RE: [Xen-devel] [PATCH][VT] Patch to allow VMX domainsto be destroyedor shut down cleanly

[Khoa Huynh <khoa@xxxxxxxxxx>]

Jiang, Yunhong wrote:
>> I think the Xin Xiaohui's patch resolved most problem for the page
>> left issue.
>>
>> Also another small issue found is the on vmx_set_cr0. On Windows, the
>> guest will enter/leave protected mode many times.  Following code
>> cause problem.
>>     if ((value & X86_CR0_PE) && (value & X86_CR0_PG) &&
>> !paging_enabled) {
>>         /*
>>          * The guest CR3 must be pointing to the guest physical.
>>          */
>>         if ( !VALID_MFN(mfn = get_mfn_from_pfn(
>>             d->arch.arch_vmx.cpu_cr3 >> PAGE_SHIFT)) ||
>>              !get_page(pfn_to_page(mfn), d->domain) )
>>           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>         {
>>             printk("Invalid CR3 value = %lx",
>>             d->arch.arch_vmx.cpu_cr3); domain_crash_synchronous(); /*
>>         need to take a clean path */ }
>>
>> We should place the get_page to when guest set cr3 when not paging,
>> i.e
>>
>>    case 3:
>>     {
>>         unsigned long old_base_mfn, mfn;
>>
>>         /*
>>          * If paging is not enabled yet, simply copy the value to CR3.
>>          */
>>         if (!vmx_paging_enabled(d)) {
>>                       ..... get page here....
>>
>
>More exactly should be:
>            if (!vmx_paging_enabled(d)) {
>                                    get_page for new mfn
>                                    put_page for old mfn
>            }

Xin's patch works for some of my Windows XP scenarios,
but there are still cases where there were still pages remaining.
I tried to implement a second patch based on your suggestion above,
but that did not work.

Do you happen to have a complete patch to fix this ?  Thanks.

Khoa H.



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel