This patch series is for "queue" branch of [1] and based on patches[2].
It brings obvious time decreasement for migration with different page
size.
The QEMU counterpart will be sent out later, thanks.
[1] git://git.kernel.org/pub/scm/virt/kvm/kvm.git
[2] https://lore.kernel.org/kvm/[email protected]/
Keqian Zhu (3):
KVM/memslot: Move the initial set of dirty bitmap to arch
KVM/arm64: Support enabling dirty log graually in small chunks
KVM/arm64: Only set bits of dirty bitmap with valid translation
entries
Documentation/virt/kvm/api.rst | 2 +-
arch/arm/include/asm/kvm_host.h | 2 -
arch/arm64/include/asm/kvm_host.h | 5 +-
arch/x86/kvm/x86.c | 3 +
virt/kvm/arm/mmu.c | 175 +++++++++++++++++++++++++-----
virt/kvm/kvm_main.c | 3 -
6 files changed, 157 insertions(+), 33 deletions(-)
--
2.19.1
When KVM_DIRTY_LOG_INITIALLY_SET is enabled, we can only report these
pages that have valid translation entries to userspace, then userspace
don't need to do zero-check on other pages during VM migration.
Under the Huawei Kunpeng 920 2.6GHz platform, I did some tests on 128G
Linux VMs with different page size.
About the time of enabling dirty log: The memory pressure is 127GB.
Page size Before After
4K 1.8ms 341ms
2M 1.8ms 4ms
1G 1.8ms 2ms
About the time of migration: The memory pressure is 3GB and the migration
bandwidth is 500MB/s.
Page size Before After
4K 21s 6s
2M 21s 6s
1G 21s 7s
Signed-off-by: Keqian Zhu <[email protected]>
---
virt/kvm/arm/mmu.c | 161 ++++++++++++++++++++++++++++++++++++++-------
1 file changed, 137 insertions(+), 24 deletions(-)
diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
index 6c84de442a0e..0c7a5faf8609 100644
--- a/virt/kvm/arm/mmu.c
+++ b/virt/kvm/arm/mmu.c
@@ -1413,34 +1413,85 @@ static bool transparent_hugepage_adjust(kvm_pfn_t *pfnp, phys_addr_t *ipap)
return false;
}
+enum s2_operation {
+ S2_OP_WP, /* write protect page tables */
+ S2_OP_MD, /* mark dirty bitmap in memslot */
+};
+
/**
- * stage2_wp_ptes - write protect PMD range
+ * mark_range_dirty - mark a range of dirty bitmap
+ * @kvm: kvm instance for the VM
+ * @addr: range start address
+ * @end: range end address
+ *
+ * note: addr and end should belong to the same memslot.
+ */
+static void mark_range_dirty(struct kvm *kvm,
+ phys_addr_t addr,
+ phys_addr_t end)
+{
+ gfn_t gfn;
+ unsigned int start, nbits;
+ struct kvm_memory_slot *memslot = NULL;
+
+ gfn = addr >> PAGE_SHIFT;
+ memslot = gfn_to_memslot(kvm, gfn);
+
+ if (memslot && memslot->dirty_bitmap) {
+ start = gfn - memslot->base_gfn;
+ nbits = DIV_ROUND_UP(end, PAGE_SIZE) - gfn;
+ bitmap_set(memslot->dirty_bitmap, start, nbits);
+ }
+}
+
+/**
+ * stage2_op_ptes - do an operation on PMD range
+ * @kvm: kvm instance for the VM
+ * @op: the operation wanted
* @pmd: pointer to pmd entry
* @addr: range start address
* @end: range end address
*/
-static void stage2_wp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end)
+static void stage2_op_ptes(struct kvm *kvm,
+ enum s2_operation op,
+ pmd_t *pmd,
+ phys_addr_t addr,
+ phys_addr_t end)
{
pte_t *pte;
pte = pte_offset_kernel(pmd, addr);
do {
- if (!pte_none(*pte)) {
+ if (pte_none(*pte))
+ continue;
+
+ switch (op) {
+ case S2_OP_WP:
if (!kvm_s2pte_readonly(pte))
kvm_set_s2pte_readonly(pte);
+ break;
+ case S2_OP_MD:
+ mark_range_dirty(kvm, addr, addr + PAGE_SIZE);
+ break;
+ default:
+ break;
}
} while (pte++, addr += PAGE_SIZE, addr != end);
}
/**
- * stage2_wp_pmds - write protect PUD range
- * kvm: kvm instance for the VM
+ * stage2_op_pmds - do an operation on PUD range
+ * @kvm: kvm instance for the VM
+ * @op: the operation wanted
* @pud: pointer to pud entry
* @addr: range start address
* @end: range end address
*/
-static void stage2_wp_pmds(struct kvm *kvm, pud_t *pud,
- phys_addr_t addr, phys_addr_t end)
+static void stage2_op_pmds(struct kvm *kvm,
+ enum s2_operation op,
+ pud_t *pud,
+ phys_addr_t addr,
+ phys_addr_t end)
{
pmd_t *pmd;
phys_addr_t next;
@@ -1449,25 +1500,40 @@ static void stage2_wp_pmds(struct kvm *kvm, pud_t *pud,
do {
next = stage2_pmd_addr_end(kvm, addr, end);
- if (!pmd_none(*pmd)) {
- if (pmd_thp_or_huge(*pmd)) {
+ if (pmd_none(*pmd))
+ continue;
+
+ if (pmd_thp_or_huge(*pmd)) {
+ switch (op) {
+ case S2_OP_WP:
if (!kvm_s2pmd_readonly(pmd))
kvm_set_s2pmd_readonly(pmd);
- } else {
- stage2_wp_ptes(pmd, addr, next);
+ break;
+ case S2_OP_MD:
+ mark_range_dirty(kvm, addr, next);
+ break;
+ default:
+ break;
}
+ } else {
+ stage2_op_ptes(kvm, op, pmd, addr, next);
}
} while (pmd++, addr = next, addr != end);
}
/**
- * stage2_wp_puds - write protect PGD range
+ * stage2_op_puds - do an operation on PGD range
+ * @kvm: kvm instance for the VM
+ * @op: the operation wanted
* @pgd: pointer to pgd entry
* @addr: range start address
* @end: range end address
*/
-static void stage2_wp_puds(struct kvm *kvm, pgd_t *pgd,
- phys_addr_t addr, phys_addr_t end)
+static void stage2_op_puds(struct kvm *kvm,
+ enum s2_operation op,
+ pgd_t *pgd,
+ phys_addr_t addr,
+ phys_addr_t end)
{
pud_t *pud;
phys_addr_t next;
@@ -1475,24 +1541,38 @@ static void stage2_wp_puds(struct kvm *kvm, pgd_t *pgd,
pud = stage2_pud_offset(kvm, pgd, addr);
do {
next = stage2_pud_addr_end(kvm, addr, end);
- if (!stage2_pud_none(kvm, *pud)) {
- if (stage2_pud_huge(kvm, *pud)) {
+ if (stage2_pud_none(kvm, *pud))
+ continue;
+
+ if (stage2_pud_huge(kvm, *pud)) {
+ switch (op) {
+ case S2_OP_WP:
if (!kvm_s2pud_readonly(pud))
kvm_set_s2pud_readonly(pud);
- } else {
- stage2_wp_pmds(kvm, pud, addr, next);
+ break;
+ case S2_OP_MD:
+ mark_range_dirty(kvm, addr, next);
+ break;
+ default:
+ break;
}
+ } else {
+ stage2_op_pmds(kvm, op, pud, addr, next);
}
} while (pud++, addr = next, addr != end);
}
/**
- * stage2_wp_range() - write protect stage2 memory region range
+ * stage2_op_range() - do an operation on stage2 memory region range
* @kvm: The KVM pointer
+ * @op: The operation wanted
* @addr: Start address of range
* @end: End address of range
*/
-static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
+static void stage2_op_range(struct kvm *kvm,
+ enum s2_operation op,
+ phys_addr_t addr,
+ phys_addr_t end)
{
pgd_t *pgd;
phys_addr_t next;
@@ -1513,7 +1593,7 @@ static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
break;
next = stage2_pgd_addr_end(kvm, addr, end);
if (stage2_pgd_present(kvm, *pgd))
- stage2_wp_puds(kvm, pgd, addr, next);
+ stage2_op_puds(kvm, op, pgd, addr, next);
} while (pgd++, addr = next, addr != end);
}
@@ -1543,11 +1623,44 @@ static void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
spin_lock(&kvm->mmu_lock);
- stage2_wp_range(kvm, start, end);
+ stage2_op_range(kvm, S2_OP_WP, start, end);
spin_unlock(&kvm->mmu_lock);
kvm_flush_remote_tlbs(kvm);
}
+/**
+ * kvm_mmu_md_memory_region() - mark dirty bitmap for memory slot
+ * @kvm: The KVM pointer
+ * @slot: The memory slot to mark dirty
+ *
+ * Called to mark dirty bitmap after memory region KVM_MEM_LOG_DIRTY_PAGES
+ * operation is called and kvm_dirty_log_manual_protect_and_init_set is
+ * true. After this function returns, a bit of dirty bitmap is set if its
+ * corresponding page table (including PUD, PMD and PTEs) is present.
+ *
+ * Afterwards read of dirty page log can be called and present PUD, PMD and
+ * PTEs can be write protected by userspace manually.
+ *
+ * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired,
+ * serializing operations for VM memory regions.
+ */
+static void kvm_mmu_md_memory_region(struct kvm *kvm, int slot)
+{
+ struct kvm_memslots *slots = kvm_memslots(kvm);
+ struct kvm_memory_slot *memslot = id_to_memslot(slots, slot);
+ phys_addr_t start, end;
+
+ if (WARN_ON_ONCE(!memslot))
+ return;
+
+ start = memslot->base_gfn << PAGE_SHIFT;
+ end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
+
+ spin_lock(&kvm->mmu_lock);
+ stage2_op_range(kvm, S2_OP_MD, start, end);
+ spin_unlock(&kvm->mmu_lock);
+}
+
/**
* kvm_mmu_write_protect_pt_masked() - write protect dirty pages
* @kvm: The KVM pointer
@@ -1567,7 +1680,7 @@ static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm,
phys_addr_t start = (base_gfn + __ffs(mask)) << PAGE_SHIFT;
phys_addr_t end = (base_gfn + __fls(mask) + 1) << PAGE_SHIFT;
- stage2_wp_range(kvm, start, end);
+ stage2_op_range(kvm, S2_OP_WP, start, end);
}
/*
@@ -2274,7 +2387,7 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
* write protect any pages because they're reported
* as dirty here.
*/
- bitmap_set(new->dirty_bitmap, 0, new->npages);
+ kvm_mmu_md_memory_region(kvm, mem->slot);
}
}
}
--
2.19.1
There is already support of enabling dirty log gradually in small chunks
for x86. This adds support for arm64. However, unlike x86, all pages,
including huge pages and normal pages, don't need to be write protected
when enabling dirty log if KVM_DIRTY_LOG_INITIALLY_SET is eanbled. For
that both huge pages and normal pages can be write protected later by
userspace manually.
Under the Huawei Kunpeng 920 2.6GHz platform, I did some tests on 128G
Linux VMs with different page size. The memory pressure is 127G in each
case. The time taken of memory_global_dirty_log_start in QEMU is listed
below:
Page Size Before After Optimization
4K 650ms 1.8ms
2M 4ms 1.8ms
1G 2ms 1.8ms
Signed-off-by: Keqian Zhu <[email protected]>
---
Documentation/virt/kvm/api.rst | 2 +-
arch/arm/include/asm/kvm_host.h | 2 --
arch/arm64/include/asm/kvm_host.h | 5 ++++-
virt/kvm/arm/mmu.c | 16 +++++++++++++---
4 files changed, 18 insertions(+), 7 deletions(-)
diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index 158d1186d103..8bfe86d86363 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -5771,7 +5771,7 @@ will be initialized to 1 when created. This also improves performance because
dirty logging can be enabled gradually in small chunks on the first call
to KVM_CLEAR_DIRTY_LOG. KVM_DIRTY_LOG_INITIALLY_SET depends on
KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE (it is also only available on
-x86 for now).
+x86 and arm64 for now).
KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 was previously available under the name
KVM_CAP_MANUAL_DIRTY_LOG_PROTECT, but the implementation had bugs that make
diff --git a/arch/arm/include/asm/kvm_host.h b/arch/arm/include/asm/kvm_host.h
index c3314b286a61..ffa815800de0 100644
--- a/arch/arm/include/asm/kvm_host.h
+++ b/arch/arm/include/asm/kvm_host.h
@@ -358,8 +358,6 @@ static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
return false;
}
-void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot);
-
struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
static inline bool kvm_arch_requires_vhe(void) { return false; }
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index d87aa609d2b6..67d372e0c75a 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -16,6 +16,7 @@
#include <linux/jump_label.h>
#include <linux/kvm_types.h>
#include <linux/percpu.h>
+#include <linux/kvm.h>
#include <asm/arch_gicv3.h>
#include <asm/barrier.h>
#include <asm/cpufeature.h>
@@ -45,6 +46,9 @@
#define KVM_REQ_VCPU_RESET KVM_ARCH_REQ(2)
#define KVM_REQ_RECORD_STEAL KVM_ARCH_REQ(3)
+#define KVM_DIRTY_LOG_MANUAL_CAPS (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \
+ KVM_DIRTY_LOG_INITIALLY_SET)
+
DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
extern unsigned int kvm_sve_max_vl;
@@ -478,7 +482,6 @@ u64 __kvm_call_hyp(void *hypfn, ...);
})
void force_vm_exit(const cpumask_t *mask);
-void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot);
int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
int exception_index);
diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
index e3b9ee268823..6c84de442a0e 100644
--- a/virt/kvm/arm/mmu.c
+++ b/virt/kvm/arm/mmu.c
@@ -1530,7 +1530,7 @@ static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
* Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired,
* serializing operations for VM memory regions.
*/
-void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
+static void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
{
struct kvm_memslots *slots = kvm_memslots(kvm);
struct kvm_memory_slot *memslot = id_to_memslot(slots, slot);
@@ -2265,8 +2265,18 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
* allocated dirty_bitmap[], dirty pages will be be tracked while the
* memory slot is write protected.
*/
- if (change != KVM_MR_DELETE && mem->flags & KVM_MEM_LOG_DIRTY_PAGES)
- kvm_mmu_wp_memory_region(kvm, mem->slot);
+ if (change != KVM_MR_DELETE && mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
+ if (!kvm_dirty_log_manual_protect_and_init_set(kvm)) {
+ kvm_mmu_wp_memory_region(kvm, mem->slot);
+ } else {
+ /*
+ * If we're with initial-all-set, we don't need to
+ * write protect any pages because they're reported
+ * as dirty here.
+ */
+ bitmap_set(new->dirty_bitmap, 0, new->npages);
+ }
+ }
}
int kvm_arch_prepare_memory_region(struct kvm *kvm,
--
2.19.1
Currently, when KVM_DIRTY_LOG_INITIALLY_SET is enabled, the initial set
of memslot dirty bitmap is located at arch shared path. By making this
implementation be architecture depended, we can realize some architecture
depended optimizations (e.g. Only mark dirty for these pages with valid
translation entries, then userspace can use this information to only send
these pages during migration).
Signed-off-by: Keqian Zhu <[email protected]>
---
arch/x86/kvm/x86.c | 3 +++
virt/kvm/kvm_main.c | 3 ---
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e54c6ad628a8..ccab8b56bc7c 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -10031,6 +10031,9 @@ static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
* See the comments in fast_page_fault().
*/
if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
+ if (kvm_dirty_log_manual_protect_and_init_set(kvm))
+ bitmap_set(new->dirty_bitmap, 0, new->npages);
+
if (kvm_x86_ops->slot_enable_log_dirty) {
kvm_x86_ops->slot_enable_log_dirty(kvm, new);
} else {
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 28eae681859f..531d96c6871e 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -1289,9 +1289,6 @@ int __kvm_set_memory_region(struct kvm *kvm,
r = kvm_alloc_dirty_bitmap(&new);
if (r)
return r;
-
- if (kvm_dirty_log_manual_protect_and_init_set(kvm))
- bitmap_set(new.dirty_bitmap, 0, new.npages);
}
r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
--
2.19.1