]> git.baikalelectronics.ru Git - kernel.git/commitdiff
kvm: x86/mmu: Support MMIO in the TDP MMU
authorBen Gardon <bgardon@google.com>
Wed, 14 Oct 2020 18:26:58 +0000 (11:26 -0700)
committerPaolo Bonzini <pbonzini@redhat.com>
Fri, 23 Oct 2020 07:42:15 +0000 (03:42 -0400)
In order to support MMIO, KVM must be able to walk the TDP paging
structures to find mappings for a given GFN. Support this walk for
the TDP MMU.

Tested by running kvm-unit-tests and KVM selftests on an Intel Haswell
machine. This series introduced no new failures.

This series can be viewed in Gerrit at:
https://linux-review.googlesource.com/c/virt/kvm/kvm/+/2538

v2: Thanks to Dan Carpenter and kernel test robot for finding that root
was used uninitialized in get_mmio_spte.

Signed-off-by: Ben Gardon <bgardon@google.com>
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Message-Id: <20201014182700.2888246-19-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/mmu/mmu.c
arch/x86/kvm/mmu/tdp_mmu.c
arch/x86/kvm/mmu/tdp_mmu.h

index 4c62ac8db16998c767931c32742cfc525692eaec..6a0941ccac34ff150d5b59b38b3f9557793c7a23 100644 (file)
@@ -3479,54 +3479,82 @@ static bool mmio_info_in_cache(struct kvm_vcpu *vcpu, u64 addr, bool direct)
        return vcpu_match_mmio_gva(vcpu, addr);
 }
 
-/* return true if reserved bit is detected on spte. */
-static bool
-walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr, u64 *sptep)
+/*
+ * Return the level of the lowest level SPTE added to sptes.
+ * That SPTE may be non-present.
+ */
+static int get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes)
 {
        struct kvm_shadow_walk_iterator iterator;
-       u64 sptes[PT64_ROOT_MAX_LEVEL], spte = 0ull;
-       struct rsvd_bits_validate *rsvd_check;
-       int root, leaf;
-       bool reserved = false;
+       int leaf = vcpu->arch.mmu->root_level;
+       u64 spte;
 
-       rsvd_check = &vcpu->arch.mmu->shadow_zero_check;
 
        walk_shadow_page_lockless_begin(vcpu);
 
-       for (shadow_walk_init(&iterator, vcpu, addr),
-                leaf = root = iterator.level;
+       for (shadow_walk_init(&iterator, vcpu, addr);
             shadow_walk_okay(&iterator);
             __shadow_walk_next(&iterator, spte)) {
+               leaf = iterator.level;
                spte = mmu_spte_get_lockless(iterator.sptep);
 
                sptes[leaf - 1] = spte;
-               leaf--;
 
                if (!is_shadow_present_pte(spte))
                        break;
 
+       }
+
+       walk_shadow_page_lockless_end(vcpu);
+
+       return leaf;
+}
+
+/* return true if reserved bit is detected on spte. */
+static bool get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr, u64 *sptep)
+{
+       u64 sptes[PT64_ROOT_MAX_LEVEL];
+       struct rsvd_bits_validate *rsvd_check;
+       int root = vcpu->arch.mmu->root_level;
+       int leaf;
+       int level;
+       bool reserved = false;
+
+       if (!VALID_PAGE(vcpu->arch.mmu->root_hpa)) {
+               *sptep = 0ull;
+               return reserved;
+       }
+
+       if (is_tdp_mmu_root(vcpu->kvm, vcpu->arch.mmu->root_hpa))
+               leaf = kvm_tdp_mmu_get_walk(vcpu, addr, sptes);
+       else
+               leaf = get_walk(vcpu, addr, sptes);
+
+       rsvd_check = &vcpu->arch.mmu->shadow_zero_check;
+
+       for (level = root; level >= leaf; level--) {
+               if (!is_shadow_present_pte(sptes[level - 1]))
+                       break;
                /*
                 * Use a bitwise-OR instead of a logical-OR to aggregate the
                 * reserved bit and EPT's invalid memtype/XWR checks to avoid
                 * adding a Jcc in the loop.
                 */
-               reserved |= __is_bad_mt_xwr(rsvd_check, spte) |
-                           __is_rsvd_bits_set(rsvd_check, spte, iterator.level);
+               reserved |= __is_bad_mt_xwr(rsvd_check, sptes[level - 1]) |
+                           __is_rsvd_bits_set(rsvd_check, sptes[level - 1],
+                                              level);
        }
 
-       walk_shadow_page_lockless_end(vcpu);
-
        if (reserved) {
                pr_err("%s: detect reserved bits on spte, addr 0x%llx, dump hierarchy:\n",
                       __func__, addr);
-               while (root > leaf) {
+               for (level = root; level >= leaf; level--)
                        pr_err("------ spte 0x%llx level %d.\n",
-                              sptes[root - 1], root);
-                       root--;
-               }
+                              sptes[level - 1], level);
        }
 
-       *sptep = spte;
+       *sptep = sptes[leaf - 1];
+
        return reserved;
 }
 
@@ -3538,7 +3566,7 @@ static int handle_mmio_page_fault(struct kvm_vcpu *vcpu, u64 addr, bool direct)
        if (mmio_info_in_cache(vcpu, addr, direct))
                return RET_PF_EMULATE;
 
-       reserved = walk_shadow_page_get_mmio_spte(vcpu, addr, &spte);
+       reserved = get_mmio_spte(vcpu, addr, &spte);
        if (WARN_ON(reserved))
                return -EINVAL;
 
index 1491e2f7a89757c63a24593315133ac9046a3bc8..5158d02b8925edadfe5d9e8fc809110b45a31459 100644 (file)
@@ -7,7 +7,10 @@
 #include "tdp_mmu.h"
 #include "spte.h"
 
+#ifdef CONFIG_X86_64
 static bool __read_mostly tdp_mmu_enabled = false;
+module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644);
+#endif
 
 static bool is_tdp_mmu_enabled(void)
 {
@@ -1128,3 +1131,21 @@ bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
        return spte_set;
 }
 
+/*
+ * Return the level of the lowest level SPTE added to sptes.
+ * That SPTE may be non-present.
+ */
+int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes)
+{
+       struct tdp_iter iter;
+       struct kvm_mmu *mmu = vcpu->arch.mmu;
+       int leaf = vcpu->arch.mmu->shadow_root_level;
+       gfn_t gfn = addr >> PAGE_SHIFT;
+
+       tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
+               leaf = iter.level;
+               sptes[leaf - 1] = iter.old_spte;
+       }
+
+       return leaf;
+}
index 6501dd2ef8e4a2ad45ed849f6ca64936cf6b616b..556e065503f69b25e00a2e582f2e3418d11b16d5 100644 (file)
@@ -43,4 +43,6 @@ void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm,
 
 bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
                                   struct kvm_memory_slot *slot, gfn_t gfn);
+
+int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes);
 #endif /* __KVM_X86_MMU_TDP_MMU_H */