On 10.08.2017 21:38, Bandan Das wrote:
> David Hildenbrand <david@xxxxxxxxxx> writes:
>
>> Don't use shifts, tag them correctly as EPTP and use better matching
>> names (PWL vs. GAW).
>
> Fair enough. This reminds me, I also don't particularly like the vmx->nested.nested...
> accesses that are spread all throughout :)
Me too, I got a bunch of patches lying around that abstract VMX features
into something like X86 features.
#define VMX_FEATURE_BASIC_INOUT (0*32 + 18)
#define VMX_FEATURE_PREEMPTION_TIMER (1*32 + 6)
#define VMX_FEATURE_POSTED_INTR (1*32 + 7)
#define VMX_FEATURE_TPR_SHADOW (2*32 + 21)
#define VMX_FEATURE_MSR_BITMAP (2*32 + 28)
#define VMX_FEATURE_SECONDARY_EXEC_CTRLS (2*32 + 31)
#define VMX_FEATURE_VIRTUALIZE_APIC_ACCESSES (3*32 + 0)
...
#define VMX_FEATURE_EXIT_LOAD_IA32_PAT (4*32 + 19)
...
#define VMX_FEATURE_ENTRY_LOAD_IA32_PAT (5*32 + 14)
...
#define VMX_FEATURE_EPT_EXECUTE_ONLY (6*32 + 0)
...
#define VMX_FEATURE_INVVPID (7*32 + 0)
...
/* whether our CPU supports a certain VMX feature */
bool vmx_has(unsigned int feature);
-> replaces most cpu_has_vmx_*
/* whether our nested CPU is allowed to use a certain VMX feature */
bool nested_vmx_has(struct kvm_vcpu *vcpu, unsigned int feature);
-> replaces most checks against
vmx->nested.nested_vmx_secondary_ctls_high and friends
/* whether L1 enabled a certain VMX feature for L2 */
bool nested_cpu_has(struct vmcs12 *vmcs12, unsigned int feature);
-> merges/replaces most checks like nested_cpu_has/nested_cpu_has2 ...
I am not yet happy with the end result. Anybody interested or has any
suggestions?
--
Thanks,
David
