Add a mechanism to switch between WrProt and D-Bit based dirty logging
while the VM is running (and possibly dirty logging is already enabled).
Switching to/from PML is not currently supported, but that can be added
later.
Signed-off-by: Junaid Shahid <junaids@xxxxxxxxxx>
---
arch/x86/include/asm/kvm_host.h | 4 +
arch/x86/include/uapi/asm/kvm.h | 1 +
arch/x86/kvm/mmu.c | 146 +++++++++++++++++++++++++++++++-
arch/x86/kvm/mmu.h | 1 +
arch/x86/kvm/vmx.c | 10 ++-
5 files changed, 158 insertions(+), 4 deletions(-)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 3da22c92a5d6..bdbc87a26662 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1131,6 +1131,8 @@ struct kvm_x86_ops {
* function can also add any un-flushed dirty state maintained by
* the hardware to the mask (e.g. if flush_log_dirty is not
* implemented.)
+ * - switch_dirty_log_mode:
+ * Switch to the given dirty log mode.
*/
void (*slot_enable_log_dirty)(struct kvm *kvm,
struct kvm_memory_slot *slot);
@@ -1141,6 +1143,7 @@ struct kvm_x86_ops {
struct kvm_memory_slot *slot,
gfn_t offset, unsigned long *mask);
int (*write_log_dirty)(struct kvm_vcpu *vcpu);
+ int (*switch_dirty_log_mode)(struct kvm *kvm, u8 mode);
/* pmu operations of sub-arch */
const struct kvm_pmu_ops *pmu_ops;
@@ -1202,6 +1205,7 @@ struct kvm_arch_async_pf {
extern struct kvm_x86_ops *kvm_x86_ops;
extern u8 kvm_default_dirty_log_mode;
+extern u8 kvm_supported_dirty_log_modes;
#define __KVM_HAVE_ARCH_VM_ALLOC
static inline struct kvm *kvm_arch_alloc_vm(void)
diff --git a/arch/x86/include/uapi/asm/kvm.h b/arch/x86/include/uapi/asm/kvm.h
index 2b1c442bffe6..ff2ed65be75c 100644
--- a/arch/x86/include/uapi/asm/kvm.h
+++ b/arch/x86/include/uapi/asm/kvm.h
@@ -420,6 +420,7 @@ struct kvm_nested_state {
__u8 data[0];
};
+#define KVM_DIRTY_LOG_MODE_DEFAULT 0
#define KVM_DIRTY_LOG_MODE_WRPROT 1
#define KVM_DIRTY_LOG_MODE_DBIT 2
#define KVM_DIRTY_LOG_MODE_PML 4
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 0839b8cfdf66..4abc75c97593 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -265,6 +265,9 @@ static u64 __read_mostly shadow_nonpresent_or_rsvd_lower_gfn_mask;
u8 __read_mostly kvm_default_dirty_log_mode;
EXPORT_SYMBOL_GPL(kvm_default_dirty_log_mode);
+u8 __read_mostly kvm_supported_dirty_log_modes;
+EXPORT_SYMBOL_GPL(kvm_supported_dirty_log_modes);
+
static void mmu_spte_set(u64 *sptep, u64 spte);
static union kvm_mmu_page_role
kvm_mmu_calc_root_page_role(struct kvm_vcpu *vcpu);
@@ -436,6 +439,7 @@ void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
if (shadow_dirty_mask == 0) {
enable_d_bit_logging = false;
+ kvm_supported_dirty_log_modes &= ~KVM_DIRTY_LOG_MODE_DBIT;
if (kvm_default_dirty_log_mode == KVM_DIRTY_LOG_MODE_DBIT)
kvm_default_dirty_log_mode = KVM_DIRTY_LOG_MODE_WRPROT;
@@ -1704,6 +1708,30 @@ kvm_mmu_shadow_dirty_mask_test_and_clear(struct kvm *kvm,
return mask;
}
+/*
+ * Test the D bit in the SPTE(s) corresponding to the GFN and return true if any
+ * SPTE has the D bit set.
+ *
+ * The MMU lock should be held before calling this function.
+ */
+bool kvm_mmu_test_shadow_dirty_mask(struct kvm *kvm,
+ struct kvm_memory_slot *slot,
+ gfn_t gfn_offset)
+{
+ struct kvm_rmap_head *rmap_head;
+ u64 *sptep;
+ struct rmap_iterator iter;
+ u64 pte_bits = 0;
+
+ rmap_head = __gfn_to_rmap(slot->base_gfn + gfn_offset,
+ PT_PAGE_TABLE_LEVEL, slot);
+
+ for_each_rmap_spte(rmap_head, &iter, sptep)
+ pte_bits |= *sptep;
+
+ return pte_bits & shadow_dirty_mask;
+}
+
/**
* Gets the dirty state (if any) for selected PT level pages from the hardware
* MMU structures and resets the hardware state to track those pages again.
@@ -6081,9 +6109,13 @@ int kvm_mmu_module_init(void)
BUILD_BUG_ON(sizeof(union kvm_mmu_role) != sizeof(u64));
kvm_mmu_reset_all_pte_masks();
- kvm_default_dirty_log_mode = enable_d_bit_logging
- ? KVM_DIRTY_LOG_MODE_DBIT
- : KVM_DIRTY_LOG_MODE_WRPROT;
+ kvm_default_dirty_log_mode = KVM_DIRTY_LOG_MODE_WRPROT;
+ kvm_supported_dirty_log_modes = KVM_DIRTY_LOG_MODE_WRPROT;
+
+ if (enable_d_bit_logging) {
+ kvm_supported_dirty_log_modes |= KVM_DIRTY_LOG_MODE_DBIT;
+ kvm_default_dirty_log_mode = KVM_DIRTY_LOG_MODE_DBIT;
+ }
pte_list_desc_cache = kmem_cache_create("pte_list_desc",
sizeof(struct pte_list_desc),
@@ -6150,3 +6182,111 @@ void kvm_mmu_module_exit(void)
unregister_shrinker(&mmu_shrinker);
mmu_audit_disable();
}
+
+static void switch_dirty_log_mode_dbit_to_wrprot(struct kvm *kvm)
+{
+ ulong i;
+ struct kvm_memslots *slots = kvm_memslots(kvm);
+ struct kvm_memory_slot *memslot;
+ bool flush = false;
+
+ kvm_for_each_memslot(memslot, slots)
+ if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES)
+ flush |= kvm_mmu_slot_leaf_remove_write_access(kvm,
+ memslot);
+ /*
+ * We need to ensure that the write-protection gets propagated to all
+ * CPUs before we transfer the dirty bits to the dirty bitmap.
+ * Otherwise, it would be possible for some other CPU to write to a
+ * page some time after we have gone over that page in the loop below
+ * and then the page wouldn't get marked in the dirty bitmap.
+ */
+ if (flush)
+ kvm_flush_remote_tlbs(kvm);
+
+ spin_lock(&kvm->mmu_lock);
+
+ kvm_for_each_memslot(memslot, slots) {
+ if (!(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES))
+ continue;
+
+ for (i = 0; i < memslot->npages; i++)
+ if (!test_bit(i, memslot->dirty_bitmap) &&
+ kvm_mmu_test_shadow_dirty_mask(kvm, memslot, i))
+ set_bit(i, memslot->dirty_bitmap);
+ }
+ spin_unlock(&kvm->mmu_lock);
+
+ kvm->arch.dirty_logging_mode = KVM_DIRTY_LOG_MODE_WRPROT;
+}
+
+static void switch_dirty_log_mode_wrprot_to_dbit(struct kvm *kvm)
+{
+ struct kvm_memslots *slots = kvm_memslots(kvm);
+ struct kvm_memory_slot *memslot;
+
+ kvm_for_each_memslot(memslot, slots)
+ if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES)
+ kvm_mmu_slot_leaf_clear_dirty(kvm, memslot);
+
+ /*
+ * No need to initiate a TLB flush here, since any page for which we
+ * cleared the dirty bit above would already be marked in the dirty
+ * bitmap. It isn't until the next get_dirty_log or enable_log_dirty
+ * that the clearing of the dirty bits needs to be propagated
+ * everywhere.
+ */
+
+ kvm->arch.dirty_logging_mode = KVM_DIRTY_LOG_MODE_DBIT;
+
+ /*
+ * As an optimization, we could also remove the write-protection from
+ * all SPTEs here, rather than incurring faults as writes happen.
+ */
+}
+
+int kvm_mmu_switch_dirty_log_mode(struct kvm *kvm, u8 mode)
+{
+ int err = 0;
+ u8 old_mode;
+
+ if (mode == KVM_DIRTY_LOG_MODE_DEFAULT)
+ mode = kvm_default_dirty_log_mode;
+
+ if (hweight8(mode) != 1)
+ return -EINVAL;
+
+ if (!(mode & kvm_supported_dirty_log_modes)) {
+ kvm_err("Dirty logging mode %u is not supported.\n", mode);
+ return -ENOTSUPP;
+ }
+
+ kvm_debug("Switching dirty logging mode from %u to %u.\n",
+ kvm->arch.dirty_logging_mode, mode);
+
+ mutex_lock(&kvm->slots_lock);
+
+ old_mode = kvm->arch.dirty_logging_mode;
+
+ if (mode != old_mode) {
+ if (mode == KVM_DIRTY_LOG_MODE_WRPROT &&
+ old_mode == KVM_DIRTY_LOG_MODE_DBIT)
+ switch_dirty_log_mode_dbit_to_wrprot(kvm);
+ else if (mode == KVM_DIRTY_LOG_MODE_DBIT &&
+ old_mode == KVM_DIRTY_LOG_MODE_WRPROT)
+ switch_dirty_log_mode_wrprot_to_dbit(kvm);
+ else if (kvm_x86_ops->switch_dirty_log_mode)
+ err = kvm_x86_ops->switch_dirty_log_mode(kvm, mode);
+ else
+ err = -ENOTSUPP;
+ }
+
+ mutex_unlock(&kvm->slots_lock);
+
+ if (err)
+ kvm_err("Trying to switch dirty logging mode from "
+ "%u to %u failed with error %d.\n",
+ old_mode, mode, err);
+
+ return err;
+}
diff --git a/arch/x86/kvm/mmu.h b/arch/x86/kvm/mmu.h
index 6d39802a666d..b27dde010ec1 100644
--- a/arch/x86/kvm/mmu.h
+++ b/arch/x86/kvm/mmu.h
@@ -213,4 +213,5 @@ void kvm_mmu_gfn_allow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
struct kvm_memory_slot *slot, u64 gfn);
int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
+int kvm_mmu_switch_dirty_log_mode(struct kvm *kvm, u8 mode);
#endif
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index c6b0477c855e..232115b84fbb 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -7992,9 +7992,17 @@ static __init int hardware_setup(void)
kvm_x86_ops->slot_disable_log_dirty = NULL;
kvm_x86_ops->flush_log_dirty = NULL;
kvm_x86_ops->get_and_reset_log_dirty = NULL;
- } else
+ } else {
kvm_default_dirty_log_mode = KVM_DIRTY_LOG_MODE_PML;
+ /*
+ * Currently, switching between PML and other modes is not
+ * supported, so if PML is enabled, it is the only available
+ * mode.
+ */
+ kvm_supported_dirty_log_modes = KVM_DIRTY_LOG_MODE_PML;
+ }
+
if (!cpu_has_vmx_preemption_timer())
kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
--
2.19.1.568.g152ad8e336-goog
