/x86/cpu/mtrr/main.c
> @@ -605,13 +605,15 @@ void mtrr_aps_sync_begin(void)
>
> void mtrr_aps_sync_end(void)
> {
> - set_mtrr(~0U, 0, 0, 0);
> + if (mtrr_if)
> + set_mtrr(~0U, 0, 0, 0);
> hold_mtrr_updates_on_aps = 0;
> }
>
> void mtrr_bp_restore(void)
Maybe I'm blind, but I cannot find any caller to mtrr_bp_restore()?
Am I missing something obvious?
Thanks, Roger.
