Re: [PATCH v2 2/5] KVM: MMU: adjust page size early if gfn used as page table

[Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>]

On 12/12/2012 08:57 AM, Marcelo Tosatti wrote:
> On Mon, Dec 10, 2012 at 05:13:03PM +0800, Xiao Guangrong wrote:
>> We have two issues in current code:
>> - if target gfn is used as its page table, guest will refault then kvm will use
>>   small page size to map it. We need two #PF to fix its shadow page table
>>
>> - sometimes, say a exception is triggered during vm-exit caused by #PF
>>   (see handle_exception() in vmx.c), we remove all the shadow pages shadowed
>>   by the target gfn before go into page fault path, it will cause infinite
>>   loop:
>>   delete shadow pages shadowed by the gfn -> try to use large page size to map
>>   the gfn -> retry the access ->...
>>
>> To fix these, We can adjust page size early if the target gfn is used as page
>> table
>>
>> Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
>> ---
>>  arch/x86/kvm/mmu.c         |   13 ++++---------
>>  arch/x86/kvm/paging_tmpl.h |   33 ++++++++++++++++++++++++++++++++-
>>  2 files changed, 36 insertions(+), 10 deletions(-)
>>
>> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
>> index 2a3c890..54fc61e 100644
>> --- a/arch/x86/kvm/mmu.c
>> +++ b/arch/x86/kvm/mmu.c
>> @@ -2380,15 +2380,10 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
>>  	if (pte_access & ACC_WRITE_MASK) {
>>
>>  		/*
>> -		 * There are two cases:
>> -		 * - the one is other vcpu creates new sp in the window
>> -		 *   between mapping_level() and acquiring mmu-lock.
>> -		 * - the another case is the new sp is created by itself
>> -		 *   (page-fault path) when guest uses the target gfn as
>> -		 *   its page table.
>> -		 * Both of these cases can be fixed by allowing guest to
>> -		 * retry the access, it will refault, then we can establish
>> -		 * the mapping by using small page.
>> +		 * Other vcpu creates new sp in the window between
>> +		 * mapping_level() and acquiring mmu-lock. We can
>> +		 * allow guest to retry the access, the mapping can
>> +		 * be fixed if guest refault.
>>  		 */
>>  		if (level > PT_PAGE_TABLE_LEVEL &&
>>  		    has_wrprotected_page(vcpu->kvm, gfn, level))
>> diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
>> index ec481e9..32d77ff 100644
>> --- a/arch/x86/kvm/paging_tmpl.h
>> +++ b/arch/x86/kvm/paging_tmpl.h
>> @@ -491,6 +491,36 @@ out_gpte_changed:
>>  	return 0;
>>  }
>>
>> + /*
>> + * To see whether the mapped gfn can write its page table in the current
>> + * mapping.
>> + *
>> + * It is the helper function of FNAME(page_fault). When guest uses large page
>> + * size to map the writable gfn which is used as current page table, we should
>> + * force kvm to use small page size to map it because new shadow page will be
>> + * created when kvm establishes shadow page table that stop kvm using large
>> + * page size. Do it early can avoid unnecessary #PF and emulation.
>> + *
>> + * Note: the PDPT page table is not checked for PAE-32 bit guest. It is ok
>> + * since the PDPT is always shadowed, that means, we can not use large page
>> + * size to map the gfn which is used as PDPT.
>> + */
>> +static bool
>> +FNAME(mapped_gfn_can_write_current_pagetable)(struct guest_walker *walker)
>> +{
>> +	int level;
>> +	gfn_t mask = ~(KVM_PAGES_PER_HPAGE(walker->level) - 1);
>> +
>> +	if (!(walker->pte_access & ACC_WRITE_MASK))
>> +		return false;
>> +
>> +	for (level = walker->level; level <= walker->max_level; level++)
>> +		if (!((walker->gfn ^ walker->table_gfn[level - 1]) & mask))
>> +			return true;
> 
> XOR won't work. Just check with sums and integer comparison, ie.
> walker->gfn + KVM_PAGES_PER_HPAGE(walker->level).

It can not work since walker->gfn is not large-page-size aligned. For example,
guest uses large page size to map 0x123000000 to physical address 0-2M, if
guest faults on 0x123001000, walker->gfn = 0x1000.

The code "if (!((walker->gfn ^ walker->table_gfn[level - 1]) & mask))" is the
same as "if (walker->gfn & mask == walker->table_gfn[level - 1] & mask)" - if
any page in the large page area used as page table, we should use 4K page size
to fix it.

In above example, if table_gfn is in the area [0, 2M), kvm is forced to use
4k page size.

> 
> Moreover, its confusing to have it checked at this level. What about
> doing at reexecute_instruction?

Hmm, this patch is trying to fix a bug described in the changelog:

======
 - sometimes, say a exception is triggered during vm-exit caused by #PF
   (see handle_exception() in vmx.c), we remove all the shadow pages shadowed
   by the target gfn before go into page fault path, it will cause infinite
   loop:
   delete shadow pages shadowed by the gfn -> try to use large page size to map
   the gfn -> retry the access ->...
======

Which is caused by this code:

	if (is_page_fault(intr_info)) {
		/* EPT won't cause page fault directly */
		BUG_ON(enable_ept);
		cr2 = vmcs_readl(EXIT_QUALIFICATION);
		trace_kvm_page_fault(cr2, error_code);

		if (kvm_event_needs_reinjection(vcpu))
			kvm_mmu_unprotect_page_virt(vcpu, cr2);
		return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
	}

This bug is introduced in commit c219346325.

Another way to fix it is doing this change:
@@ -2395,7 +2395,7 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
                 */
                if (level > PT_PAGE_TABLE_LEVEL &&
                    has_wrprotected_page(vcpu->kvm, gfn, level))
-                       goto done;
+                       return 1;

The disadvantage of this way is, it causes unnecessary emulation. For example,
if 0-2M is mapped in guest and only page 0 used as page table, any write to
[4K, 2M) will need be emulated.

Your idea?