Hypercall based pte updates are faster than faults, and also allow use
of the lazy MMU mode to batch operations.
Signed-off-by: Marcelo Tosatti <[EMAIL PROTECTED]>
Index: linux-2.6-x86-kvm/arch/x86/kernel/kvm.c
===================================================================
--- linux-2.6-x86-kvm.orig/arch/x86/kernel/kvm.c
+++ linux-2.6-x86-kvm/arch/x86/kernel/kvm.c
@@ -33,6 +33,104 @@ static void kvm_io_delay(void)
{
}
+static void kvm_mmu_write(void *dest, const void *src, size_t size)
+{
+ const uint8_t *p = src;
+ unsigned long a0 = *(unsigned long *)p;
+ unsigned long a1 = 0;
+
+ size >>= 2;
+#ifdef CONFIG_X86_32
+ if (size == 2)
+ a1 = *(u32 *)&p[4];
+#endif
+ kvm_hypercall4(KVM_HYPERCALL_MMU_WRITE, (unsigned long)dest, size, a0,
+ a1);
+}
+
+/*
+ * We only need to hook operations that are MMU writes. We hook these so that
+ * we can use lazy MMU mode to batch these operations. We could probably
+ * improve the performance of the host code if we used some of the information
+ * here to simplify processing of batched writes.
+ */
+static void kvm_set_pte(pte_t *ptep, pte_t pte)
+{
+ kvm_mmu_write(ptep, &pte, sizeof(pte));
+}
+
+static void kvm_set_pte_at(struct mm_struct *mm, unsigned long addr,
+ pte_t *ptep, pte_t pte)
+{
+ kvm_mmu_write(ptep, &pte, sizeof(pte));
+}
+
+static void kvm_set_pmd(pmd_t *pmdp, pmd_t pmd)
+{
+ kvm_mmu_write(pmdp, &pmd, sizeof(pmd));
+}
+
+#if PAGETABLE_LEVELS >= 3
+#ifdef CONFIG_X86_PAE
+static void kvm_set_pte_atomic(pte_t *ptep, pte_t pte)
+{
+ kvm_mmu_write(ptep, &pte, sizeof(pte));
+}
+
+static void kvm_set_pte_present(struct mm_struct *mm, unsigned long addr,
+ pte_t *ptep, pte_t pte)
+{
+ kvm_mmu_write(ptep, &pte, sizeof(pte));
+}
+
+static void kvm_pte_clear(struct mm_struct *mm,
+ unsigned long addr, pte_t *ptep)
+{
+ pte_t pte = __pte(0);
+ kvm_mmu_write(ptep, &pte, sizeof(pte));
+}
+
+static void kvm_pmd_clear(pmd_t *pmdp)
+{
+ pmd_t pmd = __pmd(0);
+ kvm_mmu_write(pmdp, &pmd, sizeof(pmd));
+}
+#endif
+
+static void kvm_set_pgd(pgd_t *pgdp, pgd_t pgd)
+{
+ kvm_mmu_write(pgdp, &pgd, sizeof(pgd));
+}
+
+static void kvm_pgd_clear(pgd_t *pgdp)
+{
+ pgd_t pgd = __pgd(0);
+ kvm_mmu_write(pgdp, &pgd, sizeof(pgd));
+}
+
+static void kvm_set_pud(pud_t *pudp, pud_t pud)
+{
+ kvm_mmu_write(pudp, &pud, sizeof(pud));
+}
+#if PAGETABLE_LEVELS == 4
+static void kvm_pud_clear(pud_t *pudp)
+{
+ pud_t pud = __pud(0);
+ kvm_mmu_write(pudp, &pud, sizeof(pud));
+}
+#endif
+#endif /* PAGETABLE_LEVELS >= 3 */
+
+static void kvm_flush_tlb(void)
+{
+ kvm_hypercall0(KVM_HYPERCALL_FLUSH_TLB);
+}
+
+static void kvm_release_pt(u32 pfn)
+{
+ kvm_hypercall1(KVM_HYPERCALL_RELEASE_PT, pfn << PAGE_SHIFT);
+}
+
static void paravirt_ops_setup(void)
{
pv_info.name = "KVM";
@@ -41,6 +139,28 @@ static void paravirt_ops_setup(void)
if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
pv_cpu_ops.io_delay = kvm_io_delay;
+ if (kvm_para_has_feature(KVM_FEATURE_MMU_WRITE)) {
+ pv_mmu_ops.set_pte = kvm_set_pte;
+ pv_mmu_ops.set_pte_at = kvm_set_pte_at;
+ pv_mmu_ops.set_pmd = kvm_set_pmd;
+#if PAGETABLE_LEVELS >= 3
+#ifdef CONFIG_X86_PAE
+ pv_mmu_ops.set_pte_atomic = kvm_set_pte_atomic;
+ pv_mmu_ops.set_pte_present = kvm_set_pte_present;
+ pv_mmu_ops.pte_clear = kvm_pte_clear;
+ pv_mmu_ops.pmd_clear = kvm_pmd_clear;
+#endif
+ pv_mmu_ops.set_pud = kvm_set_pud;
+ pv_mmu_ops.set_pgd = kvm_set_pgd;
+ pv_mmu_ops.pgd_clear = kvm_pgd_clear;
+#if PAGETABLE_LEVELS == 4
+ pv_mmu_ops.pud_clear = kvm_pud_clear;
+#endif
+#endif
+ pv_mmu_ops.flush_tlb_user = kvm_flush_tlb;
+ pv_mmu_ops.release_pt = kvm_release_pt;
+ pv_mmu_ops.release_pd = kvm_release_pt;
+ }
}
void __init kvm_guest_init(void)
Index: linux-2.6-x86-kvm/arch/x86/kvm/mmu.c
===================================================================
--- linux-2.6-x86-kvm.orig/arch/x86/kvm/mmu.c
+++ linux-2.6-x86-kvm/arch/x86/kvm/mmu.c
@@ -287,7 +287,7 @@ static void mmu_free_memory_cache_page(s
free_page((unsigned long)mc->objects[--mc->nobjs]);
}
-static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
+int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
{
int r;
@@ -856,7 +856,7 @@ static int kvm_mmu_unprotect_page(struct
return r;
}
-static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
+void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
{
struct kvm_mmu_page *sp;
Index: linux-2.6-x86-kvm/arch/x86/kvm/mmu.h
===================================================================
--- linux-2.6-x86-kvm.orig/arch/x86/kvm/mmu.h
+++ linux-2.6-x86-kvm/arch/x86/kvm/mmu.h
@@ -41,4 +41,7 @@ static inline int is_paging(struct kvm_v
return vcpu->arch.cr0 & X86_CR0_PG;
}
+void mmu_unshadow(struct kvm *kvm, gfn_t gfn);
+int mmu_topup_memory_caches(struct kvm_vcpu *vcpu);
+
#endif
Index: linux-2.6-x86-kvm/arch/x86/kvm/x86.c
===================================================================
--- linux-2.6-x86-kvm.orig/arch/x86/kvm/x86.c
+++ linux-2.6-x86-kvm/arch/x86/kvm/x86.c
@@ -2250,6 +2250,52 @@ int kvm_emulate_halt(struct kvm_vcpu *vc
}
EXPORT_SYMBOL_GPL(kvm_emulate_halt);
+static int kvm_hypercall_mmu_write(struct kvm_vcpu *vcpu, gva_t addr,
+ unsigned long size, unsigned long a0,
+ unsigned long a1)
+{
+ gpa_t gpa;
+ u64 value;
+
+ if (mmu_topup_memory_caches(vcpu))
+ return -KVM_EFAULT;
+
+ down_read(¤t->mm->mmap_sem);
+ gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
+ up_read(¤t->mm->mmap_sem);
+
+ if (gpa == UNMAPPED_GVA)
+ return -KVM_EFAULT;
+ if (size == 1) {
+ if (!emulator_write_phys(vcpu, gpa, &a0, sizeof(a0)))
+ return -KVM_EFAULT;
+ } else if (size == 2) {
+ if (!is_long_mode(vcpu) && is_pae(vcpu))
+ value = (u64)a1 << 32 | a0;
+ else
+ value = a0;
+ if (!emulator_write_phys(vcpu, gpa, &value, sizeof(value)))
+ return -KVM_EFAULT;
+ } else
+ return -KVM_E2BIG;
+
+ return 0;
+}
+
+static int kvm_hypercall_flush_tlb(struct kvm_vcpu *vcpu)
+{
+ kvm_x86_ops->tlb_flush(vcpu);
+ return 0;
+}
+
+static int kvm_hypercall_release_pt(struct kvm_vcpu *vcpu, gpa_t addr)
+{
+ spin_lock(&vcpu->kvm->mmu_lock);
+ mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT);
+ spin_unlock(&vcpu->kvm->mmu_lock);
+ return 0;
+}
+
int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
{
unsigned long nr, a0, a1, a2, a3, ret;
@@ -2274,6 +2320,15 @@ int kvm_emulate_hypercall(struct kvm_vcp
case KVM_HC_VAPIC_POLL_IRQ:
ret = 0;
break;
+ case KVM_HYPERCALL_MMU_WRITE:
+ ret = kvm_hypercall_mmu_write(vcpu, a0, a1, a2, a3);
+ break;
+ case KVM_HYPERCALL_FLUSH_TLB:
+ ret = kvm_hypercall_flush_tlb(vcpu);
+ break;
+ case KVM_HYPERCALL_RELEASE_PT:
+ ret = kvm_hypercall_release_pt(vcpu, a0);
+ break;
default:
ret = -KVM_ENOSYS;
break;
Index: linux-2.6-x86-kvm/include/asm-x86/kvm_para.h
===================================================================
--- linux-2.6-x86-kvm.orig/include/asm-x86/kvm_para.h
+++ linux-2.6-x86-kvm/include/asm-x86/kvm_para.h
@@ -6,6 +6,7 @@
*/
#define KVM_CPUID_SIGNATURE 0x40000000
#define KVM_FEATURE_NOP_IO_DELAY 0
+#define KVM_FEATURE_MMU_WRITE 1
/* This CPUID returns a feature bitmap in eax. Before enabling a particular
* paravirtualization, the appropriate feature bit should be checked.
@@ -15,7 +16,8 @@
#ifdef __KERNEL__
#include <asm/processor.h>
-#define KVM_PARA_FEATURES (1UL << KVM_FEATURE_NOP_IO_DELAY)
+#define KVM_PARA_FEATURES ((1UL << KVM_FEATURE_NOP_IO_DELAY) | \
+ (1UL << KVM_FEATURE_MMU_WRITE))
/* This instruction is vmcall. On non-VT architectures, it will generate a
* trap that we will then rewrite to the appropriate instruction.
Index: linux-2.6-x86-kvm/include/linux/kvm_para.h
===================================================================
--- linux-2.6-x86-kvm.orig/include/linux/kvm_para.h
+++ linux-2.6-x86-kvm/include/linux/kvm_para.h
@@ -11,8 +11,13 @@
/* Return values for hypercalls */
#define KVM_ENOSYS 1000
+#define KVM_EFAULT EFAULT
+#define KVM_E2BIG E2BIG
-#define KVM_HC_VAPIC_POLL_IRQ 1
+#define KVM_HC_VAPIC_POLL_IRQ 1
+#define KVM_HYPERCALL_MMU_WRITE 2
+#define KVM_HYPERCALL_FLUSH_TLB 3
+#define KVM_HYPERCALL_RELEASE_PT 4
/*
* hypercalls use architecture specific
--
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
kvm-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/kvm-devel
