[PATCH 1/2] kvm,ept: pass error code to handler

[PATCH 1/2] kvm,ept: pass error code to handler

[Lai Jiangshan]

handle_ept_violation() does not pass error code to
the handler tdp_page_fault().

It means tdp_page_fault() handles the page fault with ignoring
the error code, It will not handle the page fault completely correctly,
and may causes endless page fault.

{
But because there is another bad code in the kvm, This endless page faulting
never occur. The bad code: kvm always set pte writable.

negative * negative = positive, WTF!!

Next patch will fix this bad code.
}

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
---
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index c7565ea..6362dcc 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -3470,7 +3470,7 @@ static int handle_ept_violation(struct kvm_vcpu *vcpu)
 
 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
 	trace_kvm_page_fault(gpa, exit_qualification);
-	return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
+	return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, exit_qualification);
 }
 
 static u64 ept_rsvd_mask(u64 spte, int level)

Re: [PATCH 1/2] kvm,ept: pass error code to handler

[Marcelo Tosatti]

On Fri, Jun 11, 2010 at 07:30:24PM +0800, Lai Jiangshan wrote:
> handle_ept_violation() does not pass error code to
> the handler tdp_page_fault().
> 
> It means tdp_page_fault() handles the page fault with ignoring
> the error code, It will not handle the page fault completely correctly,
> and may causes endless page fault.
> 
> {
> But because there is another bad code in the kvm, This endless page faulting
> never occur. The bad code: kvm always set pte writable.
> 
> negative * negative = positive, WTF!!
> 
> Next patch will fix this bad code.
> }
> 
> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index c7565ea..6362dcc 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -3470,7 +3470,7 @@ static int handle_ept_violation(struct kvm_vcpu *vcpu)
>  
>  	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
>  	trace_kvm_page_fault(gpa, exit_qualification);
> -	return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
> +	return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, exit_qualification);

exit_qualification & 0x2, because bit 1 (write fault) of exit
qualification is the only one that matches x86's pagefault error codes.