Re: [PATCH kvm-unit-tests 1/2] VMX: checks for validity of vmxon region

[Jan Kiszka <jan.kiszka@siemens.com>]

On 2014-06-04 23:17, Bandan Das wrote:
> Verify that vmon fails with unaligned vmxon region or
> any bits set beyong the physical address width. Also verify
> failure with an invalid revision identifier.
> 
> Signed-off-by: Bandan Das <bsd@redhat.com>
> ---
>  x86/vmx.c | 46 +++++++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 43 insertions(+), 3 deletions(-)
> 
> diff --git a/x86/vmx.c b/x86/vmx.c
> index 1182eef..207eb81 100644
> --- a/x86/vmx.c
> +++ b/x86/vmx.c
> @@ -37,7 +37,7 @@
>  #include "smp.h"
>  #include "io.h"
>  
> -u32 *vmxon_region;
> +u64 *vmxon_region;
>  struct vmcs *vmcs_root;
>  u32 vpid_cnt;
>  void *guest_stack, *guest_syscall_stack;
> @@ -598,13 +598,53 @@ static int test_vmx_feature_control(void)
>  
>  static int test_vmxon(void)
>  {
> -	int ret;
> +	int ret, ret1;
>  	u64 rflags;
> +	u64 *tmp_region = vmxon_region;
> +	int width = cpuid(0x80000008).a & 0xff;
> +
> +	/* Unaligned page access */
> +	vmxon_region = (u64 *)((intptr_t)vmxon_region + 1);
> +	rflags = read_rflags() | X86_EFLAGS_CF | X86_EFLAGS_ZF;
> +	write_rflags(rflags);
> +	ret1 = vmx_on();

Can we ensure that the compiler doesn't inject any ops between
write_rflags and the vmxon that overwrite CF or ZF? If you want those
flags in a specific state, maybe it's better to pass that to vmx_on and
do this in the assembly block with vmxon.

> +	report("test vmxon with unaligned vmxon region", ret1);
> +	if (!ret1) {
> +		ret = 1;
> +		goto out;
> +	}
>  
> +	/* gpa bits beyond physical address width are set*/
> +	vmxon_region = (u64 *)((intptr_t)tmp_region | ((u64)1 << (width+1)));
> +	rflags = read_rflags() | X86_EFLAGS_CF | X86_EFLAGS_ZF;
> +	write_rflags(rflags);
> +	ret1 = vmx_on();
> +	report("test vmxon with bits set beyond physical address width", ret1);
> +	if (!ret1) {
> +		ret = 1;
> +		goto out;
> +	}
> +
> +	/* invalid revision indentifier */
> +	vmxon_region = tmp_region;
> +	*vmxon_region = 0xba9da9;
> +	rflags = read_rflags() | X86_EFLAGS_CF | X86_EFLAGS_ZF;
> +	write_rflags(rflags);
> +	ret1 = vmx_on();
> +	report("test vmxon with invalid revision identifier", ret1);
> +	if (!ret1) {
> +		ret = 1;
> +		goto out;
> +	}
> +
> +	/* and finally a valid region */
> +	*vmxon_region = basic.revision;
>  	rflags = read_rflags() | X86_EFLAGS_CF | X86_EFLAGS_ZF;
>  	write_rflags(rflags);
>  	ret = vmx_on();

Oh, this pattern is not by you! OK, but let's address this first, then
add the new tests on top.

Jan

> -	report("test vmxon", !ret);
> +	report("test vmxon with valid vmxon region", !ret);
> +
> +out:
>  	return ret;
>  }
>  
> 

-- 
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux