KVM: MMU: don't get free page number in the loop
authorXiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Fri, 4 Jun 2010 13:54:38 +0000 (21:54 +0800)
committerAvi Kivity <avi@redhat.com>
Sun, 1 Aug 2010 07:39:27 +0000 (10:39 +0300)
In the later patch, we will modify sp's zapping way like below:

kvm_mmu_prepare_zap_page A
kvm_mmu_prepare_zap_page B
kvm_mmu_prepare_zap_page C
....
kvm_mmu_commit_zap_page

[ zaped multiple sps only need to call kvm_mmu_commit_zap_page once ]

In __kvm_mmu_free_some_pages() function, the free page number is
getted form 'vcpu->kvm->arch.n_free_mmu_pages' in loop, it will
hinders us to apply kvm_mmu_prepare_zap_page() and kvm_mmu_commit_zap_page()
since kvm_mmu_prepare_zap_page() not free sp.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
arch/x86/kvm/mmu.c

index 9b849a70742da90ad6eb9417923be079a5ebf899..1aad8e713f7829efde8e7c643ad259b07afc0bdd 100644 (file)
@@ -2863,13 +2863,16 @@ EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt);
 
 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
 {
-       while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES &&
+       int free_pages;
+
+       free_pages = vcpu->kvm->arch.n_free_mmu_pages;
+       while (free_pages < KVM_REFILL_PAGES &&
               !list_empty(&vcpu->kvm->arch.active_mmu_pages)) {
                struct kvm_mmu_page *sp;
 
                sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev,
                                  struct kvm_mmu_page, link);
-               kvm_mmu_zap_page(vcpu->kvm, sp);
+               free_pages += kvm_mmu_zap_page(vcpu->kvm, sp);
                ++vcpu->kvm->stat.mmu_recycled;
        }
 }