Skip to contentSkip to conversation

KVM: x86: skip gfn_track allocation when possible

PATCH Applied David Stevens 2 participants 4 messages Sep 22, 2021 – Sep 28, 2021

Subsystems (5)
DRM drivers x86 KVM Intel i915 +1 more subsystems
lore
Share & tools
lore
#0 lore raw

v2 cover KVM: x86: skip gfn_track allocation when possible

From: David Stevens <[email protected]>

Skip allocating gfn_track arrays when tracking of guest write access to pages is not required. For VMs where the allocation can be avoided, this saves 2 bytes per 4KB of guest memory.

Write tracking is used to manage shadow page tables in three cases - when tdp is not supported, when nested virtualization is used, and for GVT-g. If tdp_enable is set and the kernel is compiled without GVT-g, then the gfn_track arrays can be allocated lazily when the shadow MMU is initialized.

v1 -> v2:
- lazily allocate gfn_track when shadow MMU is initialized, instead of looking at cpuid

David Stevens (2):
KVM: x86: add config for non-kvm users of page tracking
KVM: x86: only allocate gfn_track when necessary

 arch/x86/include/asm/kvm_host.h       |  8 ++++
 arch/x86/include/asm/kvm_page_track.h |  5 +-
 arch/x86/kvm/Kconfig                  |  3 ++
 arch/x86/kvm/mmu/mmu.c                |  7 +++
 arch/x86/kvm/mmu/page_track.c         | 69 ++++++++++++++++++++++++++-
 arch/x86/kvm/x86.c                    |  2 +-
 drivers/gpu/drm/i915/Kconfig          |  1 +
 7 files changed, 91 insertions(+), 4 deletions(-)
Signature · 1 line
2.33.0.464.g1972c5931b-goog
#1 lore raw

PATCH v2 1/2 KVM: x86: add config for non-kvm users of page tracking

From: David Stevens <[email protected]>

Add a config option that allows kvm to determine whether or not there are any external users of page tracking.

---
 arch/x86/kvm/Kconfig         | 3 +++
 drivers/gpu/drm/i915/Kconfig | 1 +
 2 files changed, 4 insertions(+)
arch/x86/kvm/Kconfig +3 -0
index ac69894eab88..619186138176 100644
@@ -129,4 +129,7 @@ config KVM_MMU_AUDIT
 	 This option adds a R/W kVM module parameter 'mmu_audit', which allows
 	 auditing of KVM MMU events at runtime.
 
+config KVM_EXTERNAL_WRITE_TRACKING
+	bool
+
 endif # VIRTUALIZATION
drivers/gpu/drm/i915/Kconfig +1 -0
index f960f5d7664e..107762427648 100644
@@ -126,6 +126,7 @@ config DRM_I915_GVT_KVMGT
 	depends on DRM_I915_GVT
 	depends on KVM
 	depends on VFIO_MDEV
+	select KVM_EXTERNAL_WRITE_TRACKING
 	default n
 	help
 	  Choose this option if you want to enable KVMGT support for
Signature · 1 line
2.33.0.464.g1972c5931b-goog
#2 lore raw

PATCH v2 2/2 KVM: x86: only allocate gfn_track when necessary

In reply to David Stevens

From: David Stevens <[email protected]>

Avoid allocating the gfn_track arrays if nothing needs them. If there are no external to KVM users of the API (i.e. no GVT-g), then page tracking is only needed for shadow page tables. This means that when tdp is enabled and there are no external users, then the gfn_track arrays can be lazily allocated when the shadow MMU is actually used. This avoid allocations equal to .05% of guest memory when nested virtualization is not used, if the kernel is compiled without GVT-g.

---
 arch/x86/include/asm/kvm_host.h       |  8 ++++
 arch/x86/include/asm/kvm_page_track.h |  5 +-
 arch/x86/kvm/mmu/mmu.c                |  7 +++
 arch/x86/kvm/mmu/page_track.c         | 69 ++++++++++++++++++++++++++-
 arch/x86/kvm/x86.c                    |  2 +-
 5 files changed, 87 insertions(+), 4 deletions(-)
arch/x86/include/asm/kvm_host.h +8 -0
index f8f48a7ec577..3e8c952e512c 100644
@@ -1212,6 +1212,14 @@ struct kvm_arch {
 	 */
 	bool memslots_have_rmaps;
 
+	/*
+	 * Set when the KVM mmu needs guest write access page tracking. If
+	 * set, the necessary gfn_track arrays have been allocated for
+	 * all memslots and should be allocated for any newly created or
+	 * modified memslots.
+	 */
+	bool memslots_mmu_write_tracking;
+
 #if IS_ENABLED(CONFIG_HYPERV)
 	hpa_t	hv_root_tdp;
 	spinlock_t hv_root_tdp_lock;
arch/x86/include/asm/kvm_page_track.h +4 -1
index 87bd6025d91d..677f1eda9934 100644
@@ -49,8 +49,11 @@ struct kvm_page_track_notifier_node {
 void kvm_page_track_init(struct kvm *kvm);
 void kvm_page_track_cleanup(struct kvm *kvm);
 
+int kvm_page_track_enable_mmu_write_tracking(struct kvm *kvm);
+
 void kvm_page_track_free_memslot(struct kvm_memory_slot *slot);
-int kvm_page_track_create_memslot(struct kvm_memory_slot *slot,
+int kvm_page_track_create_memslot(struct kvm *kvm,
+				  struct kvm_memory_slot *slot,
 				  unsigned long npages);
 
 void kvm_slot_page_track_add_page(struct kvm *kvm,
arch/x86/kvm/mmu/mmu.c +7 -0
index 2d7e61122af8..c8fda983c6a5 100644
@@ -3503,6 +3503,10 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
 	if (r)
 		return r;
 
+	r = kvm_page_track_enable_mmu_write_tracking(vcpu->kvm);
+	if (r)
+		return r;
+
 	write_lock(&vcpu->kvm->mmu_lock);
 	r = make_mmu_pages_available(vcpu);
 	if (r < 0)
@@ -5699,6 +5703,9 @@ void kvm_mmu_init_vm(struct kvm *kvm)
 		 */
 		kvm->arch.memslots_have_rmaps = true;
 
+	if (!tdp_enabled)
+		kvm->arch.memslots_mmu_write_tracking = true;
+
 	node->track_write = kvm_mmu_pte_write;
 	node->track_flush_slot = kvm_mmu_invalidate_zap_pages_in_memslot;
 	kvm_page_track_register_notifier(kvm, node);
arch/x86/kvm/mmu/page_track.c +67 -2

109 lines not on this page · raw message

arch/x86/kvm/x86.c +1 -1
index 28ef14155726..169e2359a32e 100644
@@ -11474,7 +11474,7 @@ static int kvm_alloc_memslot_metadata(struct kvm *kvm,
 		}
 	}
 
-	if (kvm_page_track_create_memslot(slot, npages))
+	if (kvm_page_track_create_memslot(kvm, slot, npages))
 		goto out_free;
 
 	return 0;
Signature · 1 line
2.33.0.464.g1972c5931b-goog
maintainer Applied
#3 lore raw
In reply to David Stevens
Quoted 12 lines · David Stevens

From: David Stevens<[email protected]>

Skip allocating gfn_track arrays when tracking of guest write access to pages is not required. For VMs where the allocation can be avoided, this saves 2 bytes per 4KB of guest memory.

Write tracking is used to manage shadow page tables in three cases - when tdp is not supported, when nested virtualization is used, and for GVT-g. If tdp_enable is set and the kernel is compiled without GVT-g, then the gfn_track arrays can be allocated lazily when the shadow MMU is initialized.

Quoted 3 lines

v1 -> v2:
- lazily allocate gfn_track when shadow MMU is initialized, instead of looking at cpuid

Queued, thanks.

paolo