[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] GP exception on vmxon

  • To: Hu Jia Yi <jyhu@xxxxxxxxx>, <xen-devel@xxxxxxxxxxxxxxxxxxx>
  • From: Keir Fraser <Keir.Fraser@xxxxxxxxxxxx>
  • Date: Fri, 11 Jan 2008 10:47:02 +0000
  • Delivery-date: Fri, 11 Jan 2008 02:47:37 -0800
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>
  • Thread-index: AchUHaTnhFD6M1JJT7O+aK2k0WyB7QAEA+vYAADRAMAAA5TvoQ==
  • Thread-topic: [Xen-devel] GP exception on vmxon

If you run in MS-DOS context you’re probably in real mode, and hence VMXON will #GP?

 -- Keir

On 11/1/08 09:25, "Hu Jia Yi" <jyhu@xxxxxxxxx> wrote:

In linux environment (FC7), the definition of vmxon and its vmxon region is the same as in xen.
In MS-DOS, I have to use 16 bit vmxon operand op-code, as follows
;compiled using NASM
%define VMXON_OPCODE  db 0xf3, 0xf, 0xc7        ;same as in xen
%define MODRM_BX_SI_06 db 0x30                       ;the physical address is in [bx+si]  ref: Intel SDM 2A Page2-6
;word   vmxon(dword* p_vmcs)
            push     bp
            mov      bp, sp
mov      bx, [bp+4]        
mov      si, 0
;return value is in ax
ja         .1
mov      ax, 0     ;fail
jmp      .2
            mov      ax, 1     ;success
            pop      bp
in C file
int start_vmx()
            int phy_vmcs[4] = {0,0,0,0}; /* to form 64bit operand and phy_vmcs[0] is at the lowest address*/
            phy_vmcs[0] =  g_vmcs_phy_addr;      /*g_vmcs_phy_addr is the physical address calculated in real mode because the base is not zero*/
            return vmxon(phy_vmcs);

Best regards,
Hu Jia Yi
Ext: 20430
Tel: 65-67510430
-----Original Message-----
From: Keir Fraser [mailto:Keir.Fraser@xxxxxxxxxxxx]
Sent: Friday, January 11, 2008 4:41 PM
To: Hu Jia Yi; xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-devel] GP exception on vmxon

How early do you run? Are you really confident that you disabled A20M? A simple test would confirm this (check whether address 2MB aliases with 1MB).

 -- Keir

On 11/1/08 06:46, "Hu Jia Yi" <jyhu@xxxxxxxxx> wrote:
Hello, I tried to write a piece of code to start vmx.
This code is directly interacting with cpu instead of with virtual cpu as in xen.
But every time I call vmxon, a GP exception happens.
Could anybody help me on this? The following is the context
1.     After booting up to the program, I disable A20M.
2.     allocate a 4kb-aligned vmxon region and calculate its physical address.
3.     setup identity page table and enter protected page mode. In this step I also set x86_cr0_ne ( cr0.bit5)
4.     call start_vmx. This start_vmx function is similar to the one in xen3.1.0
1.     test cpuid with eax = 1. ecx.vmxe(bit5) is 1.
2.     Test IA32_FEATURE_CONTROL_MSR, result is 0x05, so bit 0 and bit 2 are both 1.
3.     Set cr4.vmxe (bit13) to 1
4.     Call vmx_init_vmcs_config(). This function is the same as in xen3.1.0.
5.     Call vmxon, passing it the physical adderss calculated in step2, using the same op-code as xen
f.    stop vmx by calling vmxoff.
Using “while(1)”, I traced and found the GP exception happened in step 4.e.
>From Intel Software Development Manual 2B, I get the following conditions to throw a GP.
IF (CPL > 0) or (in A20M mode) or
(the values of CR0 and CR4 are supported in VMX operation) or
(bit 0 (lock bit) of IA32_FEATURE_CONTROL MSR is clear) or
(bit 2 of IA32_FEATURE_CONTROL MSR is clear)
THEN #GP(0);
I checked the conditions and found none of them was violated.
The results are as follows
CR0 :                                       0x80000031
IA32_VMX_CR0_FIXED0:    0x80000021
CR4 :                                       0x2250
IA32_VMX_CR4_FIXED0:    0x2000
IA32_VMX_CR4_FIXED1:    0x27FF
IA32_VMX_BASIC_MSR is 001A 0400 0000 0007
The revision ID 0x07 is assigned to the corresponding field in vmxon region in the step 4.d
My PC has a 32 bit, VT-support multi-core CPU.
I use only the BSP and haven’t dealt with multi-cpu wake-up.
Best regards,
Hu Jia Yi
Ext: 20430
Tel: 65-67510430

Xen-devel mailing list

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.