On 13.03.2024 13:24, George Dunlap wrote:
> In order to make implementation and testing tractable, we will require
> specific host functionality. Add a nested_virt bit to hvm_funcs.caps,
> and return an error if a domain is created with nested virt and this
> bit isn't set. Create VMX and SVM callbacks to be executed from
> start_nested_svm(), which is guaranteed to execute after all
Nit: nestedhvm_setup() (or, with different wording, start_nested_{svm,vmx}()).
> command-line options have been procesed.
>
> For VMX, start with always enabling it if HAP is present; this
> shouldn't change current behvior.
>
> For SVM, require some basic functionality, adding a document
> explaining the rationale.
>
> NB that only SVM CPUID bits 0-7 have been considered. Bits 10-16 may
> be considered in a follow-up patch.
>
> Signed-off-by: George Dunlap <george.dun...@cloud.com>
>[...]
> --- a/xen/arch/x86/domain.c
> +++ b/xen/arch/x86/domain.c
> @@ -673,6 +673,12 @@ int arch_sanitise_domain_config(struct
> xen_domctl_createdomain *config)
> */
> config->flags |= XEN_DOMCTL_CDF_oos_off;
>
> + if ( nested_virt && !hvm_nested_virt_supported() )
> + {
> + dprintk(XENLOG_INFO, "Nested virt requested but not available\n");
> + return -EINVAL;
> + }
> +
> if ( nested_virt && !hap )
> {
> dprintk(XENLOG_INFO, "Nested virt not supported without HAP\n");
As mentioned in reply to v1, I think what both start_nested_{svm,vmx}() check
is redundant with this latter check. I think that check isn't necessary there
(yet unlike suggested in reply to v1 I don't think anymore that the check here
can alternatively be dropped). And even if it was to be kept for some reason,
I'm having some difficulty seeing why vendor code needs to do that check, when
nestedhvm_setup() could do it for both of them.
> --- a/xen/arch/x86/hvm/nestedhvm.c
> +++ b/xen/arch/x86/hvm/nestedhvm.c
> @@ -150,6 +150,16 @@ static int __init cf_check nestedhvm_setup(void)
> __clear_bit(0x80, shadow_io_bitmap[0]);
> __clear_bit(0xed, shadow_io_bitmap[1]);
>
> + /*
> + * NB this must be called after all command-line processing has been
> + * done, so that if (for example) HAP is disabled, nested virt is
> + * disabled as well.
> + */
> + if ( cpu_has_vmx )
> + start_nested_vmx(&hvm_funcs);
> + else if ( cpu_has_svm )
> + start_nested_svm(&hvm_funcs);
Instead of passing the pointer, couldn't you have the functions return
bool, and then set hvm_funcs.caps.nested_virt from that? Passing that
pointer looks somewhat odd to me.
Is there a reason to use direct calls here rather than a true hook
(seeing that hvm_funcs must have been populated by the time we make it
here)? I understand we're (remotely) considering to switch away from
using hooks at some point, but right now this feels somewhat
inconsistent if not further justified.
> --- a/xen/arch/x86/hvm/vmx/vvmx.c
> +++ b/xen/arch/x86/hvm/vmx/vvmx.c
> @@ -2816,6 +2816,14 @@ void nvmx_set_cr_read_shadow(struct vcpu *v, unsigned
> int cr)
> __vmwrite(read_shadow_field, v->arch.hvm.nvcpu.guest_cr[cr]);
> }
>
> +void __init start_nested_vmx(struct hvm_function_table *hvm_function_table)
> +{
> + /* TODO: Require hardware support before enabling nested virt */
> + hvm_function_table->caps.nested_virt =
> + hvm_function_table->caps.hap;
> +}
> +
> +
Nit: No double blank lines please.
Jan
