KVM: x86/mmu: Allocate PAE root array when using SVM's 32-bit NPT

SVM's Nested Page Tables (NPT) reuses x86 paging for the host-controlled
page walk.  For 32-bit KVM, this means PAE paging is used even when TDP
is enabled, i.e. the PAE root array needs to be allocated.

Fixes: ee6268ba3a68 ("KVM: x86: Skip pae_root shadow allocation if tdp enabled")
Cc: stable@vger.kernel.org
Reported-by: Jiri Palecek <jpalecek@web.de>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 1e9ba81accba526b80bb698b193b405c07d48c02..d3c3d5e5ffd4999d3c1f1249167bbd8586fefb6f 100644 (file)
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -5602,14 +5602,18 @@ static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
        struct page *page;
        int i;
 
-       if (tdp_enabled)
-               return 0;
-
        /*
-        * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
-        * Therefore we need to allocate shadow page tables in the first
-        * 4GB of memory, which happens to fit the DMA32 zone.
+        * When using PAE paging, the four PDPTEs are treated as 'root' pages,
+        * while the PDP table is a per-vCPU construct that's allocated at MMU
+        * creation.  When emulating 32-bit mode, cr3 is only 32 bits even on
+        * x86_64.  Therefore we need to allocate the PDP table in the first
+        * 4GB of memory, which happens to fit the DMA32 zone.  Except for
+        * SVM's 32-bit NPT support, TDP paging doesn't use PAE paging and can
+        * skip allocating the PDP table.
         */
+       if (tdp_enabled && kvm_x86_ops->get_tdp_level(vcpu) > PT32E_ROOT_LEVEL)
+               return 0;
+
        page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_DMA32);
        if (!page)
                return -ENOMEM;