[PATCH v8 00/14] KVM TDX: TDP MMU: large page support

[PATCH v8 00/14] KVM TDX: TDP MMU: large page support

[isaku.yamahata]

From: Isaku Yamahata <isaku.yamahata@intel.com>

This patch series is based on "v19 KVM TDX: basic feature support".  It
implements large page support for TDP MMU by allowing populating of the large
page and splitting it when necessary.

No major changes from v7 instead of rebasing.

Thanks,

Changes from v7:
- Rebased to v19 TDX KVM v6.8-rc5 based patch series

Changes from v6:
- Rebased to v18 TDX KVM v6.8-rc1 based patch series
- use struct tdx_module_args
- minor improve on comment, commit message

Changes from v5:
- Switched to TDX module 1.5 base.

Chnages from v4:
- Rebased to v16 TDX KVM v6.6-rc2 base

Changes from v3:
- Rebased to v15 TDX KVM v6.5-rc1 base

Changes from v2:
- implemented page merging path
- rebased to TDX KVM v11

Changes from v1:
- implemented page merging path
- rebased to UPM v10
- rebased to TDX KVM v10
- rebased to kvm.git queue + v6.1-rc8

Isaku Yamahata (4):
  KVM: x86/tdp_mmu: Allocate private page table for large page split
  KVM: x86/tdp_mmu: Try to merge pages into a large page
  KVM: TDX: Implement merge pages into a large page
  KVM: x86/mmu: Make kvm fault handler aware of large page of private
    memslot

Sean Christopherson (1):
  KVM: Add transparent hugepage support for dedicated guest memory

Xiaoyao Li (9):
  KVM: TDX: Flush cache based on page size before TDX SEAMCALL
  KVM: TDX: Pass KVM page level to tdh_mem_page_aug()
  KVM: TDX: Pass size to reclaim_page()
  KVM: TDX: Update tdx_sept_{set,drop}_private_spte() to support large
    page
  KVM: MMU: Introduce level info in PFERR code
  KVM: TDX: Pass desired page level in err code for page fault handler
  KVM: x86/tdp_mmu: Split the large page when zap leaf
  KVM: x86/tdp_mmu, TDX: Split a large page when 4KB page within it
    converted to shared
  KVM: TDX: Allow 2MB large page for TD GUEST

 Documentation/virt/kvm/api.rst     |   7 +
 arch/x86/include/asm/kvm-x86-ops.h |   3 +
 arch/x86/include/asm/kvm_host.h    |  11 ++
 arch/x86/kvm/mmu/mmu.c             |  38 ++--
 arch/x86/kvm/mmu/mmu_internal.h    |  30 +++-
 arch/x86/kvm/mmu/tdp_iter.c        |  37 +++-
 arch/x86/kvm/mmu/tdp_iter.h        |   2 +
 arch/x86/kvm/mmu/tdp_mmu.c         | 276 ++++++++++++++++++++++++++---
 arch/x86/kvm/vmx/common.h          |   6 +-
 arch/x86/kvm/vmx/tdx.c             | 221 +++++++++++++++++------
 arch/x86/kvm/vmx/tdx_arch.h        |  21 +++
 arch/x86/kvm/vmx/tdx_errno.h       |   3 +
 arch/x86/kvm/vmx/tdx_ops.h         |  56 ++++--
 arch/x86/kvm/vmx/vmx.c             |   2 +-
 include/uapi/linux/kvm.h           |   2 +
 virt/kvm/guest_memfd.c             |  73 +++++++-
 16 files changed, 672 insertions(+), 116 deletions(-)

-- 
2.25.1

[PATCH v8 01/14] KVM: Add transparent hugepage support for dedicated guest memory

[isaku.yamahata]

From: Sean Christopherson <seanjc@google.com>

Extended guest_memfd to allow backing guest memory with transparent
hugepages.  Require userspace to opt-in via a flag even though there's no
known/anticipated use case for forcing small pages as THP is optional,
i.e. to avoid ending up in a situation where userspace is unaware that
KVM can't provide hugepages.

For simplicity, require the guest_memfd size to be a multiple of the
hugepage size, e.g. so that KVM doesn't need to do bounds checking when
deciding whether or not to allocate a huge folio.

When reporting the max order when KVM gets a pfn from guest_memfd, force
order-0 pages if the hugepage is not fully contained by the memslot
binding, e.g. if userspace requested hugepages but punches a hole in the
memslot bindings in order to emulate x86's VGA hole.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Link: https://lore.kernel.org/r/20231027182217.3615211-18-seanjc@google.com
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 Documentation/virt/kvm/api.rst |  7 ++++
 include/uapi/linux/kvm.h       |  2 +
 virt/kvm/guest_memfd.c         | 73 ++++++++++++++++++++++++++++++----
 3 files changed, 75 insertions(+), 7 deletions(-)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index 4b70d2b43532..213738a38b07 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -6312,6 +6312,8 @@ and cannot be resized  (guest_memfd files do however support PUNCH_HOLE).
 	__u64 reserved[6];
   };
 
+  #define KVM_GUEST_MEMFD_ALLOW_HUGEPAGE         (1ULL << 0)
+
 Conceptually, the inode backing a guest_memfd file represents physical memory,
 i.e. is coupled to the virtual machine as a thing, not to a "struct kvm".  The
 file itself, which is bound to a "struct kvm", is that instance's view of the
@@ -6328,6 +6330,11 @@ most one mapping per page, i.e. binding multiple memory regions to a single
 guest_memfd range is not allowed (any number of memory regions can be bound to
 a single guest_memfd file, but the bound ranges must not overlap).
 
+If KVM_GUEST_MEMFD_ALLOW_HUGEPAGE is set in flags, KVM will attempt to allocate
+and map hugepages for the guest_memfd file.  This is currently best effort.  If
+KVM_GUEST_MEMFD_ALLOW_HUGEPAGE is set, the size must be aligned to the maximum
+transparent hugepage size supported by the kernel
+
 See KVM_SET_USER_MEMORY_REGION2 for additional details.
 
 4.143 KVM_MEMORY_MAPPING
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index a7aa804ef021..47faaf71799f 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -2317,6 +2317,8 @@ struct kvm_create_guest_memfd {
 	__u64 reserved[6];
 };
 
+#define KVM_GUEST_MEMFD_ALLOW_HUGEPAGE		(1ULL << 0)
+
 #define KVM_MEMORY_MAPPING	_IOWR(KVMIO, 0xd5, struct kvm_memory_mapping)
 
 struct kvm_memory_mapping {
diff --git a/virt/kvm/guest_memfd.c b/virt/kvm/guest_memfd.c
index 3830d50b9b67..236443c3d8dc 100644
--- a/virt/kvm/guest_memfd.c
+++ b/virt/kvm/guest_memfd.c
@@ -13,14 +13,47 @@ struct kvm_gmem {
 	struct list_head entry;
 };
 
-static struct folio *kvm_gmem_get_folio(struct inode *inode, pgoff_t index)
+static struct folio *kvm_gmem_get_huge_folio(struct inode *inode, pgoff_t index)
 {
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	unsigned long huge_index = round_down(index, HPAGE_PMD_NR);
+	unsigned long flags = (unsigned long)inode->i_private;
+	struct address_space *mapping  = inode->i_mapping;
+	gfp_t gfp = mapping_gfp_mask(mapping);
 	struct folio *folio;
 
-	/* TODO: Support huge pages. */
-	folio = filemap_grab_folio(inode->i_mapping, index);
-	if (IS_ERR_OR_NULL(folio))
+	if (!(flags & KVM_GUEST_MEMFD_ALLOW_HUGEPAGE))
+		return NULL;
+
+	if (filemap_range_has_page(mapping, huge_index << PAGE_SHIFT,
+				   (huge_index + HPAGE_PMD_NR - 1) << PAGE_SHIFT))
+		return NULL;
+
+	folio = filemap_alloc_folio(gfp, HPAGE_PMD_ORDER);
+	if (!folio)
+		return NULL;
+
+	if (filemap_add_folio(mapping, folio, huge_index, gfp)) {
+		folio_put(folio);
 		return NULL;
+	}
+
+	return folio;
+#else
+	return NULL;
+#endif
+}
+
+static struct folio *kvm_gmem_get_folio(struct inode *inode, pgoff_t index)
+{
+	struct folio *folio;
+
+	folio = kvm_gmem_get_huge_folio(inode, index);
+	if (!folio) {
+		folio = filemap_grab_folio(inode->i_mapping, index);
+		if (IS_ERR_OR_NULL(folio))
+			return NULL;
+	}
 
 	/*
 	 * Use the up-to-date flag to track whether or not the memory has been
@@ -363,6 +396,7 @@ static int __kvm_gmem_create(struct kvm *kvm, loff_t size, u64 flags)
 	inode->i_mode |= S_IFREG;
 	inode->i_size = size;
 	mapping_set_gfp_mask(inode->i_mapping, GFP_HIGHUSER);
+	mapping_set_large_folios(inode->i_mapping);
 	mapping_set_unmovable(inode->i_mapping);
 	/* Unmovable mappings are supposed to be marked unevictable as well. */
 	WARN_ON_ONCE(!mapping_unevictable(inode->i_mapping));
@@ -388,12 +422,21 @@ int kvm_gmem_create(struct kvm *kvm, struct kvm_create_guest_memfd *args)
 	u64 flags = args->flags;
 	u64 valid_flags = 0;
 
+	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
+		valid_flags |= KVM_GUEST_MEMFD_ALLOW_HUGEPAGE;
+
 	if (flags & ~valid_flags)
 		return -EINVAL;
 
 	if (size <= 0 || !PAGE_ALIGNED(size))
 		return -EINVAL;
 
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if ((flags & KVM_GUEST_MEMFD_ALLOW_HUGEPAGE) &&
+	    !IS_ALIGNED(size, HPAGE_PMD_SIZE))
+		return -EINVAL;
+#endif
+
 	return __kvm_gmem_create(kvm, size, flags);
 }
 
@@ -488,7 +531,7 @@ void kvm_gmem_unbind(struct kvm_memory_slot *slot)
 int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
 		     gfn_t gfn, kvm_pfn_t *pfn, int *max_order)
 {
-	pgoff_t index = gfn - slot->base_gfn + slot->gmem.pgoff;
+	pgoff_t index, huge_index;
 	struct kvm_gmem *gmem;
 	struct folio *folio;
 	struct page *page;
@@ -501,6 +544,7 @@ int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
 
 	gmem = file->private_data;
 
+	index = gfn - slot->base_gfn + slot->gmem.pgoff;
 	if (WARN_ON_ONCE(xa_load(&gmem->bindings, index) != slot)) {
 		r = -EIO;
 		goto out_fput;
@@ -520,9 +564,24 @@ int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
 	page = folio_file_page(folio, index);
 
 	*pfn = page_to_pfn(page);
-	if (max_order)
-		*max_order = 0;
+	if (!max_order)
+		goto success;
+
+	*max_order = compound_order(compound_head(page));
+	if (!*max_order)
+		goto success;
 
+	/*
+	 * The folio can be mapped with a hugepage if and only if the folio is
+	 * fully contained by the range the memslot is bound to.  Note, the
+	 * caller is responsible for handling gfn alignment, this only deals
+	 * with the file binding.
+	 */
+	huge_index = ALIGN(index, 1ull << *max_order);
+	if (huge_index < ALIGN(slot->gmem.pgoff, 1ull << *max_order) ||
+	    huge_index + (1ull << *max_order) > slot->gmem.pgoff + slot->npages)
+		*max_order = 0;
+success:
 	r = 0;
 
 out_unlock:
-- 
2.25.1

[PATCH v8 02/14] KVM: TDX: Flush cache based on page size before TDX SEAMCALL

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

tdh_mem_page_aug() will support 2MB large page in the near future.  Cache
flush also needs to be 2MB instead of 4KB in such cases.  Introduce a
helper function to flush cache with page size info in preparation for large
pages.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
Reviewed-by: Binbin Wu <binbin.wu@linux.intel.com>
---
v6:
- catch up tdx_seamcall() change
---
 arch/x86/kvm/vmx/tdx_ops.h | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kvm/vmx/tdx_ops.h b/arch/x86/kvm/vmx/tdx_ops.h
index d27f281152cb..3af124711e98 100644
--- a/arch/x86/kvm/vmx/tdx_ops.h
+++ b/arch/x86/kvm/vmx/tdx_ops.h
@@ -6,6 +6,7 @@
 
 #include <linux/compiler.h>
 
+#include <asm/pgtable_types.h>
 #include <asm/cacheflush.h>
 #include <asm/asm.h>
 #include <asm/kvm_host.h>
@@ -50,6 +51,11 @@ static inline int pg_level_to_tdx_sept_level(enum pg_level level)
 	return level - 1;
 }
 
+static inline void tdx_clflush_page(hpa_t addr, enum pg_level level)
+{
+	clflush_cache_range(__va(addr), KVM_HPAGE_SIZE(level));
+}
+
 /*
  * TDX module acquires its internal lock for resources.  It doesn't spin to get
  * locks because of its restrictions of allowed execution time.  Instead, it
@@ -87,7 +93,7 @@ static inline u64 tdh_mng_addcx(hpa_t tdr, hpa_t addr)
 		.rdx = tdr,
 	};
 
-	clflush_cache_range(__va(addr), PAGE_SIZE);
+	tdx_clflush_page(addr, PG_LEVEL_4K);
 	return tdx_seamcall(TDH_MNG_ADDCX, &in, NULL);
 }
 
@@ -101,7 +107,7 @@ static inline u64 tdh_mem_page_add(hpa_t tdr, gpa_t gpa, hpa_t hpa, hpa_t source
 		.r9 = source,
 	};
 
-	clflush_cache_range(__va(hpa), PAGE_SIZE);
+	tdx_clflush_page(hpa, PG_LEVEL_4K);
 	return tdx_seamcall_sept(TDH_MEM_PAGE_ADD, &in, out);
 }
 
@@ -114,7 +120,7 @@ static inline u64 tdh_mem_sept_add(hpa_t tdr, gpa_t gpa, int level, hpa_t page,
 		.r8 = page,
 	};
 
-	clflush_cache_range(__va(page), PAGE_SIZE);
+	tdx_clflush_page(page, PG_LEVEL_4K);
 	return tdx_seamcall_sept(TDH_MEM_SEPT_ADD, &in, out);
 }
 
@@ -147,7 +153,7 @@ static inline u64 tdh_vp_addcx(hpa_t tdvpr, hpa_t addr)
 		.rdx = tdvpr,
 	};
 
-	clflush_cache_range(__va(addr), PAGE_SIZE);
+	tdx_clflush_page(addr, PG_LEVEL_4K);
 	return tdx_seamcall(TDH_VP_ADDCX, &in, NULL);
 }
 
@@ -160,7 +166,7 @@ static inline u64 tdh_mem_page_relocate(hpa_t tdr, gpa_t gpa, hpa_t hpa,
 		.r8 = hpa,
 	};
 
-	clflush_cache_range(__va(hpa), PAGE_SIZE);
+	tdx_clflush_page(hpa, PG_LEVEL_4K);
 	return tdx_seamcall_sept(TDH_MEM_PAGE_RELOCATE, &in, out);
 }
 
@@ -173,7 +179,7 @@ static inline u64 tdh_mem_page_aug(hpa_t tdr, gpa_t gpa, hpa_t hpa,
 		.r8 = hpa,
 	};
 
-	clflush_cache_range(__va(hpa), PAGE_SIZE);
+	tdx_clflush_page(hpa, PG_LEVEL_4K);
 	return tdx_seamcall_sept(TDH_MEM_PAGE_AUG, &in, out);
 }
 
@@ -204,7 +210,7 @@ static inline u64 tdh_mng_create(hpa_t tdr, int hkid)
 		.rdx = hkid,
 	};
 
-	clflush_cache_range(__va(tdr), PAGE_SIZE);
+	tdx_clflush_page(tdr, PG_LEVEL_4K);
 	return tdx_seamcall(TDH_MNG_CREATE, &in, NULL);
 }
 
@@ -215,7 +221,7 @@ static inline u64 tdh_vp_create(hpa_t tdr, hpa_t tdvpr)
 		.rdx = tdr,
 	};
 
-	clflush_cache_range(__va(tdvpr), PAGE_SIZE);
+	tdx_clflush_page(tdvpr, PG_LEVEL_4K);
 	return tdx_seamcall(TDH_VP_CREATE, &in, NULL);
 }
 
-- 
2.25.1

[PATCH v8 03/14] KVM: TDX: Pass KVM page level to tdh_mem_page_aug()

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

Level info is needed in tdx_clflush_page() to generate the correct page
size.

Besides, explicitly pass level info to SEAMCALL instead of assuming
it's zero. It works naturally when 2MB support lands.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
v7:
- Don't pass level to tdh_mem_page_add() as it supports only 4K page.
- catch up for change of tdx_seamcall()
---
 arch/x86/kvm/vmx/tdx.c     |  2 +-
 arch/x86/kvm/vmx/tdx_ops.h | 12 +++++++++---
 2 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index a71093f7c3e3..fd992966379c 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -1470,7 +1470,7 @@ static int tdx_mem_page_aug(struct kvm *kvm, gfn_t gfn,
 	union tdx_sept_entry entry;
 	u64 err;
 
-	err = tdh_mem_page_aug(kvm_tdx->tdr_pa, gpa, hpa, &out);
+	err = tdh_mem_page_aug(kvm_tdx->tdr_pa, gpa, tdx_level, hpa, &out);
 	if (unlikely(err == TDX_ERROR_SEPT_BUSY)) {
 		tdx_unpin(kvm, pfn);
 		return -EAGAIN;
diff --git a/arch/x86/kvm/vmx/tdx_ops.h b/arch/x86/kvm/vmx/tdx_ops.h
index 3af124711e98..ef4748943ac7 100644
--- a/arch/x86/kvm/vmx/tdx_ops.h
+++ b/arch/x86/kvm/vmx/tdx_ops.h
@@ -51,6 +51,11 @@ static inline int pg_level_to_tdx_sept_level(enum pg_level level)
 	return level - 1;
 }
 
+static inline enum pg_level tdx_sept_level_to_pg_level(int tdx_level)
+{
+	return tdx_level + 1;
+}
+
 static inline void tdx_clflush_page(hpa_t addr, enum pg_level level)
 {
 	clflush_cache_range(__va(addr), KVM_HPAGE_SIZE(level));
@@ -100,6 +105,7 @@ static inline u64 tdh_mng_addcx(hpa_t tdr, hpa_t addr)
 static inline u64 tdh_mem_page_add(hpa_t tdr, gpa_t gpa, hpa_t hpa, hpa_t source,
 				   struct tdx_module_args *out)
 {
+	/* TDH.MEM.PAGE.ADD() suports only 4K page. tdx 4K page level = 0 */
 	struct tdx_module_args in = {
 		.rcx = gpa,
 		.rdx = tdr,
@@ -170,16 +176,16 @@ static inline u64 tdh_mem_page_relocate(hpa_t tdr, gpa_t gpa, hpa_t hpa,
 	return tdx_seamcall_sept(TDH_MEM_PAGE_RELOCATE, &in, out);
 }
 
-static inline u64 tdh_mem_page_aug(hpa_t tdr, gpa_t gpa, hpa_t hpa,
+static inline u64 tdh_mem_page_aug(hpa_t tdr, gpa_t gpa, int level, hpa_t hpa,
 				   struct tdx_module_args *out)
 {
 	struct tdx_module_args in = {
-		.rcx = gpa,
+		.rcx = gpa | level,
 		.rdx = tdr,
 		.r8 = hpa,
 	};
 
-	tdx_clflush_page(hpa, PG_LEVEL_4K);
+	tdx_clflush_page(hpa, tdx_sept_level_to_pg_level(level));
 	return tdx_seamcall_sept(TDH_MEM_PAGE_AUG, &in, out);
 }
 
-- 
2.25.1

[PATCH v8 04/14] KVM: TDX: Pass size to reclaim_page()

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

A 2MB large page can be tdh_mem_page_aug()'ed to TD directly. In this case,
it needs to reclaim and clear the page as 2MB size.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
v5:
- Change type of page size from int to unsigned long
---
 arch/x86/kvm/vmx/tdx.c | 27 +++++++++++++++------------
 1 file changed, 15 insertions(+), 12 deletions(-)

diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index fd992966379c..8205d68ed477 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -233,12 +233,13 @@ static void tdx_disassociate_vp_on_cpu(struct kvm_vcpu *vcpu)
 	smp_call_function_single(cpu, tdx_disassociate_vp_arg, vcpu, 1);
 }
 
-static void tdx_clear_page(unsigned long page_pa)
+static void tdx_clear_page(unsigned long page_pa, unsigned long size)
 {
 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
 	void *page = __va(page_pa);
 	unsigned long i;
 
+	WARN_ON_ONCE(size % PAGE_SIZE);
 	/*
 	 * When re-assign one page from old keyid to a new keyid, MOVDIR64B is
 	 * required to clear/write the page with new keyid to prevent integrity
@@ -247,7 +248,7 @@ static void tdx_clear_page(unsigned long page_pa)
 	 * clflush doesn't flush cache with HKID set.  The cache line could be
 	 * poisoned (even without MKTME-i), clear the poison bit.
 	 */
-	for (i = 0; i < PAGE_SIZE; i += 64)
+	for (i = 0; i < size; i += 64)
 		movdir64b(page + i, zero_page);
 	/*
 	 * MOVDIR64B store uses WC buffer.  Prevent following memory reads
@@ -256,7 +257,7 @@ static void tdx_clear_page(unsigned long page_pa)
 	__mb();
 }
 
-static int __tdx_reclaim_page(hpa_t pa)
+static int __tdx_reclaim_page(hpa_t pa, enum pg_level level)
 {
 	struct tdx_module_args out;
 	u64 err;
@@ -275,17 +276,19 @@ static int __tdx_reclaim_page(hpa_t pa)
 		pr_tdx_error(TDH_PHYMEM_PAGE_RECLAIM, err, &out);
 		return -EIO;
 	}
+	/* out.r8 == tdx sept page level */
+	WARN_ON_ONCE(out.r8 != pg_level_to_tdx_sept_level(level));
 
 	return 0;
 }
 
-static int tdx_reclaim_page(hpa_t pa)
+static int tdx_reclaim_page(hpa_t pa, enum pg_level level)
 {
 	int r;
 
-	r = __tdx_reclaim_page(pa);
+	r = __tdx_reclaim_page(pa, level);
 	if (!r)
-		tdx_clear_page(pa);
+		tdx_clear_page(pa, KVM_HPAGE_SIZE(level));
 	return r;
 }
 
@@ -299,7 +302,7 @@ static void tdx_reclaim_control_page(unsigned long td_page_pa)
 	 * was already flushed by TDH.PHYMEM.CACHE.WB before here, So
 	 * cache doesn't need to be flushed again.
 	 */
-	if (tdx_reclaim_page(td_page_pa))
+	if (tdx_reclaim_page(td_page_pa, PG_LEVEL_4K))
 		/*
 		 * Leak the page on failure:
 		 * tdx_reclaim_page() returns an error if and only if there's an
@@ -530,7 +533,7 @@ void tdx_vm_free(struct kvm *kvm)
 
 	if (!kvm_tdx->tdr_pa)
 		return;
-	if (__tdx_reclaim_page(kvm_tdx->tdr_pa))
+	if (__tdx_reclaim_page(kvm_tdx->tdr_pa, PG_LEVEL_4K))
 		return;
 	/*
 	 * TDX module maps TDR with TDX global HKID.  TDX module may access TDR
@@ -543,7 +546,7 @@ void tdx_vm_free(struct kvm *kvm)
 		pr_tdx_error(TDH_PHYMEM_PAGE_WBINVD, err, NULL);
 		return;
 	}
-	tdx_clear_page(kvm_tdx->tdr_pa);
+	tdx_clear_page(kvm_tdx->tdr_pa, PAGE_SIZE);
 
 	free_page((unsigned long)__va(kvm_tdx->tdr_pa));
 	kvm_tdx->tdr_pa = 0;
@@ -1586,7 +1589,7 @@ static int tdx_sept_drop_private_spte(struct kvm *kvm, gfn_t gfn,
 		 * The HKID assigned to this TD was already freed and cache
 		 * was already flushed. We don't have to flush again.
 		 */
-		err = tdx_reclaim_page(hpa);
+		err = tdx_reclaim_page(hpa, level);
 		if (KVM_BUG_ON(err, kvm))
 			return -EIO;
 		tdx_unpin(kvm, pfn);
@@ -1619,7 +1622,7 @@ static int tdx_sept_drop_private_spte(struct kvm *kvm, gfn_t gfn,
 		pr_tdx_error(TDH_PHYMEM_PAGE_WBINVD, err, NULL);
 		return -EIO;
 	}
-	tdx_clear_page(hpa);
+	tdx_clear_page(hpa, PAGE_SIZE);
 	tdx_unpin(kvm, pfn);
 	return 0;
 }
@@ -1753,7 +1756,7 @@ static int tdx_sept_free_private_spt(struct kvm *kvm, gfn_t gfn,
 	 * already flushed. We don't have to flush again.
 	 */
 	if (!is_hkid_assigned(kvm_tdx))
-		return tdx_reclaim_page(__pa(private_spt));
+		return tdx_reclaim_page(__pa(private_spt), PG_LEVEL_4K);
 
 	/*
 	 * free_private_spt() is (obviously) called when a shadow page is being
-- 
2.25.1

[PATCH v8 05/14] KVM: TDX: Update tdx_sept_{set,drop}_private_spte() to support large page

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

Allow large page level AUG and REMOVE for TDX pages.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/vmx/tdx.c | 68 ++++++++++++++++++++++--------------------
 1 file changed, 35 insertions(+), 33 deletions(-)

diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index 8205d68ed477..d73a32588ad8 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -1454,11 +1454,12 @@ void tdx_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa, int pgd_level)
 	td_vmcs_write64(to_tdx(vcpu), SHARED_EPT_POINTER, root_hpa);
 }
 
-static void tdx_unpin(struct kvm *kvm, kvm_pfn_t pfn)
+static void tdx_unpin(struct kvm *kvm, kvm_pfn_t pfn, enum pg_level level)
 {
-	struct page *page = pfn_to_page(pfn);
+	int i;
 
-	put_page(page);
+	for (i = 0; i < KVM_PAGES_PER_HPAGE(level); i++)
+		put_page(pfn_to_page(pfn + i));
 }
 
 static int tdx_mem_page_aug(struct kvm *kvm, gfn_t gfn,
@@ -1475,7 +1476,7 @@ static int tdx_mem_page_aug(struct kvm *kvm, gfn_t gfn,
 
 	err = tdh_mem_page_aug(kvm_tdx->tdr_pa, gpa, tdx_level, hpa, &out);
 	if (unlikely(err == TDX_ERROR_SEPT_BUSY)) {
-		tdx_unpin(kvm, pfn);
+		tdx_unpin(kvm, pfn, level);
 		return -EAGAIN;
 	}
 	if (unlikely(err == (TDX_EPT_ENTRY_STATE_INCORRECT | TDX_OPERAND_ID_RCX))) {
@@ -1484,7 +1485,7 @@ static int tdx_mem_page_aug(struct kvm *kvm, gfn_t gfn,
 		if (level_state.level == tdx_level &&
 		    level_state.state == TDX_SEPT_PENDING &&
 		    entry.leaf && entry.pfn == pfn && entry.sve) {
-			tdx_unpin(kvm, pfn);
+			tdx_unpin(kvm, pfn, level);
 			WARN_ON_ONCE(!(to_kvm_tdx(kvm)->attributes &
 				       TDX_TD_ATTR_SEPT_VE_DISABLE));
 			return -EAGAIN;
@@ -1492,7 +1493,7 @@ static int tdx_mem_page_aug(struct kvm *kvm, gfn_t gfn,
 	}
 	if (KVM_BUG_ON(err, kvm)) {
 		pr_tdx_error(TDH_MEM_PAGE_AUG, err, &out);
-		tdx_unpin(kvm, pfn);
+		tdx_unpin(kvm, pfn, level);
 		return -EIO;
 	}
 
@@ -1519,7 +1520,7 @@ static int tdx_mem_page_add(struct kvm *kvm, gfn_t gfn,
 		return -EINVAL;
 
 	if (KVM_BUG_ON(!kvm_tdx->source_page, kvm)) {
-		tdx_unpin(kvm, pfn);
+		tdx_unpin(kvm, pfn, level);
 		return -EINVAL;
 	}
 
@@ -1537,7 +1538,7 @@ static int tdx_mem_page_add(struct kvm *kvm, gfn_t gfn,
 	 * fail with parameters user provided.
 	 */
 	if (err) {
-		tdx_unpin(kvm, pfn);
+		tdx_unpin(kvm, pfn, level);
 		return -EIO;
 	}
 
@@ -1548,10 +1549,7 @@ static int tdx_sept_set_private_spte(struct kvm *kvm, gfn_t gfn,
 				     enum pg_level level, kvm_pfn_t pfn)
 {
 	struct kvm_tdx *kvm_tdx = to_kvm_tdx(kvm);
-
-	/* TODO: handle large pages. */
-	if (KVM_BUG_ON(level != PG_LEVEL_4K, kvm))
-		return -EINVAL;
+	int i;
 
 	/*
 	 * Because restricted mem doesn't support page migration with
@@ -1561,7 +1559,8 @@ static int tdx_sept_set_private_spte(struct kvm *kvm, gfn_t gfn,
 	 * TODO: Once restricted mem introduces callback on page migration,
 	 * implement it and remove get_page/put_page().
 	 */
-	get_page(pfn_to_page(pfn));
+	for (i = 0; i < KVM_PAGES_PER_HPAGE(level); i++)
+		get_page(pfn_to_page(pfn + i));
 
 	if (likely(is_td_finalized(kvm_tdx)))
 		return tdx_mem_page_aug(kvm, gfn, level, pfn);
@@ -1578,11 +1577,9 @@ static int tdx_sept_drop_private_spte(struct kvm *kvm, gfn_t gfn,
 	gpa_t gpa = gfn_to_gpa(gfn);
 	hpa_t hpa = pfn_to_hpa(pfn);
 	hpa_t hpa_with_hkid;
+	int r = 0;
 	u64 err;
-
-	/* TODO: handle large pages. */
-	if (KVM_BUG_ON(level != PG_LEVEL_4K, kvm))
-		return -EINVAL;
+	int i;
 
 	if (unlikely(!is_hkid_assigned(kvm_tdx))) {
 		/*
@@ -1592,7 +1589,7 @@ static int tdx_sept_drop_private_spte(struct kvm *kvm, gfn_t gfn,
 		err = tdx_reclaim_page(hpa, level);
 		if (KVM_BUG_ON(err, kvm))
 			return -EIO;
-		tdx_unpin(kvm, pfn);
+		tdx_unpin(kvm, pfn, level);
 		return 0;
 	}
 
@@ -1609,22 +1606,27 @@ static int tdx_sept_drop_private_spte(struct kvm *kvm, gfn_t gfn,
 		return -EIO;
 	}
 
-	hpa_with_hkid = set_hkid_to_hpa(hpa, (u16)kvm_tdx->hkid);
-	do {
-		/*
-		 * TDX_OPERAND_BUSY can happen on locking PAMT entry.  Because
-		 * this page was removed above, other thread shouldn't be
-		 * repeatedly operating on this page.  Just retry loop.
-		 */
-		err = tdh_phymem_page_wbinvd(hpa_with_hkid);
-	} while (unlikely(err == (TDX_OPERAND_BUSY | TDX_OPERAND_ID_RCX)));
-	if (KVM_BUG_ON(err, kvm)) {
-		pr_tdx_error(TDH_PHYMEM_PAGE_WBINVD, err, NULL);
-		return -EIO;
+	for (i = 0; i < KVM_PAGES_PER_HPAGE(level); i++) {
+		hpa_with_hkid = set_hkid_to_hpa(hpa, (u16)kvm_tdx->hkid);
+		do {
+			/*
+			 * TDX_OPERAND_BUSY can happen on locking PAMT entry.
+			 * Because this page was removed above, other thread
+			 * shouldn't be repeatedly operating on this page.
+			 * Simple retry should work.
+			 */
+			err = tdh_phymem_page_wbinvd(hpa_with_hkid);
+		} while (unlikely(err == (TDX_OPERAND_BUSY | TDX_OPERAND_ID_RCX)));
+		if (KVM_BUG_ON(err, kvm)) {
+			pr_tdx_error(TDH_PHYMEM_PAGE_WBINVD, err, NULL);
+			r = -EIO;
+		} else {
+			tdx_clear_page(hpa, PAGE_SIZE);
+			tdx_unpin(kvm, pfn + i, PG_LEVEL_4K);
+		}
+		hpa += PAGE_SIZE;
 	}
-	tdx_clear_page(hpa, PAGE_SIZE);
-	tdx_unpin(kvm, pfn);
-	return 0;
+	return r;
 }
 
 static int tdx_sept_link_private_spt(struct kvm *kvm, gfn_t gfn,
-- 
2.25.1

[PATCH v8 06/14] KVM: MMU: Introduce level info in PFERR code

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

For TDX, EPT violation can happen when TDG.MEM.PAGE.ACCEPT.
And TDG.MEM.PAGE.ACCEPT contains the desired accept page level of TD guest.

1. KVM can map it with 4KB page while TD guest wants to accept 2MB page.

  TD guest will get TDX_PAGE_SIZE_MISMATCH and it should try to accept
  4KB size.

2. KVM can map it with 2MB page while TD guest wants to accept 4KB page.

  KVM needs to honor it because
  a) there is no way to tell guest KVM maps it as 2MB size. And
  b) guest accepts it in 4KB size since guest knows some other 4KB page
     in the same 2MB range will be used as shared page.

For case 2, it need to pass desired page level to KVM MMU page fault
handler. Use bit 29:31 of kvm PF error code for this purpose.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/include/asm/kvm_host.h | 5 +++++
 arch/x86/kvm/mmu/mmu.c          | 5 +++++
 2 files changed, 10 insertions(+)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index e4d40e31fc31..c864a1ff2eb1 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -262,6 +262,8 @@ enum x86_intercept_stage;
 #define PFERR_FETCH_BIT 4
 #define PFERR_PK_BIT 5
 #define PFERR_SGX_BIT 15
+#define PFERR_LEVEL_START_BIT 29
+#define PFERR_LEVEL_END_BIT 31
 #define PFERR_GUEST_FINAL_BIT 32
 #define PFERR_GUEST_PAGE_BIT 33
 #define PFERR_GUEST_ENC_BIT 34
@@ -274,6 +276,7 @@ enum x86_intercept_stage;
 #define PFERR_FETCH_MASK	BIT(PFERR_FETCH_BIT)
 #define PFERR_PK_MASK		BIT(PFERR_PK_BIT)
 #define PFERR_SGX_MASK		BIT(PFERR_SGX_BIT)
+#define PFERR_LEVEL_MASK	GENMASK_ULL(PFERR_LEVEL_END_BIT, PFERR_LEVEL_START_BIT)
 #define PFERR_GUEST_FINAL_MASK	BIT_ULL(PFERR_GUEST_FINAL_BIT)
 #define PFERR_GUEST_PAGE_MASK	BIT_ULL(PFERR_GUEST_PAGE_BIT)
 #define PFERR_GUEST_ENC_MASK	BIT_ULL(PFERR_GUEST_ENC_BIT)
@@ -283,6 +286,8 @@ enum x86_intercept_stage;
 				 PFERR_WRITE_MASK |		\
 				 PFERR_PRESENT_MASK)
 
+#define PFERR_LEVEL(err_code)	(((err_code) & PFERR_LEVEL_MASK) >> PFERR_LEVEL_START_BIT)
+
 /* apic attention bits */
 #define KVM_APIC_CHECK_VAPIC	0
 /*
diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index b8d6ce02e66d..081df7855065 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -4625,6 +4625,11 @@ bool __kvm_mmu_honors_guest_mtrrs(bool vm_has_noncoherent_dma)
 
 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
 {
+	u8 err_level = PFERR_LEVEL(fault->error_code);
+
+	if (err_level)
+		fault->max_level = min(fault->max_level, err_level);
+
 	/*
 	 * If the guest's MTRRs may be used to compute the "real" memtype,
 	 * restrict the mapping level to ensure KVM uses a consistent memtype
-- 
2.25.1

[PATCH v8 07/14] KVM: TDX: Pass desired page level in err code for page fault handler

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

For TDX, EPT violation can happen when TDG.MEM.PAGE.ACCEPT.
And TDG.MEM.PAGE.ACCEPT contains the desired accept page level of TD guest.

1. KVM can map it with 4KB page while TD guest wants to accept 2MB page.

  TD geust will get TDX_PAGE_SIZE_MISMATCH and it should try to accept
  4KB size.

2. KVM can map it with 2MB page while TD guest wants to accept 4KB page.

  KVM needs to honor it because
  a) there is no way to tell guest KVM maps it as 2MB size. And
  b) guest accepts it in 4KB size since guest knows some other 4KB page
     in the same 2MB range will be used as shared page.

For case 2, it need to pass desired page level to MMU's
page_fault_handler. Use bit 29:31 of kvm PF error code for this purpose.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/vmx/common.h   |  6 +++++-
 arch/x86/kvm/vmx/tdx.c      | 18 ++++++++++++++++--
 arch/x86/kvm/vmx/tdx_arch.h | 19 +++++++++++++++++++
 arch/x86/kvm/vmx/vmx.c      |  2 +-
 4 files changed, 41 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/vmx/common.h b/arch/x86/kvm/vmx/common.h
index 027aa4175d2c..787f59c44abc 100644
--- a/arch/x86/kvm/vmx/common.h
+++ b/arch/x86/kvm/vmx/common.h
@@ -67,7 +67,8 @@ static inline void vmx_handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu,
 }
 
 static inline int __vmx_handle_ept_violation(struct kvm_vcpu *vcpu, gpa_t gpa,
-					     unsigned long exit_qualification)
+					     unsigned long exit_qualification,
+					     int err_page_level)
 {
 	u64 error_code;
 
@@ -90,6 +91,9 @@ static inline int __vmx_handle_ept_violation(struct kvm_vcpu *vcpu, gpa_t gpa,
 	if (kvm_is_private_gpa(vcpu->kvm, gpa))
 		error_code |= PFERR_GUEST_ENC_MASK;
 
+	if (err_page_level > PG_LEVEL_NONE)
+		error_code |= (err_page_level << PFERR_LEVEL_START_BIT) & PFERR_LEVEL_MASK;
+
 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
 }
 
diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index d73a32588ad8..6941e9483e7e 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -1812,7 +1812,20 @@ void tdx_deliver_interrupt(struct kvm_lapic *apic, int delivery_mode,
 
 static int tdx_handle_ept_violation(struct kvm_vcpu *vcpu)
 {
+	union tdx_ext_exit_qualification ext_exit_qual;
 	unsigned long exit_qual;
+	int err_page_level = 0;
+
+	ext_exit_qual.full = tdexit_ext_exit_qual(vcpu);
+
+	if (ext_exit_qual.type >= NUM_EXT_EXIT_QUAL) {
+		pr_err("EPT violation at gpa 0x%lx, with invalid ext exit qualification type 0x%x\n",
+			tdexit_gpa(vcpu), ext_exit_qual.type);
+		kvm_vm_bugged(vcpu->kvm);
+		return 0;
+	} else if (ext_exit_qual.type == EXT_EXIT_QUAL_ACCEPT) {
+		err_page_level = tdx_sept_level_to_pg_level(ext_exit_qual.req_sept_level);
+	}
 
 	if (kvm_is_private_gpa(vcpu->kvm, tdexit_gpa(vcpu))) {
 		/*
@@ -1839,7 +1852,7 @@ static int tdx_handle_ept_violation(struct kvm_vcpu *vcpu)
 	}
 
 	trace_kvm_page_fault(vcpu, tdexit_gpa(vcpu), exit_qual);
-	return __vmx_handle_ept_violation(vcpu, tdexit_gpa(vcpu), exit_qual);
+	return __vmx_handle_ept_violation(vcpu, tdexit_gpa(vcpu), exit_qual, err_page_level);
 }
 
 static int tdx_handle_ept_misconfig(struct kvm_vcpu *vcpu)
@@ -3027,7 +3040,8 @@ int tdx_pre_memory_mapping(struct kvm_vcpu *vcpu,
 
 	/* TDX supports only 4K to pre-populate. */
 	*max_level = PG_LEVEL_4K;
-	*error_code = TDX_SEPT_PFERR;
+	*error_code = TDX_SEPT_PFERR |
+		((PG_LEVEL_4K << PFERR_LEVEL_START_BIT) & PFERR_LEVEL_MASK);
 
 	r = get_user_pages_fast(mapping->source, 1, 0, &page);
 	if (r < 0)
diff --git a/arch/x86/kvm/vmx/tdx_arch.h b/arch/x86/kvm/vmx/tdx_arch.h
index 87ef22e9cd49..19f2deafde5b 100644
--- a/arch/x86/kvm/vmx/tdx_arch.h
+++ b/arch/x86/kvm/vmx/tdx_arch.h
@@ -221,6 +221,25 @@ union tdx_sept_level_state {
 	u64 raw;
 };
 
+union tdx_ext_exit_qualification {
+	struct {
+		u64 type		:  4;
+		u64 reserved0		: 28;
+		u64 req_sept_level	:  3;
+		u64 err_sept_level	:  3;
+		u64 err_sept_state	:  8;
+		u64 err_sept_is_leaf	:  1;
+		u64 reserved1		: 17;
+	};
+	u64 full;
+};
+
+enum tdx_ext_exit_qualification_type {
+	EXT_EXIT_QUAL_NONE = 0,
+	EXT_EXIT_QUAL_ACCEPT = 1,
+	NUM_EXT_EXIT_QUAL,
+};
+
 /*
  * Global scope metadata field ID.
  * See Table "Global Scope Metadata", TDX module 1.5 ABI spec.
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index f8a00a766c40..a2004a0feb1c 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -5752,7 +5752,7 @@ static int handle_ept_violation(struct kvm_vcpu *vcpu)
 	if (unlikely(allow_smaller_maxphyaddr && !kvm_vcpu_is_legal_gpa(vcpu, gpa)))
 		return kvm_emulate_instruction(vcpu, 0);
 
-	return __vmx_handle_ept_violation(vcpu, gpa, exit_qualification);
+	return __vmx_handle_ept_violation(vcpu, gpa, exit_qualification, PG_LEVEL_NONE);
 }
 
 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
-- 
2.25.1

[PATCH v8 08/14] KVM: x86/tdp_mmu: Allocate private page table for large page split

[isaku.yamahata]

From: Isaku Yamahata <isaku.yamahata@intel.com>

Make tdp_mmu_alloc_sp_split() aware of private page table.

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/mmu/mmu_internal.h | 9 +++++++++
 arch/x86/kvm/mmu/tdp_mmu.c      | 8 ++++++--
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/mmu/mmu_internal.h b/arch/x86/kvm/mmu/mmu_internal.h
index 9e2c7c6d85bf..9aa4c6ffa207 100644
--- a/arch/x86/kvm/mmu/mmu_internal.h
+++ b/arch/x86/kvm/mmu/mmu_internal.h
@@ -201,6 +201,15 @@ static inline void kvm_mmu_alloc_private_spt(struct kvm_vcpu *vcpu, struct kvm_m
 	}
 }
 
+static inline int kvm_alloc_private_spt_for_split(struct kvm_mmu_page *sp, gfp_t gfp)
+{
+	gfp &= ~__GFP_ZERO;
+	sp->private_spt = (void *)__get_free_page(gfp);
+	if (!sp->private_spt)
+		return -ENOMEM;
+	return 0;
+}
+
 static inline void kvm_mmu_free_private_spt(struct kvm_mmu_page *sp)
 {
 	if (sp->private_spt)
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 1a0e4baa8311..66de875d3de1 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -1593,8 +1593,12 @@ static struct kvm_mmu_page *__tdp_mmu_alloc_sp_for_split(gfp_t gfp, union kvm_mm
 
 	sp->role = role;
 	sp->spt = (void *)__get_free_page(gfp);
-	/* TODO: large page support for private GPA. */
-	WARN_ON_ONCE(kvm_mmu_page_role_is_private(role));
+	if (kvm_mmu_page_role_is_private(role)) {
+		if (kvm_alloc_private_spt_for_split(sp, gfp)) {
+			free_page((unsigned long)sp->spt);
+			sp->spt = NULL;
+		}
+	}
 	if (!sp->spt) {
 		kmem_cache_free(mmu_page_header_cache, sp);
 		return NULL;
-- 
2.25.1

[PATCH v8 09/14] KVM: x86/tdp_mmu: Split the large page when zap leaf

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

When TDX enabled, a large page cannot be zapped if it contains mixed
pages. In this case, it has to split the large page.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
---
v7:
- remote unnecessary tlb shoot down in tdp_mmu_zap_leafs() to free unused
  split_sp.
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/mmu/mmu.c          |  6 ++--
 arch/x86/kvm/mmu/mmu_internal.h |  9 +++++
 arch/x86/kvm/mmu/tdp_mmu.c      | 60 ++++++++++++++++++++++++++++++---
 3 files changed, 68 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 081df7855065..fa7fabc410c4 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -7473,8 +7473,8 @@ bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
 	return kvm_unmap_gfn_range(kvm, range);
 }
 
-static bool hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn,
-				int level)
+bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn,
+			     int level)
 {
 	return lpage_info_slot(gfn, slot, level)->disallow_lpage & KVM_LPAGE_MIXED_FLAG;
 }
@@ -7501,7 +7501,7 @@ static bool hugepage_has_attrs(struct kvm *kvm, struct kvm_memory_slot *slot,
 		return kvm_range_has_memory_attributes(kvm, start, end, attrs);
 
 	for (gfn = start; gfn < end; gfn += KVM_PAGES_PER_HPAGE(level - 1)) {
-		if (hugepage_test_mixed(slot, gfn, level - 1) ||
+		if (kvm_hugepage_test_mixed(slot, gfn, level - 1) ||
 		    attrs != kvm_get_memory_attributes(kvm, gfn))
 			return false;
 	}
diff --git a/arch/x86/kvm/mmu/mmu_internal.h b/arch/x86/kvm/mmu/mmu_internal.h
index 9aa4c6ffa207..315c123affaf 100644
--- a/arch/x86/kvm/mmu/mmu_internal.h
+++ b/arch/x86/kvm/mmu/mmu_internal.h
@@ -430,4 +430,13 @@ void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
 void track_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
 void untrack_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
 
+#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
+bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn, int level);
+#else
+static inline bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn, int level)
+{
+	return false;
+}
+#endif
+
 #endif /* __KVM_X86_MMU_INTERNAL_H */
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 66de875d3de1..e3682794adda 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -953,6 +953,14 @@ bool kvm_tdp_mmu_zap_sp(struct kvm *kvm, struct kvm_mmu_page *sp)
 	return true;
 }
 
+
+static struct kvm_mmu_page *tdp_mmu_alloc_sp_for_split(struct kvm *kvm,
+						       struct tdp_iter *iter,
+						       bool shared);
+
+static int tdp_mmu_split_huge_page(struct kvm *kvm, struct tdp_iter *iter,
+				   struct kvm_mmu_page *sp, bool shared);
+
 /*
  * If can_yield is true, will release the MMU lock and reschedule if the
  * scheduler needs the CPU or there is contention on the MMU lock. If this
@@ -964,14 +972,16 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
 			      gfn_t start, gfn_t end, bool can_yield, bool flush,
 			      bool zap_private)
 {
+	bool is_private = is_private_sp(root);
+	struct kvm_mmu_page *split_sp = NULL;
 	struct tdp_iter iter;
 
 	end = min(end, tdp_mmu_max_gfn_exclusive());
 
 	lockdep_assert_held_write(&kvm->mmu_lock);
 
-	WARN_ON_ONCE(zap_private && !is_private_sp(root));
-	if (!zap_private && is_private_sp(root))
+	WARN_ON_ONCE(zap_private && !is_private);
+	if (!zap_private && is_private)
 		return false;
 
 	/*
@@ -995,12 +1005,56 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
 		    !is_last_spte(iter.old_spte, iter.level))
 			continue;
 
+		if (is_private && kvm_gfn_shared_mask(kvm) &&
+		    is_large_pte(iter.old_spte)) {
+			gfn_t gfn = iter.gfn & ~kvm_gfn_shared_mask(kvm);
+			gfn_t mask = KVM_PAGES_PER_HPAGE(iter.level) - 1;
+			struct kvm_memory_slot *slot;
+			struct kvm_mmu_page *sp;
+
+			slot = gfn_to_memslot(kvm, gfn);
+			if (kvm_hugepage_test_mixed(slot, gfn, iter.level) ||
+			    (gfn & mask) < start ||
+			    end < (gfn & mask) + KVM_PAGES_PER_HPAGE(iter.level)) {
+				WARN_ON_ONCE(!can_yield);
+				if (split_sp) {
+					sp = split_sp;
+					split_sp = NULL;
+					sp->role = tdp_iter_child_role(&iter);
+				} else {
+					WARN_ON(iter.yielded);
+					if (flush && can_yield) {
+						kvm_flush_remote_tlbs(kvm);
+						flush = false;
+					}
+					sp = tdp_mmu_alloc_sp_for_split(kvm, &iter, false);
+					if (iter.yielded) {
+						split_sp = sp;
+						continue;
+					}
+				}
+				KVM_BUG_ON(!sp, kvm);
+
+				tdp_mmu_init_sp(sp, iter.sptep, iter.gfn);
+				if (tdp_mmu_split_huge_page(kvm, &iter, sp, false)) {
+					/* force retry on this gfn. */
+					iter.yielded = true;
+					split_sp = sp;
+				} else
+					flush = true;
+				continue;
+			}
+		}
+
 		tdp_mmu_iter_set_spte(kvm, &iter, SHADOW_NONPRESENT_VALUE);
 		flush = true;
 	}
 
 	rcu_read_unlock();
 
+	if (split_sp)
+		tdp_mmu_free_sp(split_sp);
+
 	/*
 	 * Because this flow zaps _only_ leaf SPTEs, the caller doesn't need
 	 * to provide RCU protection as no 'struct kvm_mmu_page' will be freed.
@@ -1617,8 +1671,6 @@ static struct kvm_mmu_page *tdp_mmu_alloc_sp_for_split(struct kvm *kvm,
 	kvm_lockdep_assert_mmu_lock_held(kvm, shared);
 	KVM_BUG_ON(kvm_mmu_page_role_is_private(role) !=
 		   is_private_sptep(iter->sptep), kvm);
-	/* TODO: Large page isn't supported for private SPTE yet. */
-	KVM_BUG_ON(kvm_mmu_page_role_is_private(role), kvm);
 
 	/*
 	 * Since we are allocating while under the MMU lock we have to be
-- 
2.25.1

[PATCH v8 10/14] KVM: x86/tdp_mmu, TDX: Split a large page when 4KB page within it converted to shared

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

When mapping the shared page for TDX, it needs to zap private alias.

In the case that private page is mapped as large page (2MB), it can be
removed directly only when the whole 2MB is converted to shared.
Otherwise, it has to split 2MB page into 512 4KB page, and only remove
the pages that converted to shared.

When a present large leaf spte switches to present non-leaf spte, TDX needs
to split the corresponding SEPT page to reflect it.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
v7:
- catch up for tdx_seamcall() change
- typo in a comment of __set_private_spte_present()
- improved a comment in tdx_sept_split_private_spt()

v6:
- repeat TDH.MEM.PAGE.DEMOTE on TDX_INTERRUPTED_RESTARTABLE

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/include/asm/kvm-x86-ops.h |  1 +
 arch/x86/include/asm/kvm_host.h    |  2 ++
 arch/x86/kvm/mmu/tdp_mmu.c         | 21 ++++++++++++++++-----
 arch/x86/kvm/vmx/tdx.c             | 27 +++++++++++++++++++++++++--
 arch/x86/kvm/vmx/tdx_arch.h        |  1 +
 arch/x86/kvm/vmx/tdx_errno.h       |  1 +
 arch/x86/kvm/vmx/tdx_ops.h         | 13 +++++++++++++
 7 files changed, 59 insertions(+), 7 deletions(-)

diff --git a/arch/x86/include/asm/kvm-x86-ops.h b/arch/x86/include/asm/kvm-x86-ops.h
index 3a7140129855..ada6865100ee 100644
--- a/arch/x86/include/asm/kvm-x86-ops.h
+++ b/arch/x86/include/asm/kvm-x86-ops.h
@@ -105,6 +105,7 @@ KVM_X86_OP_OPTIONAL_RET0(get_mt_mask)
 KVM_X86_OP(load_mmu_pgd)
 KVM_X86_OP_OPTIONAL(link_private_spt)
 KVM_X86_OP_OPTIONAL(free_private_spt)
+KVM_X86_OP_OPTIONAL(split_private_spt)
 KVM_X86_OP_OPTIONAL(set_private_spte)
 KVM_X86_OP_OPTIONAL(remove_private_spte)
 KVM_X86_OP_OPTIONAL(zap_private_spte)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index c864a1ff2eb1..9c9742cc469c 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1766,6 +1766,8 @@ struct kvm_x86_ops {
 				void *private_spt);
 	int (*free_private_spt)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
 				void *private_spt);
+	int (*split_private_spt)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
+				  void *private_spt);
 	int (*set_private_spte)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
 				 kvm_pfn_t pfn);
 	int (*remove_private_spte)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index e3682794adda..0ac2a4911fd1 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -588,23 +588,34 @@ static int __must_check __set_private_spte_present(struct kvm *kvm, tdp_ptep_t s
 {
 	bool was_present = is_shadow_present_pte(old_spte);
 	bool is_present = is_shadow_present_pte(new_spte);
+	bool was_leaf = was_present && is_last_spte(old_spte, level);
 	bool is_leaf = is_present && is_last_spte(new_spte, level);
 	kvm_pfn_t new_pfn = spte_to_pfn(new_spte);
+	void *private_spt;
 	int ret = 0;
 
 	lockdep_assert_held(&kvm->mmu_lock);
-	/* TDP MMU doesn't change present -> present */
-	KVM_BUG_ON(was_present, kvm);
 
 	/*
 	 * Use different call to either set up middle level
 	 * private page table, or leaf.
 	 */
-	if (is_leaf)
+	if (level > PG_LEVEL_4K && was_leaf && !is_leaf) {
+		/*
+		 * splitting large page into 4KB.
+		 * tdp_mmu_split_huge_page() => tdp_mmu_link_sp()
+		 */
+		private_spt = get_private_spt(gfn, new_spte, level);
+		KVM_BUG_ON(!private_spt, kvm);
+		ret = static_call(kvm_x86_zap_private_spte)(kvm, gfn, level);
+		kvm_flush_remote_tlbs(kvm);
+		if (!ret)
+			ret = static_call(kvm_x86_split_private_spt)(kvm, gfn,
+								     level, private_spt);
+	} else if (is_leaf)
 		ret = static_call(kvm_x86_set_private_spte)(kvm, gfn, level, new_pfn);
 	else {
-		void *private_spt = get_private_spt(gfn, new_spte, level);
-
+		private_spt = get_private_spt(gfn, new_spte, level);
 		KVM_BUG_ON(!private_spt, kvm);
 		ret = static_call(kvm_x86_link_private_spt)(kvm, gfn, level, private_spt);
 	}
diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index 6941e9483e7e..88af64658a9c 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -1650,6 +1650,30 @@ static int tdx_sept_link_private_spt(struct kvm *kvm, gfn_t gfn,
 	return 0;
 }
 
+static int tdx_sept_split_private_spt(struct kvm *kvm, gfn_t gfn,
+				      enum pg_level level, void *private_spt)
+{
+	int tdx_level = pg_level_to_tdx_sept_level(level);
+	struct kvm_tdx *kvm_tdx = to_kvm_tdx(kvm);
+	gpa_t gpa = gfn_to_gpa(gfn) & KVM_HPAGE_MASK(level);
+	hpa_t hpa = __pa(private_spt);
+	struct tdx_module_args out;
+	u64 err;
+
+	/* See comment in tdx_sept_set_private_spte() to pin pages. */
+	do {
+		err = tdh_mem_page_demote(kvm_tdx->tdr_pa, gpa, tdx_level, hpa, &out);
+	} while (err == TDX_INTERRUPTED_RESTARTABLE);
+	if (unlikely(err == TDX_ERROR_SEPT_BUSY))
+		return -EAGAIN;
+	if (KVM_BUG_ON(err, kvm)) {
+		pr_tdx_error(TDH_MEM_PAGE_DEMOTE, err, &out);
+		return -EIO;
+	}
+
+	return 0;
+}
+
 static int tdx_sept_zap_private_spte(struct kvm *kvm, gfn_t gfn,
 				      enum pg_level level)
 {
@@ -1663,8 +1687,6 @@ static int tdx_sept_zap_private_spte(struct kvm *kvm, gfn_t gfn,
 	if (unlikely(!is_hkid_assigned(kvm_tdx)))
 		return 0;
 
-	/* For now large page isn't supported yet. */
-	WARN_ON_ONCE(level != PG_LEVEL_4K);
 	err = tdh_mem_range_block(kvm_tdx->tdr_pa, gpa, tdx_level, &out);
 	if (unlikely(err == TDX_ERROR_SEPT_BUSY))
 		return -EAGAIN;
@@ -3308,6 +3330,7 @@ int __init tdx_hardware_setup(struct kvm_x86_ops *x86_ops)
 
 	x86_ops->link_private_spt = tdx_sept_link_private_spt;
 	x86_ops->free_private_spt = tdx_sept_free_private_spt;
+	x86_ops->split_private_spt = tdx_sept_split_private_spt;
 	x86_ops->set_private_spte = tdx_sept_set_private_spte;
 	x86_ops->remove_private_spte = tdx_sept_remove_private_spte;
 	x86_ops->zap_private_spte = tdx_sept_zap_private_spte;
diff --git a/arch/x86/kvm/vmx/tdx_arch.h b/arch/x86/kvm/vmx/tdx_arch.h
index 19f2deafde5b..bb324f744bbf 100644
--- a/arch/x86/kvm/vmx/tdx_arch.h
+++ b/arch/x86/kvm/vmx/tdx_arch.h
@@ -21,6 +21,7 @@
 #define TDH_MNG_CREATE			9
 #define TDH_VP_CREATE			10
 #define TDH_MNG_RD			11
+#define TDH_MEM_PAGE_DEMOTE		15
 #define TDH_MR_EXTEND			16
 #define TDH_MR_FINALIZE			17
 #define TDH_VP_FLUSH			18
diff --git a/arch/x86/kvm/vmx/tdx_errno.h b/arch/x86/kvm/vmx/tdx_errno.h
index 5366bf476d2c..416708e6cbb7 100644
--- a/arch/x86/kvm/vmx/tdx_errno.h
+++ b/arch/x86/kvm/vmx/tdx_errno.h
@@ -11,6 +11,7 @@
  */
 #define TDX_NON_RECOVERABLE_VCPU		0x4000000100000000ULL
 #define TDX_INTERRUPTED_RESUMABLE		0x8000000300000000ULL
+#define TDX_INTERRUPTED_RESTARTABLE		0x8000000400000000ULL
 #define TDX_OPERAND_INVALID			0xC000010000000000ULL
 #define TDX_OPERAND_BUSY			0x8000020000000000ULL
 #define TDX_PREVIOUS_TLB_EPOCH_BUSY		0x8000020100000000ULL
diff --git a/arch/x86/kvm/vmx/tdx_ops.h b/arch/x86/kvm/vmx/tdx_ops.h
index ef4748943ac7..d8f0d9aa7439 100644
--- a/arch/x86/kvm/vmx/tdx_ops.h
+++ b/arch/x86/kvm/vmx/tdx_ops.h
@@ -241,6 +241,19 @@ static inline u64 tdh_mng_rd(hpa_t tdr, u64 field, struct tdx_module_args *out)
 	return tdx_seamcall(TDH_MNG_RD, &in, out);
 }
 
+static inline u64 tdh_mem_page_demote(hpa_t tdr, gpa_t gpa, int level, hpa_t page,
+				      struct tdx_module_args *out)
+{
+	struct tdx_module_args in = {
+		.rcx = gpa | level,
+		.rdx = tdr,
+		.r8 = page,
+	};
+
+	tdx_clflush_page(page, PG_LEVEL_4K);
+	return tdx_seamcall_sept(TDH_MEM_PAGE_DEMOTE, &in, out);
+}
+
 static inline u64 tdh_mr_extend(hpa_t tdr, gpa_t gpa,
 				struct tdx_module_args *out)
 {
-- 
2.25.1

[PATCH v8 11/14] KVM: x86/tdp_mmu: Try to merge pages into a large page

[isaku.yamahata]

From: Isaku Yamahata <isaku.yamahata@intel.com>

When a large page is passed to the KVM page fault handler and some of sub
pages are already populated, try to merge sub pages into a large page.
This situation can happen when the guest converts small pages into shared
and convert it back into private.

When a large page is passed to KVM mmu page fault handler and the spte
corresponding to the page is non-leaf (one or more of sub pages are already
populated at lower page level), the current kvm mmu zaps non-leaf spte at a
large page level, and populate a leaf spte at that level.  Thus small pages
are converted into a large page.  However, it doesn't work for TDX because
zapping and re-populating results in zeroing page content.  Instead,
populate all small pages and merge them into a large page.

Merging pages into a large page can fail when some sub pages are accepted
and some are not.  In such case, with the assumption that guest tries to
accept at large page size for performance when possible, don't try to be
smart to identify which page is still pending, map all pages at lower page
level, and let vcpu re-execute.

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
v7:
- typo freezed => frozen
- return 0 when page is merged into 2M large page instead of -EAGAIN

v5:
- Fix memory leak

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/include/asm/kvm-x86-ops.h |   2 +
 arch/x86/include/asm/kvm_host.h    |   4 +
 arch/x86/kvm/mmu/tdp_iter.c        |  37 ++++--
 arch/x86/kvm/mmu/tdp_iter.h        |   2 +
 arch/x86/kvm/mmu/tdp_mmu.c         | 176 ++++++++++++++++++++++++++++-
 5 files changed, 211 insertions(+), 10 deletions(-)

diff --git a/arch/x86/include/asm/kvm-x86-ops.h b/arch/x86/include/asm/kvm-x86-ops.h
index ada6865100ee..6741cc518dae 100644
--- a/arch/x86/include/asm/kvm-x86-ops.h
+++ b/arch/x86/include/asm/kvm-x86-ops.h
@@ -106,9 +106,11 @@ KVM_X86_OP(load_mmu_pgd)
 KVM_X86_OP_OPTIONAL(link_private_spt)
 KVM_X86_OP_OPTIONAL(free_private_spt)
 KVM_X86_OP_OPTIONAL(split_private_spt)
+KVM_X86_OP_OPTIONAL(merge_private_spt)
 KVM_X86_OP_OPTIONAL(set_private_spte)
 KVM_X86_OP_OPTIONAL(remove_private_spte)
 KVM_X86_OP_OPTIONAL(zap_private_spte)
+KVM_X86_OP_OPTIONAL(unzap_private_spte)
 KVM_X86_OP(has_wbinvd_exit)
 KVM_X86_OP(get_l2_tsc_offset)
 KVM_X86_OP(get_l2_tsc_multiplier)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 9c9742cc469c..a02b14be186c 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -147,6 +147,7 @@
 #define KVM_MAX_HUGEPAGE_LEVEL	PG_LEVEL_1G
 #define KVM_NR_PAGE_SIZES	(KVM_MAX_HUGEPAGE_LEVEL - PG_LEVEL_4K + 1)
 #define KVM_HPAGE_GFN_SHIFT(x)	(((x) - 1) * 9)
+#define KVM_HPAGE_GFN_MASK(x)	(~((1UL << KVM_HPAGE_GFN_SHIFT(x)) - 1))
 #define KVM_HPAGE_SHIFT(x)	(PAGE_SHIFT + KVM_HPAGE_GFN_SHIFT(x))
 #define KVM_HPAGE_SIZE(x)	(1UL << KVM_HPAGE_SHIFT(x))
 #define KVM_HPAGE_MASK(x)	(~(KVM_HPAGE_SIZE(x) - 1))
@@ -1768,11 +1769,14 @@ struct kvm_x86_ops {
 				void *private_spt);
 	int (*split_private_spt)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
 				  void *private_spt);
+	int (*merge_private_spt)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
+				 void *private_spt);
 	int (*set_private_spte)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
 				 kvm_pfn_t pfn);
 	int (*remove_private_spte)(struct kvm *kvm, gfn_t gfn, enum pg_level level,
 				    kvm_pfn_t pfn);
 	int (*zap_private_spte)(struct kvm *kvm, gfn_t gfn, enum pg_level level);
+	int (*unzap_private_spte)(struct kvm *kvm, gfn_t gfn, enum pg_level level);
 
 	bool (*has_wbinvd_exit)(void);
 
diff --git a/arch/x86/kvm/mmu/tdp_iter.c b/arch/x86/kvm/mmu/tdp_iter.c
index 04c247bfe318..c4a18703f88a 100644
--- a/arch/x86/kvm/mmu/tdp_iter.c
+++ b/arch/x86/kvm/mmu/tdp_iter.c
@@ -71,6 +71,14 @@ tdp_ptep_t spte_to_child_pt(u64 spte, int level)
 	return (tdp_ptep_t)__va(spte_to_pfn(spte) << PAGE_SHIFT);
 }
 
+static void step_down(struct tdp_iter *iter, tdp_ptep_t child_pt)
+{
+	iter->level--;
+	iter->pt_path[iter->level - 1] = child_pt;
+	iter->gfn = gfn_round_for_level(iter->next_last_level_gfn, iter->level);
+	tdp_iter_refresh_sptep(iter);
+}
+
 /*
  * Steps down one level in the paging structure towards the goal GFN. Returns
  * true if the iterator was able to step down a level, false otherwise.
@@ -92,14 +100,28 @@ static bool try_step_down(struct tdp_iter *iter)
 	if (!child_pt)
 		return false;
 
-	iter->level--;
-	iter->pt_path[iter->level - 1] = child_pt;
-	iter->gfn = gfn_round_for_level(iter->next_last_level_gfn, iter->level);
-	tdp_iter_refresh_sptep(iter);
-
+	step_down(iter, child_pt);
 	return true;
 }
 
+/* Steps down for frozen spte.  Don't re-read sptep because it was frozen. */
+void tdp_iter_step_down(struct tdp_iter *iter, tdp_ptep_t child_pt)
+{
+	WARN_ON_ONCE(!child_pt);
+	WARN_ON_ONCE(iter->yielded);
+	WARN_ON_ONCE(iter->level == iter->min_level);
+
+	step_down(iter, child_pt);
+}
+
+void tdp_iter_step_side(struct tdp_iter *iter)
+{
+	iter->gfn += KVM_PAGES_PER_HPAGE(iter->level);
+	iter->next_last_level_gfn = iter->gfn;
+	iter->sptep++;
+	iter->old_spte = kvm_tdp_mmu_read_spte(iter->sptep);
+}
+
 /*
  * Steps to the next entry in the current page table, at the current page table
  * level. The next entry could point to a page backing guest memory or another
@@ -117,10 +139,7 @@ static bool try_step_side(struct tdp_iter *iter)
 	    (SPTE_ENT_PER_PAGE - 1))
 		return false;
 
-	iter->gfn += KVM_PAGES_PER_HPAGE(iter->level);
-	iter->next_last_level_gfn = iter->gfn;
-	iter->sptep++;
-	iter->old_spte = kvm_tdp_mmu_read_spte(iter->sptep);
+	tdp_iter_step_side(iter);
 
 	return true;
 }
diff --git a/arch/x86/kvm/mmu/tdp_iter.h b/arch/x86/kvm/mmu/tdp_iter.h
index a9c9cd0db20a..ca00db799a50 100644
--- a/arch/x86/kvm/mmu/tdp_iter.h
+++ b/arch/x86/kvm/mmu/tdp_iter.h
@@ -134,6 +134,8 @@ void tdp_iter_start(struct tdp_iter *iter, struct kvm_mmu_page *root,
 		    int min_level, gfn_t next_last_level_gfn);
 void tdp_iter_next(struct tdp_iter *iter);
 void tdp_iter_restart(struct tdp_iter *iter);
+void tdp_iter_step_side(struct tdp_iter *iter);
+void tdp_iter_step_down(struct tdp_iter *iter, tdp_ptep_t child_pt);
 
 static inline union kvm_mmu_page_role tdp_iter_child_role(struct tdp_iter *iter)
 {
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 0ac2a4911fd1..556974361d36 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -1205,6 +1205,180 @@ void kvm_tdp_mmu_invalidate_all_roots(struct kvm *kvm, bool skip_private)
 	}
 }
 
+static int tdp_mmu_iter_step_side(int i, struct tdp_iter *iter)
+{
+	i++;
+
+	/*
+	 * if i = SPTE_ENT_PER_PAGE, tdp_iter_step_side() results
+	 * in reading the entry beyond the last entry.
+	 */
+	if (i < SPTE_ENT_PER_PAGE)
+		tdp_iter_step_side(iter);
+
+	return i;
+}
+
+static int tdp_mmu_merge_private_spt(struct kvm_vcpu *vcpu,
+				     struct kvm_page_fault *fault,
+				     struct tdp_iter *iter, u64 new_spte)
+{
+	u64 *sptep = rcu_dereference(iter->sptep);
+	u64 old_spte = iter->old_spte;
+	struct kvm_mmu_page *child_sp;
+	struct kvm *kvm = vcpu->kvm;
+	struct tdp_iter child_iter;
+	int level = iter->level;
+	gfn_t gfn = iter->gfn;
+	tdp_ptep_t child_pt;
+	u64 child_spte;
+	int ret = 0;
+	int i;
+
+	/*
+	 * TDX KVM supports only 2MB large page.  It's not supported to merge
+	 * 2MB pages into 1GB page at the moment.
+	 */
+	WARN_ON_ONCE(fault->goal_level != PG_LEVEL_2M);
+	WARN_ON_ONCE(iter->level != PG_LEVEL_2M);
+	WARN_ON_ONCE(!is_large_pte(new_spte));
+
+	/* Freeze the spte to prevent other threads from working spte. */
+	if (!try_cmpxchg64(sptep, &iter->old_spte, REMOVED_SPTE))
+		return -EBUSY;
+
+	/*
+	 * Step down to the child spte.  Because tdp_iter_next() assumes the
+	 * parent spte isn't frozen, do it manually.
+	 */
+	child_pt = spte_to_child_pt(iter->old_spte, iter->level);
+	child_sp = sptep_to_sp(child_pt);
+	WARN_ON_ONCE(child_sp->role.level != PG_LEVEL_4K);
+	WARN_ON_ONCE(!kvm_mmu_page_role_is_private(child_sp->role));
+
+	/* Don't modify iter as the caller will use iter after this function. */
+	child_iter = *iter;
+	/* Adjust the target gfn to the head gfn of the large page. */
+	child_iter.next_last_level_gfn &= -KVM_PAGES_PER_HPAGE(level);
+	tdp_iter_step_down(&child_iter, child_pt);
+
+	/*
+	 * All child pages are required to be populated for merging them into a
+	 * large page.  Populate all child spte.
+	 */
+	for (i = 0; i < SPTE_ENT_PER_PAGE; i = tdp_mmu_iter_step_side(i, &child_iter)) {
+		int tmp;
+
+		WARN_ON_ONCE(child_iter.level != PG_LEVEL_4K);
+
+		if (is_shadow_present_pte(child_iter.old_spte)) {
+			/* TODO: relocate page for huge page. */
+			if (WARN_ON_ONCE(spte_to_pfn(child_iter.old_spte) !=
+					 spte_to_pfn(new_spte) + i)) {
+				if (!ret)
+					ret = -EAGAIN;
+				continue;
+			}
+			/*
+			 * When SEPT_VE_DISABLE=true and the page state is
+			 * pending, this case can happen.  Just resume the vcpu
+			 * again with the expectation for other vcpu to accept
+			 * this page.
+			 */
+			if (child_iter.gfn == fault->gfn) {
+				if (!ret)
+					ret = -EAGAIN;
+			}
+			continue;
+		}
+
+		child_spte = make_huge_page_split_spte(kvm, new_spte, child_sp->role, i);
+		/*
+		 * Because other thread may have started to operate on this spte
+		 * before freezing the parent spte,  Use atomic version to
+		 * prevent race.
+		 */
+		tmp = tdp_mmu_set_spte_atomic(vcpu->kvm, &child_iter, child_spte);
+		if (tmp == -EBUSY || tmp == -EAGAIN) {
+			/*
+			 * There was a race condition.  Populate remaining 4K
+			 * spte to resolve fault->gfn to guarantee the forward
+			 * progress.
+			 */
+			if (!ret)
+				ret = tmp;
+		} else if (tmp) {
+			ret = tmp;
+			goto out;
+		}
+	}
+	if (ret)
+		goto out;
+
+	/* Prevent the Secure-EPT entry from being used. */
+	ret = static_call(kvm_x86_zap_private_spte)(kvm, gfn, level);
+	if (ret)
+		goto out;
+	kvm_flush_remote_tlbs_range(kvm, gfn & KVM_HPAGE_GFN_MASK(level),
+				    KVM_PAGES_PER_HPAGE(level));
+
+	/* Merge pages into a large page. */
+	ret = static_call(kvm_x86_merge_private_spt)(kvm, gfn, level,
+						     kvm_mmu_private_spt(child_sp));
+	/*
+	 * Failed to merge pages because some pages are accepted and some are
+	 * pending.  Since the child page was mapped above, let vcpu run.
+	 */
+	if (ret) {
+		if (static_call(kvm_x86_unzap_private_spte)(kvm, gfn, level))
+			old_spte = SHADOW_NONPRESENT_VALUE |
+				(spte_to_pfn(old_spte) << PAGE_SHIFT) |
+				PT_PAGE_SIZE_MASK;
+		goto out;
+	}
+
+	/* Update stats manually as we don't use tdp_mmu_set_spte{, _atomic}(). */
+	kvm_update_page_stats(kvm, level - 1, -SPTE_ENT_PER_PAGE);
+	kvm_update_page_stats(kvm, level, 1);
+
+	/* Unfreeze spte. */
+	iter->old_spte = new_spte;
+	__kvm_tdp_mmu_write_spte(sptep, new_spte);
+
+	/*
+	 * Free unused child sp.  Secure-EPT page was already freed at TDX level
+	 * by kvm_x86_merge_private_spt().
+	 */
+	tdp_unaccount_mmu_page(kvm, child_sp);
+	tdp_mmu_free_sp(child_sp);
+	return 0;
+
+out:
+	iter->old_spte = old_spte;
+	__kvm_tdp_mmu_write_spte(sptep, old_spte);
+	return ret;
+}
+
+static int __tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu,
+					     struct kvm_page_fault *fault,
+					     struct tdp_iter *iter, u64 new_spte)
+{
+	/*
+	 * The private page has smaller-size pages.  For example, the child
+	 * pages was converted from shared to page, and now it can be mapped as
+	 * a large page.  Try to merge small pages into a large page.
+	 */
+	if (fault->slot &&
+	    kvm_gfn_shared_mask(vcpu->kvm) &&
+	    iter->level > PG_LEVEL_4K &&
+	    kvm_is_private_gpa(vcpu->kvm, fault->addr) &&
+	    is_shadow_present_pte(iter->old_spte) &&
+	    !is_large_pte(iter->old_spte))
+		return tdp_mmu_merge_private_spt(vcpu, fault, iter, new_spte);
+
+	return tdp_mmu_set_spte_atomic(vcpu->kvm, iter, new_spte);
+}
+
 /*
  * Installs a last-level SPTE to handle a TDP page fault.
  * (NPT/EPT violation/misconfiguration)
@@ -1246,7 +1420,7 @@ static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu,
 
 	if (new_spte == iter->old_spte)
 		ret = RET_PF_SPURIOUS;
-	else if (tdp_mmu_set_spte_atomic(vcpu->kvm, iter, new_spte))
+	else if (__tdp_mmu_map_handle_target_level(vcpu, fault, iter, new_spte))
 		return RET_PF_RETRY;
 	else if (is_shadow_present_pte(iter->old_spte) &&
 		 !is_last_spte(iter->old_spte, iter->level))
-- 
2.25.1

[PATCH v8 12/14] KVM: TDX: Implement merge pages into a large page

[isaku.yamahata]

From: Isaku Yamahata <isaku.yamahata@intel.com>

Implement merge_private_stp callback.

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
v7:
- Fix subject, x86/tdp_mmu => TDX
- comment: use ulink instead of free for clarity

v6:
- repeat TDH.MEM.PAGE.PROMOTE() on TDX_INTERRUPTED_RESTARTABLE

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/vmx/tdx.c       | 74 ++++++++++++++++++++++++++++++++++++
 arch/x86/kvm/vmx/tdx_arch.h  |  1 +
 arch/x86/kvm/vmx/tdx_errno.h |  2 +
 arch/x86/kvm/vmx/tdx_ops.h   | 11 ++++++
 4 files changed, 88 insertions(+)

diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index 88af64658a9c..5b4d94a6c6e2 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -1674,6 +1674,51 @@ static int tdx_sept_split_private_spt(struct kvm *kvm, gfn_t gfn,
 	return 0;
 }
 
+static int tdx_sept_merge_private_spt(struct kvm *kvm, gfn_t gfn,
+				      enum pg_level level, void *private_spt)
+{
+	int tdx_level = pg_level_to_tdx_sept_level(level);
+	struct kvm_tdx *kvm_tdx = to_kvm_tdx(kvm);
+	struct tdx_module_args out;
+	gpa_t gpa = gfn_to_gpa(gfn) & KVM_HPAGE_MASK(level);
+	u64 err;
+
+	/* See comment in tdx_sept_set_private_spte() */
+	do {
+		err = tdh_mem_page_promote(kvm_tdx->tdr_pa, gpa, tdx_level, &out);
+	} while (err == TDX_INTERRUPTED_RESTARTABLE);
+	if (unlikely(err == TDX_ERROR_SEPT_BUSY))
+		return -EAGAIN;
+	if (unlikely(err == (TDX_EPT_INVALID_PROMOTE_CONDITIONS |
+			     TDX_OPERAND_ID_RCX)))
+		/*
+		 * Some pages are accepted, some pending.  Need to wait for TD
+		 * to accept all pages.  Tell it the caller.
+		 */
+		return -EAGAIN;
+	if (KVM_BUG_ON(err, kvm)) {
+		pr_tdx_error(TDH_MEM_PAGE_PROMOTE, err, &out);
+		return -EIO;
+	}
+	WARN_ON_ONCE(out.rcx != __pa(private_spt));
+
+	/*
+	 * TDH.MEM.PAGE.PROMOTE unlinks the Secure-EPT page for the lower level.
+	 * Flush cache for reuse.
+	 */
+	do {
+		err = tdh_phymem_page_wbinvd(set_hkid_to_hpa(__pa(private_spt),
+							     to_kvm_tdx(kvm)->hkid));
+	} while (unlikely(err == (TDX_OPERAND_BUSY | TDX_OPERAND_ID_RCX)));
+	if (WARN_ON_ONCE(err)) {
+		pr_tdx_error(TDH_PHYMEM_PAGE_WBINVD, err, NULL);
+		return -EIO;
+	}
+
+	tdx_clear_page(__pa(private_spt), PAGE_SIZE);
+	return 0;
+}
+
 static int tdx_sept_zap_private_spte(struct kvm *kvm, gfn_t gfn,
 				      enum pg_level level)
 {
@@ -1770,6 +1815,33 @@ static void tdx_track(struct kvm *kvm)
 
 }
 
+static int tdx_sept_unzap_private_spte(struct kvm *kvm, gfn_t gfn,
+				       enum pg_level level)
+{
+	int tdx_level = pg_level_to_tdx_sept_level(level);
+	struct kvm_tdx *kvm_tdx = to_kvm_tdx(kvm);
+	gpa_t gpa = gfn_to_gpa(gfn) & KVM_HPAGE_MASK(level);
+	struct tdx_module_args out;
+	u64 err;
+
+	do {
+		err = tdh_mem_range_unblock(kvm_tdx->tdr_pa, gpa, tdx_level, &out);
+
+		/*
+		 * tdh_mem_range_block() is accompanied with tdx_track() via kvm
+		 * remote tlb flush.  Wait for the caller of
+		 * tdh_mem_range_block() to complete TDX track.
+		 */
+	} while (err == (TDX_TLB_TRACKING_NOT_DONE | TDX_OPERAND_ID_SEPT));
+	if (unlikely(err == TDX_ERROR_SEPT_BUSY))
+		return -EAGAIN;
+	if (KVM_BUG_ON(err, kvm)) {
+		pr_tdx_error(TDH_MEM_RANGE_UNBLOCK, err, &out);
+		return -EIO;
+	}
+	return 0;
+}
+
 static int tdx_sept_free_private_spt(struct kvm *kvm, gfn_t gfn,
 				     enum pg_level level, void *private_spt)
 {
@@ -3331,9 +3403,11 @@ int __init tdx_hardware_setup(struct kvm_x86_ops *x86_ops)
 	x86_ops->link_private_spt = tdx_sept_link_private_spt;
 	x86_ops->free_private_spt = tdx_sept_free_private_spt;
 	x86_ops->split_private_spt = tdx_sept_split_private_spt;
+	x86_ops->merge_private_spt = tdx_sept_merge_private_spt;
 	x86_ops->set_private_spte = tdx_sept_set_private_spte;
 	x86_ops->remove_private_spte = tdx_sept_remove_private_spte;
 	x86_ops->zap_private_spte = tdx_sept_zap_private_spte;
+	x86_ops->unzap_private_spte = tdx_sept_unzap_private_spte;
 
 	return 0;
 
diff --git a/arch/x86/kvm/vmx/tdx_arch.h b/arch/x86/kvm/vmx/tdx_arch.h
index bb324f744bbf..a320f6d45731 100644
--- a/arch/x86/kvm/vmx/tdx_arch.h
+++ b/arch/x86/kvm/vmx/tdx_arch.h
@@ -29,6 +29,7 @@
 #define TDH_MNG_KEY_FREEID		20
 #define TDH_MNG_INIT			21
 #define TDH_VP_INIT			22
+#define TDH_MEM_PAGE_PROMOTE		23
 #define TDH_MEM_SEPT_RD			25
 #define TDH_VP_RD			26
 #define TDH_MNG_KEY_RECLAIMID		27
diff --git a/arch/x86/kvm/vmx/tdx_errno.h b/arch/x86/kvm/vmx/tdx_errno.h
index 416708e6cbb7..799d7166e69a 100644
--- a/arch/x86/kvm/vmx/tdx_errno.h
+++ b/arch/x86/kvm/vmx/tdx_errno.h
@@ -21,6 +21,8 @@
 #define TDX_KEY_CONFIGURED			0x0000081500000000ULL
 #define TDX_NO_HKID_READY_TO_WBCACHE		0x0000082100000000ULL
 #define TDX_FLUSHVP_NOT_DONE			0x8000082400000000ULL
+#define TDX_TLB_TRACKING_NOT_DONE		0xC0000B0800000000ULL
+#define TDX_EPT_INVALID_PROMOTE_CONDITIONS	0xC0000B0900000000ULL
 #define TDX_EPT_ENTRY_STATE_INCORRECT		0xC0000B0D00000000ULL
 
 /*
diff --git a/arch/x86/kvm/vmx/tdx_ops.h b/arch/x86/kvm/vmx/tdx_ops.h
index d8f0d9aa7439..bf660eefa9e0 100644
--- a/arch/x86/kvm/vmx/tdx_ops.h
+++ b/arch/x86/kvm/vmx/tdx_ops.h
@@ -254,6 +254,17 @@ static inline u64 tdh_mem_page_demote(hpa_t tdr, gpa_t gpa, int level, hpa_t pag
 	return tdx_seamcall_sept(TDH_MEM_PAGE_DEMOTE, &in, out);
 }
 
+static inline u64 tdh_mem_page_promote(hpa_t tdr, gpa_t gpa, int level,
+				       struct tdx_module_args *out)
+{
+	struct tdx_module_args in = {
+		.rcx = gpa | level,
+		.rdx = tdr,
+	};
+
+	return tdx_seamcall_sept(TDH_MEM_PAGE_PROMOTE, &in, out);
+}
+
 static inline u64 tdh_mr_extend(hpa_t tdr, gpa_t gpa,
 				struct tdx_module_args *out)
 {
-- 
2.25.1

[PATCH v8 13/14] KVM: x86/mmu: Make kvm fault handler aware of large page of private memslot

[isaku.yamahata]

From: Isaku Yamahata <isaku.yamahata@intel.com>

struct kvm_page_fault.req_level is the page level which takes care of the
faulted-in page size.  For now its calculation is only for the conventional
kvm memslot by host_pfn_mapping_level() that traverses page table.

However, host_pfn_mapping_level() cannot be used for private kvm memslot
because private pages of private kvm memlost aren't mapped into user
virtual address space.  Instead, page order is given when getting pfn.
Remember it in struct kvm_page_fault and use it.

Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/mmu/mmu.c          | 27 ++++++++++++++-------------
 arch/x86/kvm/mmu/mmu_internal.h | 12 +++++++++++-
 arch/x86/kvm/mmu/tdp_mmu.c      |  2 +-
 3 files changed, 26 insertions(+), 15 deletions(-)

diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index fa7fabc410c4..3c41861b4b3d 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -3154,10 +3154,10 @@ static int host_pfn_mapping_level(struct kvm *kvm, gfn_t gfn,
 
 static int __kvm_mmu_max_mapping_level(struct kvm *kvm,
 				       const struct kvm_memory_slot *slot,
-				       gfn_t gfn, int max_level, bool is_private)
+				       gfn_t gfn, int max_level, int host_level,
+				       bool is_private)
 {
 	struct kvm_lpage_info *linfo;
-	int host_level;
 
 	max_level = min(max_level, max_huge_page_level);
 	for ( ; max_level > PG_LEVEL_4K; max_level--) {
@@ -3166,24 +3166,23 @@ static int __kvm_mmu_max_mapping_level(struct kvm *kvm,
 			break;
 	}
 
-	if (is_private)
-		return max_level;
-
 	if (max_level == PG_LEVEL_4K)
 		return PG_LEVEL_4K;
 
-	host_level = host_pfn_mapping_level(kvm, gfn, slot);
+	if (!is_private) {
+		WARN_ON_ONCE(host_level != PG_LEVEL_NONE);
+		host_level = host_pfn_mapping_level(kvm, gfn, slot);
+	}
+	WARN_ON_ONCE(host_level == PG_LEVEL_NONE);
 	return min(host_level, max_level);
 }
 
 int kvm_mmu_max_mapping_level(struct kvm *kvm,
 			      const struct kvm_memory_slot *slot, gfn_t gfn,
-			      int max_level)
+			      int max_level, bool faultin_private)
 {
-	bool is_private = kvm_slot_can_be_private(slot) &&
-			  kvm_mem_is_private(kvm, gfn);
-
-	return __kvm_mmu_max_mapping_level(kvm, slot, gfn, max_level, is_private);
+	return __kvm_mmu_max_mapping_level(kvm, slot, gfn, max_level,
+					   PG_LEVEL_NONE, faultin_private);
 }
 
 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
@@ -3208,7 +3207,8 @@ void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault
 	 */
 	fault->req_level = __kvm_mmu_max_mapping_level(vcpu->kvm, slot,
 						       fault->gfn, fault->max_level,
-						       fault->is_private);
+						       fault->host_level,
+						       kvm_is_faultin_private(fault));
 	if (fault->req_level == PG_LEVEL_4K || fault->huge_page_disallowed)
 		return;
 
@@ -4349,6 +4349,7 @@ static int kvm_faultin_pfn_private(struct kvm_vcpu *vcpu,
 	}
 
 	max_level = kvm_max_level_for_order(max_order);
+	fault->host_level = max_level;
 	r = static_call(kvm_x86_gmem_max_level)(vcpu->kvm, fault->pfn,
 						fault->gfn, fault->is_private,
 						&max_level);
@@ -6818,7 +6819,7 @@ static bool kvm_mmu_zap_collapsible_spte(struct kvm *kvm,
 		 */
 		if (sp->role.direct &&
 		    sp->role.level < kvm_mmu_max_mapping_level(kvm, slot, sp->gfn,
-							       PG_LEVEL_NUM)) {
+							       PG_LEVEL_NUM, false)) {
 			kvm_zap_one_rmap_spte(kvm, rmap_head, sptep);
 
 			if (kvm_available_flush_remote_tlbs_range())
diff --git a/arch/x86/kvm/mmu/mmu_internal.h b/arch/x86/kvm/mmu/mmu_internal.h
index 315c123affaf..9d56f9ab16f7 100644
--- a/arch/x86/kvm/mmu/mmu_internal.h
+++ b/arch/x86/kvm/mmu/mmu_internal.h
@@ -327,6 +327,9 @@ struct kvm_page_fault {
 	 * is changing its own translation in the guest page tables.
 	 */
 	bool write_fault_to_shadow_pgtable;
+
+	/* valid only for private memslot && private gfn */
+	enum pg_level host_level;
 };
 
 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
@@ -421,7 +424,7 @@ static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
 
 int kvm_mmu_max_mapping_level(struct kvm *kvm,
 			      const struct kvm_memory_slot *slot, gfn_t gfn,
-			      int max_level);
+			      int max_level, bool faultin_private);
 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
 void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level);
 
@@ -439,4 +442,11 @@ static inline bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t g
 }
 #endif
 
+static inline bool kvm_is_faultin_private(const struct kvm_page_fault *fault)
+{
+	if (IS_ENABLED(CONFIG_KVM_GENERIC_PRIVATE_MEM))
+		return fault->is_private && kvm_slot_can_be_private(fault->slot);
+	return false;
+}
+
 #endif /* __KVM_X86_MMU_INTERNAL_H */
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 556974361d36..d6ce8496803f 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -2183,7 +2183,7 @@ static void zap_collapsible_spte_range(struct kvm *kvm,
 			continue;
 
 		max_mapping_level = kvm_mmu_max_mapping_level(kvm, slot,
-							      iter.gfn, PG_LEVEL_NUM);
+							      iter.gfn, PG_LEVEL_NUM, false);
 		if (max_mapping_level < iter.level)
 			continue;
 
-- 
2.25.1

[PATCH v8 14/14] KVM: TDX: Allow 2MB large page for TD GUEST

[isaku.yamahata]

From: Xiaoyao Li <xiaoyao.li@intel.com>

Now that everything is there to support 2MB page for TD guest.  Because TDX
module TDH.MEM.PAGE.AUG supports 4KB page and 2MB page, set struct
kvm_arch.tdp_max_page_level to 2MB page level.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 arch/x86/kvm/mmu/tdp_mmu.c | 9 ++-------
 arch/x86/kvm/vmx/tdx.c     | 7 +++++--
 2 files changed, 7 insertions(+), 9 deletions(-)

diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index d6ce8496803f..8663e163427c 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -1544,14 +1544,9 @@ int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
 
 		sp->nx_huge_page_disallowed = fault->huge_page_disallowed;
 
-		if (is_shadow_present_pte(iter.old_spte)) {
-			/*
-			 * TODO: large page support.
-			 * Doesn't support large page for TDX now
-			 */
-			KVM_BUG_ON(is_private_sptep(iter.sptep), vcpu->kvm);
+		if (is_shadow_present_pte(iter.old_spte))
 			r = tdp_mmu_split_huge_page(kvm, &iter, sp, true);
-		} else
+		else
 			r = tdp_mmu_link_sp(kvm, &iter, sp, true);
 
 		/*
diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
index 5b4d94a6c6e2..1a94b1072068 100644
--- a/arch/x86/kvm/vmx/tdx.c
+++ b/arch/x86/kvm/vmx/tdx.c
@@ -3109,8 +3109,11 @@ int tdx_gmem_max_level(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn,
 	if (!is_private)
 		return 0;
 
-	/* TODO: Enable 2mb and 1gb large page support. */
-	*max_level = min(*max_level, PG_LEVEL_4K);
+	/*
+	 * TDH.MEM.PAGE.AUG supports up to 2MB page.
+	 * TODO: Enable 1gb large page support.
+	 */
+	*max_level = min(*max_level, PG_LEVEL_2M);
 	return 0;
 }
 
-- 
2.25.1

Re: [PATCH v8 00/14] KVM TDX: TDP MMU: large page support

[Yin, Fengwei]

Hi Isaku,

On 2/26/2024 4:29 PM, isaku.yamahata@intel.com wrote:
> From: Isaku Yamahata <isaku.yamahata@intel.com>
> 
> This patch series is based on "v19 KVM TDX: basic feature support".  It
> implements large page support for TDP MMU by allowing populating of the large
> page and splitting it when necessary.
To test the hugepage for TDX guest, we need to apply Qemu patch
from Xiaoyao:
https://lore.kernel.org/qemu-devel/20231115071519.2864957-4-xiaoyao.li@intel.com/

According to Xiaoyao, it's still under discussion. So he didn't
send updated version patch. For folks want to try this series,
it may be better to mention above link in this cover letter?

Test in my side showed several benchmarks got 10+% performance
gain which is really nice. So:
Tested-by: Yin Fengwei <fengwei.yin@intel.com>


Regards
Yin, Fengwei

> 
> No major changes from v7 instead of rebasing.
> 
> Thanks,
> 
> Changes from v7:
> - Rebased to v19 TDX KVM v6.8-rc5 based patch series
> 
> Changes from v6:
> - Rebased to v18 TDX KVM v6.8-rc1 based patch series
> - use struct tdx_module_args
> - minor improve on comment, commit message
> 
> Changes from v5:
> - Switched to TDX module 1.5 base.
> 
> Chnages from v4:
> - Rebased to v16 TDX KVM v6.6-rc2 base
> 
> Changes from v3:
> - Rebased to v15 TDX KVM v6.5-rc1 base
> 
> Changes from v2:
> - implemented page merging path
> - rebased to TDX KVM v11
> 
> Changes from v1:
> - implemented page merging path
> - rebased to UPM v10
> - rebased to TDX KVM v10
> - rebased to kvm.git queue + v6.1-rc8
> 
> Isaku Yamahata (4):
>    KVM: x86/tdp_mmu: Allocate private page table for large page split
>    KVM: x86/tdp_mmu: Try to merge pages into a large page
>    KVM: TDX: Implement merge pages into a large page
>    KVM: x86/mmu: Make kvm fault handler aware of large page of private
>      memslot
> 
> Sean Christopherson (1):
>    KVM: Add transparent hugepage support for dedicated guest memory
> 
> Xiaoyao Li (9):
>    KVM: TDX: Flush cache based on page size before TDX SEAMCALL
>    KVM: TDX: Pass KVM page level to tdh_mem_page_aug()
>    KVM: TDX: Pass size to reclaim_page()
>    KVM: TDX: Update tdx_sept_{set,drop}_private_spte() to support large
>      page
>    KVM: MMU: Introduce level info in PFERR code
>    KVM: TDX: Pass desired page level in err code for page fault handler
>    KVM: x86/tdp_mmu: Split the large page when zap leaf
>    KVM: x86/tdp_mmu, TDX: Split a large page when 4KB page within it
>      converted to shared
>    KVM: TDX: Allow 2MB large page for TD GUEST
> 
>   Documentation/virt/kvm/api.rst     |   7 +
>   arch/x86/include/asm/kvm-x86-ops.h |   3 +
>   arch/x86/include/asm/kvm_host.h    |  11 ++
>   arch/x86/kvm/mmu/mmu.c             |  38 ++--
>   arch/x86/kvm/mmu/mmu_internal.h    |  30 +++-
>   arch/x86/kvm/mmu/tdp_iter.c        |  37 +++-
>   arch/x86/kvm/mmu/tdp_iter.h        |   2 +
>   arch/x86/kvm/mmu/tdp_mmu.c         | 276 ++++++++++++++++++++++++++---
>   arch/x86/kvm/vmx/common.h          |   6 +-
>   arch/x86/kvm/vmx/tdx.c             | 221 +++++++++++++++++------
>   arch/x86/kvm/vmx/tdx_arch.h        |  21 +++
>   arch/x86/kvm/vmx/tdx_errno.h       |   3 +
>   arch/x86/kvm/vmx/tdx_ops.h         |  56 ++++--
>   arch/x86/kvm/vmx/vmx.c             |   2 +-
>   include/uapi/linux/kvm.h           |   2 +
>   virt/kvm/guest_memfd.c             |  73 +++++++-
>   16 files changed, 672 insertions(+), 116 deletions(-)
> 

Re: [PATCH v8 00/14] KVM TDX: TDP MMU: large page support

[Isaku Yamahata]

On Wed, Mar 27, 2024 at 08:53:50AM +0800,
"Yin, Fengwei" <fengwei.yin@intel.com> wrote:

> Hi Isaku,
> 
> On 2/26/2024 4:29 PM, isaku.yamahata@intel.com wrote:
> > From: Isaku Yamahata <isaku.yamahata@intel.com>
> > 
> > This patch series is based on "v19 KVM TDX: basic feature support".  It
> > implements large page support for TDP MMU by allowing populating of the large
> > page and splitting it when necessary.
> To test the hugepage for TDX guest, we need to apply Qemu patch
> from Xiaoyao:
> https://lore.kernel.org/qemu-devel/20231115071519.2864957-4-xiaoyao.li@intel.com/
> 
> According to Xiaoyao, it's still under discussion. So he didn't
> send updated version patch. For folks want to try this series,
> it may be better to mention above link in this cover letter?

Makes sense.  Let me include it from the next versions.


> Test in my side showed several benchmarks got 10+% performance
> gain which is really nice. So:

So nice.

> Tested-by: Yin Fengwei <fengwei.yin@intel.com>

Thanks,
-- 
Isaku Yamahata <isaku.yamahata@intel.com>