[RFC] respect the referenced bit of KVM guest pages?

Commit Message

Avi Kivity Aug. 5, 2009, 8:17 a.m. UTC

On 08/05/2009 10:58 AM, Avi Kivity wrote:
> On 08/05/2009 05:40 AM, Wu Fengguang wrote:
>> Greetings,
>>
>> Jeff Dike found that many KVM pages are being refaulted in 2.6.29:
>>
>> "Lots of pages between discarded due to memory pressure only to be
>> faulted back in soon after. These pages are nearly all stack pages.
>> This is not consistent - sometimes there are relatively few such pages
>> and they are spread out between processes."
>>
>> The refaults can be drastically reduced by the following patch, which
>> respects the referenced bit of all anonymous pages (including the KVM
>> pages).
>>
>> However it risks reintroducing the problem addressed by commit 7e9cd4842
>> (fix reclaim scalability problem by ignoring the referenced bit,
>> mainly the pte young bit). I wonder if there are better solutions?
>
> How do you distinguish between kvm pages and non-kvm anonymous pages?  
> More importantly, why should you?
>
> Jeff, do you see the refaults on Nehalem systems?  If so, that's 
> likely due to the lack of an accessed bit on EPT pagetables.  It would 
> be interesting to compare with Barcelona  (which does).
>
> If that's indeed the case, we can have the EPT ageing mechanism give 
> pages a bit more time around by using an available bit in the EPT PTEs 
> to return accessed on the first pass and not-accessed on the second.
>

The attached patch implements this.

Comments

Rik van Riel Aug. 5, 2009, 2:33 p.m. UTC | #1

Avi Kivity wrote:

> The attached patch implements this.

The attached page requires each page to go around twice
before it is evicted, but they will still get evicted in
the order in which they were made present.

FIFO page replacement was shown to be a bad idea in the
1960's and it is still a terrible idea today.

Avi Kivity Aug. 5, 2009, 3:37 p.m. UTC | #2

On 08/05/2009 05:33 PM, Rik van Riel wrote:
> Avi Kivity wrote:
>
>> The attached patch implements this.
>
> The attached page requires each page to go around twice
> before it is evicted, but they will still get evicted in
> the order in which they were made present.
>
> FIFO page replacement was shown to be a bad idea in the
> 1960's and it is still a terrible idea today.
>

Which is why we have accessed bits in page tables... but emulating the 
accessed bit via RWX (note no present bit in EPT) is better than 
ignoring it.

Patch

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 7b53614..310938a 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -195,6 +195,7 @@  static u64 __read_mostly shadow_x_mask;	/* mutual exclusive with nx_mask */
 static u64 __read_mostly shadow_user_mask;
 static u64 __read_mostly shadow_accessed_mask;
 static u64 __read_mostly shadow_dirty_mask;
+static int __read_mostly shadow_accessed_shift;
 
 static inline u64 rsvd_bits(int s, int e)
 {
@@ -219,6 +220,8 @@  void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
 {
 	shadow_user_mask = user_mask;
 	shadow_accessed_mask = accessed_mask;
+	shadow_accessed_shift
+		= find_first_bit((void *)&shadow_accessed_mask, 64);
 	shadow_dirty_mask = dirty_mask;
 	shadow_nx_mask = nx_mask;
 	shadow_x_mask = x_mask;
@@ -817,11 +820,11 @@  static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp)
 	while (spte) {
 		int _young;
 		u64 _spte = *spte;
-		BUG_ON(!(_spte & PT_PRESENT_MASK));
-		_young = _spte & PT_ACCESSED_MASK;
+		BUG_ON(!(_spte & shadow_accessed_mask));
+		_young = _spte & shadow_accessed_mask;
 		if (_young) {
 			young = 1;
-			clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
+			clear_bit(shadow_accessed_shift, (unsigned long *)spte);
 		}
 		spte = rmap_next(kvm, rmapp, spte);
 	}
@@ -2572,7 +2575,7 @@  static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn)
 	    && shadow_accessed_mask
 	    && !(*spte & shadow_accessed_mask)
 	    && is_shadow_present_pte(*spte))
-		set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
+		set_bit(shadow_accessed_shift, (unsigned long *)spte);
 }
 
 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 0ba706e..bc99367 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -4029,7 +4029,7 @@  static int __init vmx_init(void)
 		bypass_guest_pf = 0;
 		kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
 			VMX_EPT_WRITABLE_MASK);
-		kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
+		kvm_mmu_set_mask_ptes(0ull, 1ull << 63, 0ull, 0ull,
 				VMX_EPT_EXECUTABLE_MASK);
 		kvm_enable_tdp();
 	} else