Re: [Xen-devel] [PATCHv4 8/8] x86: reduce struct hvm_domain size

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 30.04.15 at 17:33, <david.vrabel@xxxxxxxxxx> wrote:
> Pack struct hvm_domain to reduce it by 8 bytes.  Thus reducing the
> size of struct domain by 8 bytes.

Is that really true _after_ the change to ticket locks?

> --- a/xen/include/asm-x86/hvm/domain.h
> +++ b/xen/include/asm-x86/hvm/domain.h
> @@ -116,12 +116,6 @@ struct hvm_domain {
>      /* VRAM dirty support.  Protect with the domain paging lock. */
>      struct sh_dirty_vram *dirty_vram;
>  
> -    /* If one of vcpus of this domain is in no_fill_mode or
> -     * mtrr/pat between vcpus is not the same, set is_in_uc_mode
> -     */
> -    spinlock_t             uc_lock;
> -    bool_t                 is_in_uc_mode;
> -
>      /* Pass-through */
>      struct hvm_iommu       hvm_iommu;

Here the block sits between two 8-byte aligned fields.

> @@ -137,6 +131,12 @@ struct hvm_domain {
>      bool_t                 is_s3_suspended;
>      bool_t                 introspection_enabled;
>  
> +    /* If one of vcpus of this domain is in no_fill_mode or
> +     * mtrr/pat between vcpus is not the same, set is_in_uc_mode
> +     */
> +    bool_t                 is_in_uc_mode;
> +    spinlock_t             uc_lock;
> +
>      /*
>       * TSC value that VCPUs use to calculate their tsc_offset value.
>       * Used during initialization and save/restore.

And here it follows 5 bool_t-s, and is being followed by an 8-byte
aligned field. I.e. without ticket locks it exactly fills the 3 byte gap,
but with ticket locks it requires a second 8-byte slot.

Additionally I wonder whether the reduced distance between
uc_lock and msixtbl_list_lock would now lead to (or, going forward,
at least risk) them being on the same cache line.

Jan


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