[PATCH] KVM: x86/mmu: Recurse down to 1GB level when zapping pages in a range

[Paolo Bonzini]

The recursive zapping that was reintroduced by reverting "KVM: x86/mmu:
Zap only TDP MMU leafs in kvm_zap_gfn_range()" can be expensive.  Allow
zap_gfn_range to recurse down to the PDPTE level, so that periodic
yielding is possible with a finer granularity.

Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
---
 arch/x86/kvm/mmu/tdp_mmu.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 87d8910c9ac2..53689603078a 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -926,8 +926,10 @@ static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
 	/*
 	 * No need to try to step down in the iterator when zapping all SPTEs,
 	 * zapping the top-level non-leaf SPTEs will recurse on their children.
+	 * Do not do it above the 1GB level, to avoid making tdp_mmu_set_spte's
+	 * recursion too expensive and allow yielding.
 	 */
-	int min_level = zap_all ? root->role.level : PG_LEVEL_4K;
+	int min_level = zap_all ? PG_LEVEL_1G : PG_LEVEL_4K;
 
 	end = min(end, tdp_mmu_max_gfn_host());
 
-- 
2.31.1