[PATCH 0/6 v4] KVM support for 1GB pages

[PATCH 0/6 v4] KVM support for 1GB pages

[Joerg Roedel]

Hi,

this is the forth version of the patches for KVM to support 1GB pages.
This code still only can make use of 1GB pages with nested paging
enabled. I will give the shadow paging code another debug round soon.
Please comment or consider to apply these patches.

Thanks,
	
	Joerg

shortlog:

Joerg Roedel (6):
      kvm/mmu: make rmap code aware of mapping levels
      kvm/mmu: rename is_largepage_backed to mapping_level
      kvm/mmu: make direct mapping paths aware of mapping levels
      kvm/mmu: make page walker aware of mapping levels
      kvm/mmu: enable gbpages by increasing nr of pagesizes
      kvm x86: report 1GB page support to userspace

diffstat:

 arch/x86/include/asm/kvm_host.h |    6 +-
 arch/x86/kvm/mmu.c              |  234 ++++++++++++++++++++++++--------------
 arch/x86/kvm/paging_tmpl.h      |   62 ++++++-----
 arch/x86/kvm/svm.c              |    6 +
 arch/x86/kvm/vmx.c              |    6 +
 arch/x86/kvm/x86.c              |    3 +-
 6 files changed, 199 insertions(+), 118 deletions(-)

[PATCH 1/6] kvm/mmu: make rmap code aware of mapping levels

[Joerg Roedel]

This patch removes the largepage parameter from the rmap_add function.
Together with rmap_remove this function now uses the role.level field to
find determine if the page is a huge page.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kvm/mmu.c |   53 +++++++++++++++++++++++++++------------------------
 1 files changed, 28 insertions(+), 25 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 231d880..0434990 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -479,19 +479,19 @@ static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn)
  * Note: gfn must be unaliased before this function get called
  */
 
-static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage)
+static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level)
 {
 	struct kvm_memory_slot *slot;
 	unsigned long idx;
 
 	slot = gfn_to_memslot(kvm, gfn);
-	if (!lpage)
+	if (likely(level == PT_PAGE_TABLE_LEVEL))
 		return &slot->rmap[gfn - slot->base_gfn];
 
-	idx = (gfn / KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL)) -
-	      (slot->base_gfn / KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL));
+	idx = (gfn / KVM_PAGES_PER_HPAGE(level)) -
+		(slot->base_gfn / KVM_PAGES_PER_HPAGE(level));
 
-	return &slot->lpage_info[0][idx].rmap_pde;
+	return &slot->lpage_info[level - 2][idx].rmap_pde;
 }
 
 /*
@@ -503,7 +503,7 @@ static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage)
  * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
  * containing more mappings.
  */
-static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage)
+static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
 {
 	struct kvm_mmu_page *sp;
 	struct kvm_rmap_desc *desc;
@@ -515,7 +515,7 @@ static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage)
 	gfn = unalias_gfn(vcpu->kvm, gfn);
 	sp = page_header(__pa(spte));
 	sp->gfns[spte - sp->spt] = gfn;
-	rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage);
+	rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level);
 	if (!*rmapp) {
 		rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
 		*rmapp = (unsigned long)spte;
@@ -582,7 +582,7 @@ static void rmap_remove(struct kvm *kvm, u64 *spte)
 		kvm_release_pfn_dirty(pfn);
 	else
 		kvm_release_pfn_clean(pfn);
-	rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], is_large_pte(*spte));
+	rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], sp->role.level);
 	if (!*rmapp) {
 		printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
 		BUG();
@@ -645,10 +645,10 @@ static int rmap_write_protect(struct kvm *kvm, u64 gfn)
 {
 	unsigned long *rmapp;
 	u64 *spte;
-	int write_protected = 0;
+	int i, write_protected = 0;
 
 	gfn = unalias_gfn(kvm, gfn);
-	rmapp = gfn_to_rmap(kvm, gfn, 0);
+	rmapp = gfn_to_rmap(kvm, gfn, PT_PAGE_TABLE_LEVEL);
 
 	spte = rmap_next(kvm, rmapp, NULL);
 	while (spte) {
@@ -670,21 +670,24 @@ static int rmap_write_protect(struct kvm *kvm, u64 gfn)
 	}
 
 	/* check for huge page mappings */
-	rmapp = gfn_to_rmap(kvm, gfn, 1);
-	spte = rmap_next(kvm, rmapp, NULL);
-	while (spte) {
-		BUG_ON(!spte);
-		BUG_ON(!(*spte & PT_PRESENT_MASK));
-		BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK));
-		pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
-		if (is_writeble_pte(*spte)) {
-			rmap_remove(kvm, spte);
-			--kvm->stat.lpages;
-			__set_spte(spte, shadow_trap_nonpresent_pte);
-			spte = NULL;
-			write_protected = 1;
+	for (i = PT_DIRECTORY_LEVEL;
+	     i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
+		rmapp = gfn_to_rmap(kvm, gfn, i);
+		spte = rmap_next(kvm, rmapp, NULL);
+		while (spte) {
+			BUG_ON(!spte);
+			BUG_ON(!(*spte & PT_PRESENT_MASK));
+			BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK));
+			pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
+			if (is_writeble_pte(*spte)) {
+				rmap_remove(kvm, spte);
+				--kvm->stat.lpages;
+				__set_spte(spte, shadow_trap_nonpresent_pte);
+				spte = NULL;
+				write_protected = 1;
+			}
+			spte = rmap_next(kvm, rmapp, spte);
 		}
-		spte = rmap_next(kvm, rmapp, spte);
 	}
 
 	return write_protected;
@@ -1793,7 +1796,7 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 
 	page_header_update_slot(vcpu->kvm, sptep, gfn);
 	if (!was_rmapped) {
-		rmap_add(vcpu, sptep, gfn, largepage);
+		rmap_add(vcpu, sptep, gfn);
 		if (!is_rmap_spte(*sptep))
 			kvm_release_pfn_clean(pfn);
 	} else {
-- 
1.6.3.3

[PATCH 2/6] kvm/mmu: rename is_largepage_backed to mapping_level

[Joerg Roedel]

With the new name and the corresponding backend changes this function
can now support multiple hugepage sizes.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kvm/mmu.c         |  100 +++++++++++++++++++++++++++++--------------
 arch/x86/kvm/paging_tmpl.h |    4 +-
 2 files changed, 69 insertions(+), 35 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 0434990..7de9f41 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -393,37 +393,52 @@ static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
  * Return the pointer to the largepage write count for a given
  * gfn, handling slots that are not large page aligned.
  */
-static int *slot_largepage_idx(gfn_t gfn, struct kvm_memory_slot *slot)
+static int *slot_largepage_idx(gfn_t gfn,
+			       struct kvm_memory_slot *slot,
+			       int level)
 {
 	unsigned long idx;
 
-	idx = (gfn / KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL)) -
-	      (slot->base_gfn / KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL));
-	return &slot->lpage_info[0][idx].write_count;
+	idx = (gfn / KVM_PAGES_PER_HPAGE(level)) -
+	      (slot->base_gfn / KVM_PAGES_PER_HPAGE(level));
+	return &slot->lpage_info[level - 2][idx].write_count;
 }
 
 static void account_shadowed(struct kvm *kvm, gfn_t gfn)
 {
+	struct kvm_memory_slot *slot;
 	int *write_count;
+	int i;
 
 	gfn = unalias_gfn(kvm, gfn);
-	write_count = slot_largepage_idx(gfn,
-					 gfn_to_memslot_unaliased(kvm, gfn));
-	*write_count += 1;
+
+	slot = gfn_to_memslot_unaliased(kvm, gfn);
+	for (i = PT_DIRECTORY_LEVEL;
+	     i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
+		write_count   = slot_largepage_idx(gfn, slot, i);
+		*write_count += 1;
+	}
 }
 
 static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn)
 {
+	struct kvm_memory_slot *slot;
 	int *write_count;
+	int i;
 
 	gfn = unalias_gfn(kvm, gfn);
-	write_count = slot_largepage_idx(gfn,
-					 gfn_to_memslot_unaliased(kvm, gfn));
-	*write_count -= 1;
-	WARN_ON(*write_count < 0);
+	for (i = PT_DIRECTORY_LEVEL;
+	     i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
+		slot          = gfn_to_memslot_unaliased(kvm, gfn);
+		write_count   = slot_largepage_idx(gfn, slot, i);
+		*write_count -= 1;
+		WARN_ON(*write_count < 0);
+	}
 }
 
-static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn)
+static int has_wrprotected_page(struct kvm *kvm,
+				gfn_t gfn,
+				int level)
 {
 	struct kvm_memory_slot *slot;
 	int *largepage_idx;
@@ -431,47 +446,67 @@ static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn)
 	gfn = unalias_gfn(kvm, gfn);
 	slot = gfn_to_memslot_unaliased(kvm, gfn);
 	if (slot) {
-		largepage_idx = slot_largepage_idx(gfn, slot);
+		largepage_idx = slot_largepage_idx(gfn, slot, level);
 		return *largepage_idx;
 	}
 
 	return 1;
 }
 
-static int host_largepage_backed(struct kvm *kvm, gfn_t gfn)
+static int host_mapping_level(struct kvm *kvm, gfn_t gfn)
 {
+	unsigned long page_size = PAGE_SIZE;
 	struct vm_area_struct *vma;
 	unsigned long addr;
-	int ret = 0;
+	int i, ret = 0;
 
 	addr = gfn_to_hva(kvm, gfn);
 	if (kvm_is_error_hva(addr))
-		return ret;
+		return page_size;
 
 	down_read(&current->mm->mmap_sem);
 	vma = find_vma(current->mm, addr);
-	if (vma && is_vm_hugetlb_page(vma))
-		ret = 1;
+	if (!vma)
+		goto out;
+
+	page_size = vma_kernel_pagesize(vma);
+
+out:
 	up_read(&current->mm->mmap_sem);
 
+	for (i = PT_PAGE_TABLE_LEVEL;
+	     i < (PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES); ++i) {
+		if (page_size >= KVM_HPAGE_SIZE(i))
+			ret = i;
+		else
+			break;
+	}
+
 	return ret;
 }
 
-static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn)
+static int mapping_level(struct kvm_vcpu *vcpu, gfn_t large_gfn)
 {
 	struct kvm_memory_slot *slot;
-
-	if (has_wrprotected_page(vcpu->kvm, large_gfn))
-		return 0;
-
-	if (!host_largepage_backed(vcpu->kvm, large_gfn))
-		return 0;
+	int host_level;
+	int level = PT_PAGE_TABLE_LEVEL;
 
 	slot = gfn_to_memslot(vcpu->kvm, large_gfn);
 	if (slot && slot->dirty_bitmap)
-		return 0;
+		return PT_PAGE_TABLE_LEVEL;
 
-	return 1;
+	host_level = host_mapping_level(vcpu->kvm, large_gfn);
+
+	if (host_level == PT_PAGE_TABLE_LEVEL)
+		return host_level;
+
+	for (level = PT_DIRECTORY_LEVEL; level <= host_level; ++level) {
+
+		if (has_wrprotected_page(vcpu->kvm, large_gfn, level))
+			break;
+	}
+
+	return level - 1;
 }
 
 /*
@@ -1712,7 +1747,7 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 	if ((pte_access & ACC_WRITE_MASK)
 	    || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
 
-		if (largepage && has_wrprotected_page(vcpu->kvm, gfn)) {
+		if (largepage && has_wrprotected_page(vcpu->kvm, gfn, 1)) {
 			ret = 1;
 			spte = shadow_trap_nonpresent_pte;
 			goto set_pte;
@@ -1860,8 +1895,7 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 	pfn_t pfn;
 	unsigned long mmu_seq;
 
-	if (is_largepage_backed(vcpu, gfn &
-			~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1))) {
+	if (mapping_level(vcpu, gfn) == PT_DIRECTORY_LEVEL) {
 		gfn &= ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
 		largepage = 1;
 	}
@@ -2067,8 +2101,7 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 	if (r)
 		return r;
 
-	if (is_largepage_backed(vcpu, gfn &
-			~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1))) {
+	if (mapping_level(vcpu, gfn) == PT_DIRECTORY_LEVEL) {
 		gfn &= ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
 		largepage = 1;
 	}
@@ -2470,7 +2503,8 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 		return;
 	gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
 
-	if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) {
+	if (is_large_pte(gpte) &&
+	    (mapping_level(vcpu, gfn) == PT_DIRECTORY_LEVEL)) {
 		gfn &= ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
 		vcpu->arch.update_pte.largepage = 1;
 	}
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index 36ac6d7..44f0346 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -407,8 +407,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	if (walker.level == PT_DIRECTORY_LEVEL) {
 		gfn_t large_gfn;
 		large_gfn = walker.gfn &
-			~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
-		if (is_largepage_backed(vcpu, large_gfn)) {
+			    ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
+		if (mapping_level(vcpu, large_gfn) == PT_DIRECTORY_LEVEL) {
 			walker.gfn = large_gfn;
 			largepage = 1;
 		}
-- 
1.6.3.3

[PATCH 3/6] kvm/mmu: make direct mapping paths aware of mapping levels

[Joerg Roedel]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/kvm_host.h |    2 +-
 arch/x86/kvm/mmu.c              |   74 ++++++++++++++++++++++----------------
 arch/x86/kvm/paging_tmpl.h      |    6 ++--
 3 files changed, 47 insertions(+), 35 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 30b625d..1fa1ff0 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -315,7 +315,7 @@ struct kvm_vcpu_arch {
 	struct {
 		gfn_t gfn;	/* presumed gfn during guest pte update */
 		pfn_t pfn;	/* pfn corresponding to that gfn */
-		int largepage;
+		int level;
 		unsigned long mmu_seq;
 	} update_pte;
 
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 7de9f41..d42185a 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -257,7 +257,7 @@ static int is_last_spte(u64 pte, int level)
 {
 	if (level == PT_PAGE_TABLE_LEVEL)
 		return 1;
-	if (level == PT_DIRECTORY_LEVEL && is_large_pte(pte))
+	if (is_large_pte(pte))
 		return 1;
 	return 0;
 }
@@ -746,7 +746,7 @@ static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
 			  int (*handler)(struct kvm *kvm, unsigned long *rmapp))
 {
-	int i;
+	int i, j;
 	int retval = 0;
 
 	/*
@@ -765,11 +765,15 @@ static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
 		end = start + (memslot->npages << PAGE_SHIFT);
 		if (hva >= start && hva < end) {
 			gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT;
-			int idx = gfn_offset /
-			          KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL);
+
 			retval |= handler(kvm, &memslot->rmap[gfn_offset]);
-			retval |= handler(kvm,
-					&memslot->lpage_info[0][idx].rmap_pde);
+
+			for (j = 0; j < KVM_NR_PAGE_SIZES - 1; ++j) {
+				int idx = gfn_offset;
+				idx /= KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL + j);
+				retval |= handler(kvm,
+					&memslot->lpage_info[j][idx].rmap_pde);
+			}
 		}
 	}
 
@@ -1713,7 +1717,7 @@ static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
 
 static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 		    unsigned pte_access, int user_fault,
-		    int write_fault, int dirty, int largepage,
+		    int write_fault, int dirty, int level,
 		    gfn_t gfn, pfn_t pfn, bool speculative,
 		    bool can_unsync)
 {
@@ -1736,7 +1740,7 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 		spte |= shadow_nx_mask;
 	if (pte_access & ACC_USER_MASK)
 		spte |= shadow_user_mask;
-	if (largepage)
+	if (level > PT_PAGE_TABLE_LEVEL)
 		spte |= PT_PAGE_SIZE_MASK;
 	if (tdp_enabled)
 		spte |= kvm_x86_ops->get_mt_mask(vcpu, gfn,
@@ -1747,7 +1751,8 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 	if ((pte_access & ACC_WRITE_MASK)
 	    || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
 
-		if (largepage && has_wrprotected_page(vcpu->kvm, gfn, 1)) {
+		if (level > PT_PAGE_TABLE_LEVEL &&
+		    has_wrprotected_page(vcpu->kvm, gfn, level)) {
 			ret = 1;
 			spte = shadow_trap_nonpresent_pte;
 			goto set_pte;
@@ -1785,7 +1790,7 @@ set_pte:
 static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 			 unsigned pt_access, unsigned pte_access,
 			 int user_fault, int write_fault, int dirty,
-			 int *ptwrite, int largepage, gfn_t gfn,
+			 int *ptwrite, int level, gfn_t gfn,
 			 pfn_t pfn, bool speculative)
 {
 	int was_rmapped = 0;
@@ -1801,7 +1806,8 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 		 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
 		 * the parent of the now unreachable PTE.
 		 */
-		if (largepage && !is_large_pte(*sptep)) {
+		if (level > PT_PAGE_TABLE_LEVEL &&
+		    !is_large_pte(*sptep)) {
 			struct kvm_mmu_page *child;
 			u64 pte = *sptep;
 
@@ -1814,8 +1820,9 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 		} else
 			was_rmapped = 1;
 	}
+
 	if (set_spte(vcpu, sptep, pte_access, user_fault, write_fault,
-		      dirty, largepage, gfn, pfn, speculative, true)) {
+		      dirty, level, gfn, pfn, speculative, true)) {
 		if (write_fault)
 			*ptwrite = 1;
 		kvm_x86_ops->tlb_flush(vcpu);
@@ -1851,7 +1858,7 @@ static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
 }
 
 static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
-			int largepage, gfn_t gfn, pfn_t pfn)
+			int level, gfn_t gfn, pfn_t pfn)
 {
 	struct kvm_shadow_walk_iterator iterator;
 	struct kvm_mmu_page *sp;
@@ -1859,11 +1866,10 @@ static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
 	gfn_t pseudo_gfn;
 
 	for_each_shadow_entry(vcpu, (u64)gfn << PAGE_SHIFT, iterator) {
-		if (iterator.level == PT_PAGE_TABLE_LEVEL
-		    || (largepage && iterator.level == PT_DIRECTORY_LEVEL)) {
+		if (iterator.level == level) {
 			mmu_set_spte(vcpu, iterator.sptep, ACC_ALL, ACC_ALL,
 				     0, write, 1, &pt_write,
-				     largepage, gfn, pfn, false);
+				     level, gfn, pfn, false);
 			++vcpu->stat.pf_fixed;
 			break;
 		}
@@ -1891,14 +1897,20 @@ static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 {
 	int r;
-	int largepage = 0;
+	int level;
 	pfn_t pfn;
 	unsigned long mmu_seq;
 
-	if (mapping_level(vcpu, gfn) == PT_DIRECTORY_LEVEL) {
-		gfn &= ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
-		largepage = 1;
-	}
+	level = mapping_level(vcpu, gfn);
+
+	/*
+	 * This path builds a PAE pagetable - so we can map 2mb pages at
+	 * maximum. Therefore check if the level is larger than that.
+	 */
+	if (level > PT_DIRECTORY_LEVEL)
+		level = PT_DIRECTORY_LEVEL;
+
+	gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1);
 
 	mmu_seq = vcpu->kvm->mmu_notifier_seq;
 	smp_rmb();
@@ -1914,7 +1926,7 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 	if (mmu_notifier_retry(vcpu, mmu_seq))
 		goto out_unlock;
 	kvm_mmu_free_some_pages(vcpu);
-	r = __direct_map(vcpu, v, write, largepage, gfn, pfn);
+	r = __direct_map(vcpu, v, write, level, gfn, pfn);
 	spin_unlock(&vcpu->kvm->mmu_lock);
 
 
@@ -2090,7 +2102,7 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 {
 	pfn_t pfn;
 	int r;
-	int largepage = 0;
+	int level;
 	gfn_t gfn = gpa >> PAGE_SHIFT;
 	unsigned long mmu_seq;
 
@@ -2101,10 +2113,10 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 	if (r)
 		return r;
 
-	if (mapping_level(vcpu, gfn) == PT_DIRECTORY_LEVEL) {
-		gfn &= ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
-		largepage = 1;
-	}
+	level = mapping_level(vcpu, gfn);
+
+	gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1);
+
 	mmu_seq = vcpu->kvm->mmu_notifier_seq;
 	smp_rmb();
 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
@@ -2117,7 +2129,7 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
 		goto out_unlock;
 	kvm_mmu_free_some_pages(vcpu);
 	r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
-			 largepage, gfn, pfn);
+			 level, gfn, pfn);
 	spin_unlock(&vcpu->kvm->mmu_lock);
 
 	return r;
@@ -2424,7 +2436,7 @@ static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
 				  const void *new)
 {
 	if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
-		if (!vcpu->arch.update_pte.largepage ||
+		if (vcpu->arch.update_pte.level == PT_PAGE_TABLE_LEVEL ||
 		    sp->role.glevels == PT32_ROOT_LEVEL) {
 			++vcpu->kvm->stat.mmu_pde_zapped;
 			return;
@@ -2474,7 +2486,7 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 	u64 gpte = 0;
 	pfn_t pfn;
 
-	vcpu->arch.update_pte.largepage = 0;
+	vcpu->arch.update_pte.level = PT_PAGE_TABLE_LEVEL;
 
 	if (bytes != 4 && bytes != 8)
 		return;
@@ -2506,7 +2518,7 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 	if (is_large_pte(gpte) &&
 	    (mapping_level(vcpu, gfn) == PT_DIRECTORY_LEVEL)) {
 		gfn &= ~(KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL) - 1);
-		vcpu->arch.update_pte.largepage = 1;
+		vcpu->arch.update_pte.level = PT_DIRECTORY_LEVEL;
 	}
 	vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq;
 	smp_rmb();
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index 44f0346..b167f0d 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -253,7 +253,7 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
 	pt_element_t gpte;
 	unsigned pte_access;
 	pfn_t pfn;
-	int largepage = vcpu->arch.update_pte.largepage;
+	int level = vcpu->arch.update_pte.level;
 
 	gpte = *(const pt_element_t *)pte;
 	if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
@@ -272,7 +272,7 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
 		return;
 	kvm_get_pfn(pfn);
 	mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
-		     gpte & PT_DIRTY_MASK, NULL, largepage,
+		     gpte & PT_DIRTY_MASK, NULL, level,
 		     gpte_to_gfn(gpte), pfn, true);
 }
 
@@ -306,7 +306,7 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 				     gw->pte_access & access,
 				     user_fault, write_fault,
 				     gw->ptes[gw->level-1] & PT_DIRTY_MASK,
-				     ptwrite, largepage,
+				     ptwrite, level,
 				     gw->gfn, pfn, false);
 			break;
 		}
-- 
1.6.3.3

[PATCH 4/6] kvm/mmu: make page walker aware of mapping levels

[Joerg Roedel]

The page walker may be used with nested paging too when accessing mmio
areas.  Make it support the additional page-level too.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kvm/mmu.c         |   13 +++++++++++
 arch/x86/kvm/paging_tmpl.h |   52 +++++++++++++++++++++++--------------------
 2 files changed, 41 insertions(+), 24 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index d42185a..ddee679 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -108,6 +108,9 @@ module_param(oos_shadow, bool, 0644);
 
 #define PT32_LEVEL_MASK(level) \
 		(((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
+#define PT32_LVL_OFFSET_MASK(level) \
+	(PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
+						* PT32_LEVEL_BITS))) - 1))
 
 #define PT32_INDEX(address, level)\
 	(((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
@@ -116,10 +119,19 @@ module_param(oos_shadow, bool, 0644);
 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
 #define PT64_DIR_BASE_ADDR_MASK \
 	(PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
+#define PT64_LVL_ADDR_MASK(level) \
+	(PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
+						* PT64_LEVEL_BITS))) - 1))
+#define PT64_LVL_OFFSET_MASK(level) \
+	(PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
+						* PT64_LEVEL_BITS))) - 1))
 
 #define PT32_BASE_ADDR_MASK PAGE_MASK
 #define PT32_DIR_BASE_ADDR_MASK \
 	(PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
+#define PT32_LVL_ADDR_MASK(level) \
+	(PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
+					    * PT32_LEVEL_BITS))) - 1))
 
 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
 			| PT64_NX_MASK)
@@ -130,6 +142,7 @@ module_param(oos_shadow, bool, 0644);
 #define PFERR_RSVD_MASK (1U << 3)
 #define PFERR_FETCH_MASK (1U << 4)
 
+#define PT_PDPE_LEVEL 3
 #define PT_DIRECTORY_LEVEL 2
 #define PT_PAGE_TABLE_LEVEL 1
 
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index b167f0d..578276e 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -27,7 +27,8 @@
 	#define guest_walker guest_walker64
 	#define FNAME(name) paging##64_##name
 	#define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
-	#define PT_DIR_BASE_ADDR_MASK PT64_DIR_BASE_ADDR_MASK
+	#define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
+	#define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
 	#define PT_INDEX(addr, level) PT64_INDEX(addr, level)
 	#define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level)
 	#define PT_LEVEL_BITS PT64_LEVEL_BITS
@@ -43,7 +44,8 @@
 	#define guest_walker guest_walker32
 	#define FNAME(name) paging##32_##name
 	#define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
-	#define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK
+	#define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl)
+	#define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl)
 	#define PT_INDEX(addr, level) PT32_INDEX(addr, level)
 	#define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
 	#define PT_LEVEL_BITS PT32_LEVEL_BITS
@@ -53,8 +55,8 @@
 	#error Invalid PTTYPE value
 #endif
 
-#define gpte_to_gfn FNAME(gpte_to_gfn)
-#define gpte_to_gfn_pde FNAME(gpte_to_gfn_pde)
+#define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
+#define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL)
 
 /*
  * The guest_walker structure emulates the behavior of the hardware page
@@ -71,14 +73,9 @@ struct guest_walker {
 	u32 error_code;
 };
 
-static gfn_t gpte_to_gfn(pt_element_t gpte)
+static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
 {
-	return (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT;
-}
-
-static gfn_t gpte_to_gfn_pde(pt_element_t gpte)
-{
-	return (gpte & PT_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT;
+	return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
 }
 
 static bool FNAME(cmpxchg_gpte)(struct kvm *kvm,
@@ -189,18 +186,24 @@ walk:
 
 		walker->ptes[walker->level - 1] = pte;
 
-		if (walker->level == PT_PAGE_TABLE_LEVEL) {
-			walker->gfn = gpte_to_gfn(pte);
-			break;
-		}
-
-		if (walker->level == PT_DIRECTORY_LEVEL
-		    && (pte & PT_PAGE_SIZE_MASK)
-		    && (PTTYPE == 64 || is_pse(vcpu))) {
-			walker->gfn = gpte_to_gfn_pde(pte);
-			walker->gfn += PT_INDEX(addr, PT_PAGE_TABLE_LEVEL);
-			if (PTTYPE == 32 && is_cpuid_PSE36())
+		if ((walker->level == PT_PAGE_TABLE_LEVEL) ||
+		    ((walker->level == PT_DIRECTORY_LEVEL) &&
+				(pte & PT_PAGE_SIZE_MASK)  &&
+				(PTTYPE == 64 || is_pse(vcpu))) ||
+		    ((walker->level == PT_PDPE_LEVEL) &&
+				(pte & PT_PAGE_SIZE_MASK)  &&
+				is_long_mode(vcpu))) {
+			int lvl = walker->level;
+
+			walker->gfn = gpte_to_gfn_lvl(pte, lvl);
+			walker->gfn += (addr & PT_LVL_OFFSET_MASK(lvl))
+					>> PAGE_SHIFT;
+
+			if (PTTYPE == 32 &&
+			    walker->level == PT_DIRECTORY_LEVEL &&
+			    is_cpuid_PSE36())
 				walker->gfn += pse36_gfn_delta(pte);
+
 			break;
 		}
 
@@ -609,9 +612,10 @@ static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
 #undef PT_BASE_ADDR_MASK
 #undef PT_INDEX
 #undef PT_LEVEL_MASK
-#undef PT_DIR_BASE_ADDR_MASK
+#undef PT_LVL_ADDR_MASK
+#undef PT_LVL_OFFSET_MASK
 #undef PT_LEVEL_BITS
 #undef PT_MAX_FULL_LEVELS
 #undef gpte_to_gfn
-#undef gpte_to_gfn_pde
+#undef gpte_to_gfn_lvl
 #undef CMPXCHG
-- 
1.6.3.3

[PATCH 5/6] kvm/mmu: enable gbpages by increasing nr of pagesizes

[Joerg Roedel]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/kvm_host.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 1fa1ff0..be2315b 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -55,7 +55,7 @@
 #define UNMAPPED_GVA (~(gpa_t)0)
 
 /* KVM Hugepage definitions for x86 */
-#define KVM_NR_PAGE_SIZES	2
+#define KVM_NR_PAGE_SIZES	3
 #define KVM_HPAGE_SHIFT(x)	(PAGE_SHIFT + (((x) - 1) * 9))
 #define KVM_HPAGE_SIZE(x)	(1UL << KVM_HPAGE_SHIFT(x))
 #define KVM_HPAGE_MASK(x)	(~(KVM_HPAGE_SIZE(x) - 1))
-- 
1.6.3.3

[PATCH 6/6] kvm x86: report 1GB page support to userspace

[Joerg Roedel]

If userspace knows that the kernel part supports 1GB pages it can enable
the corresponding cpuid bit so that guests actually use GB pages.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/kvm_host.h |    2 ++
 arch/x86/kvm/svm.c              |    6 ++++++
 arch/x86/kvm/vmx.c              |    6 ++++++
 arch/x86/kvm/x86.c              |    3 ++-
 4 files changed, 16 insertions(+), 1 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index be2315b..f1f9149 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -528,6 +528,8 @@ struct kvm_x86_ops {
 	int (*set_tss_addr)(struct kvm *kvm, unsigned int addr);
 	int (*get_tdp_level)(void);
 	u64 (*get_mt_mask)(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio);
+	bool (*gb_page_enable)(void);
+
 	const struct trace_print_flags *exit_reasons_str;
 };
 
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index fc14bdf..b0fc342 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -2755,6 +2755,11 @@ static const struct trace_print_flags svm_exit_reasons_str[] = {
 	{ -1, NULL }
 };
 
+static bool svm_gb_page_enable(void)
+{
+	return npt_enabled;
+}
+
 static struct kvm_x86_ops svm_x86_ops = {
 	.cpu_has_kvm_support = has_svm,
 	.disabled_by_bios = is_disabled,
@@ -2818,6 +2823,7 @@ static struct kvm_x86_ops svm_x86_ops = {
 	.get_mt_mask = svm_get_mt_mask,
 
 	.exit_reasons_str = svm_exit_reasons_str,
+	.gb_page_enable = svm_gb_page_enable,
 };
 
 static int __init svm_init(void)
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 3a75db3..02529fb 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -3903,6 +3903,11 @@ static const struct trace_print_flags vmx_exit_reasons_str[] = {
 	{ -1, NULL }
 };
 
+static bool vmx_gb_page_enable(void)
+{
+	return false;
+}
+
 static struct kvm_x86_ops vmx_x86_ops = {
 	.cpu_has_kvm_support = cpu_has_kvm_support,
 	.disabled_by_bios = vmx_disabled_by_bios,
@@ -3964,6 +3969,7 @@ static struct kvm_x86_ops vmx_x86_ops = {
 	.get_mt_mask = vmx_get_mt_mask,
 
 	.exit_reasons_str = vmx_exit_reasons_str,
+	.gb_page_enable = vmx_gb_page_enable,
 };
 
 static int __init vmx_init(void)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 95fa45c..ae9017b 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1450,6 +1450,7 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 			 u32 index, int *nent, int maxnent)
 {
 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
+	unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
 #ifdef CONFIG_X86_64
 	unsigned f_lm = F(LM);
 #else
@@ -1474,7 +1475,7 @@ static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
-		F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
+		F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 	/* cpuid 1.ecx */
 	const u32 kvm_supported_word4_x86_features =
-- 
1.6.3.3

Re: [PATCH 0/6 v4] KVM support for 1GB pages

[Avi Kivity]

On 07/07/2009 07:41 PM, Joerg Roedel wrote:
> Hi,
>
> this is the forth version of the patches for KVM to support 1GB pages.
> This code still only can make use of 1GB pages with nested paging
> enabled. I will give the shadow paging code another debug round soon.
> Please comment or consider to apply these patches.
>
>    

I would really like to have shadow 1GB work, or at least gain some 
understanding of why it doesn't.

Can you detail the changes compared to the previous round?

-- 
error compiling committee.c: too many arguments to function

Re: [PATCH 0/6 v4] KVM support for 1GB pages

[Joerg Roedel]

On Tue, Jul 07, 2009 at 07:47:14PM +0300, Avi Kivity wrote:
> On 07/07/2009 07:41 PM, Joerg Roedel wrote:
> >Hi,
> >
> >this is the forth version of the patches for KVM to support 1GB pages.
> >This code still only can make use of 1GB pages with nested paging
> >enabled. I will give the shadow paging code another debug round soon.
> >Please comment or consider to apply these patches.
> >
> 
> I would really like to have shadow 1GB work, or at least gain some
> understanding of why it doesn't.
> 
> Can you detail the changes compared to the previous round?

Until now I addressed the review comments from the previous round. The
biggest change is patch #4 which was completly rewritten.
I understand that it would be nicer to have shadow paging working too.
But since most users will use this feature only on processors with
nested paging (only processors with nested paging support gbpages) I
don't think it should be a show-stopper. YMMV.
Anyway, I will further try to get shadow paging support running and
stable.

Joerg

-- 
           | Advanced Micro Devices GmbH
 Operating | Karl-Hammerschmidt-Str. 34, 85609 Dornach bei München
 System    | 
 Research  | Geschäftsführer: Thomas M. McCoy, Giuliano Meroni
 Center    | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
           | Registergericht München, HRB Nr. 43632

Re: [PATCH 0/6 v4] KVM support for 1GB pages

[Avi Kivity]

On 07/07/2009 07:52 PM, Joerg Roedel wrote:
>> I would really like to have shadow 1GB work, or at least gain some
>> understanding of why it doesn't.
>>
>> Can you detail the changes compared to the previous round?
>>      
>
> Until now I addressed the review comments from the previous round. The
> biggest change is patch #4 which was completly rewritten.
> I understand that it would be nicer to have shadow paging working too.
> But since most users will use this feature only on processors with
> nested paging (only processors with nested paging support gbpages) I
> don't think it should be a show-stopper. YMMV.
>    

Right.  I don't think 1GB+shadow is useful in its own right.  It's not a 
barrier for merging (but still very desirable).  I do want more 
information on how it fails though.

> Anyway, I will further try to get shadow paging support running and
> stable.
>    

Great, thanks.

-- 
error compiling committee.c: too many arguments to function