On Mon, Sep 8, 2014 at 8:07 PM, Andrew Cooper <andrew.cooper3@citrix.com> wrote:

On 08/09/14 13:02, Zhuo Song wrote:
> * Since there would not be 32-bit hypervisor, we do not need
>   hypervisor_is_64bit() again.
>
> * Remove xen_64bit from xc_cpuid_pv_policy().
>
> * Because is_64bit only depends on is_pae, only use is_pae for both
>   vendor specific functions.
>
> * Move conditions for LM/NX masking into architectural logic
>
> Signed-off-by: Zhuo Song <songzhuo.sz@alibaba-inc.com>
> ---
>  tools/libxc/xc_cpuid_x86.c | 37 ++++++++++++++-----------------------
>  1 file changed, 14 insertions(+), 23 deletions(-)
>
> diff --git a/tools/libxc/xc_cpuid_x86.c b/tools/libxc/xc_cpuid_x86.c
> index 6b81641..710fd61 100644
> --- a/tools/libxc/xc_cpuid_x86.c
> +++ b/tools/libxc/xc_cpuid_x86.c
> @@ -34,13 +34,6 @@
>  #define DEF_MAX_INTELEXT  0x80000008u
>  #define DEF_MAX_AMDEXT    0x8000001cu
>
> -static int hypervisor_is_64bit(xc_interface *xch)
> -{
> -    xen_capabilities_info_t xen_caps = "";
> -    return ((xc_version(xch, XENVER_capabilities, &xen_caps) == 0) &&
> -            (strstr(xen_caps, "x86_64") != NULL));
> -}
> -
>  static void cpuid(const unsigned int *input, unsigned int *regs)
>  {
>      unsigned int count = (input[1] == XEN_CPUID_INPUT_UNUSED) ? 0 : input[1];
> @@ -95,13 +88,11 @@ static void amd_xc_cpuid_policy(
>          break;
>
>      case 0x80000001: {
> -        int is_64bit = hypervisor_is_64bit(xch) && is_pae;
> -
>          if ( !is_pae )
>              clear_bit(X86_FEATURE_PAE, regs[3]);
>
>          /* Filter all other features according to a whitelist. */
> -        regs[2] &= ((is_64bit ? bitmaskof(X86_FEATURE_LAHF_LM) : 0) |
> +        regs[2] &= (bitmaskof(X86_FEATURE_LAHF_LM) |
>                      bitmaskof(X86_FEATURE_CMP_LEGACY) |
>                      (is_nestedhvm ? bitmaskof(X86_FEATURE_SVM) : 0) |
>                      bitmaskof(X86_FEATURE_CR8_LEGACY) |
> @@ -116,8 +107,8 @@ static void amd_xc_cpuid_policy(
>                      bitmaskof(X86_FEATURE_TBM) |
>                      bitmaskof(X86_FEATURE_DBEXT));
>          regs[3] &= (0x0183f3ff | /* features shared with 0x00000001:EDX */
> -                    (is_pae ? bitmaskof(X86_FEATURE_NX) : 0) |
> -                    (is_64bit ? bitmaskof(X86_FEATURE_LM) : 0) |
> +                    bitmaskof(X86_FEATURE_NX) |
> +                    bitmaskof(X86_FEATURE_LM) |

You are changing the behaviour here, due to dropping is_pae.

This will break VM migrate.

~Andrew

I move it to architectural logic as I said. See:

In my opinion, for LM:

> +            clear_bit(X86_FEATURE_LAHF_LM, regs[2]);
> +            clear_bit(X86_FEATURE_LM, regs[3]);

for NX:

clear_bit(X86_FEATURE_NX, regs[3]); 

should have done the work, so we do not need to do it again both in amd_xc_cpuid_policy or intel_xc_cpuid_policy

Zhuo