Re: [PATCH 16/16] KVM: arm64: pkvm: Unshare guest structs during teardown

[Quentin Perret <qperret@xxxxxxxxxx>]

On Monday 18 Oct 2021 at 11:32:13 (+0100), Quentin Perret wrote:
> Another option is to take a refcount on 'current' from
> kvm_arch_vcpu_run_map_fp() before sharing thread-specific structs with
> the hyp and release the refcount of the previous task after unsharing.
> But that means we'll have to also drop the refcount when the vcpu
> gets destroyed, as well as explicitly unshare at that point. Shouldn't
> be too bad I think. Thoughts?

Something like the below seems to work OK on my setup, including
SIGKILL'ing the guest and such. How much do you hate it?

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index f8be56d5342b..50598d704c71 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -322,6 +322,7 @@ struct kvm_vcpu_arch {
 
 	struct thread_info *host_thread_info;	/* hyp VA */
 	struct user_fpsimd_state *host_fpsimd_state;	/* hyp VA */
+	struct task_struct *parent_task;
 
 	struct {
 		/* {Break,watch}point registers */
@@ -738,6 +739,7 @@ int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
+void kvm_vcpu_unshare_task_fp(struct kvm_vcpu *vcpu);
 
 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
 {
diff --git a/arch/arm64/kvm/fpsimd.c b/arch/arm64/kvm/fpsimd.c
index 2fe1128d9f3d..27afeebbe1cb 100644
--- a/arch/arm64/kvm/fpsimd.c
+++ b/arch/arm64/kvm/fpsimd.c
@@ -15,6 +15,22 @@
 #include <asm/kvm_mmu.h>
 #include <asm/sysreg.h>
 
+void kvm_vcpu_unshare_task_fp(struct kvm_vcpu *vcpu)
+{
+	struct task_struct *p = vcpu->arch.parent_task;
+	struct user_fpsimd_state *fpsimd;
+	struct thread_info *ti;
+
+	if (!static_branch_likely(&kvm_protected_mode_initialized) || !p)
+		return;
+
+	ti = &p->thread_info;
+	kvm_unshare_hyp(ti, ti + 1);
+	fpsimd = &p->thread.uw.fpsimd_state;
+	kvm_unshare_hyp(fpsimd, fpsimd + 1);
+	put_task_struct(p);
+}
+
 /*
  * Called on entry to KVM_RUN unless this vcpu previously ran at least
  * once and the most recent prior KVM_RUN for this vcpu was called from
@@ -31,6 +47,8 @@ int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu)
 	struct thread_info *ti = &current->thread_info;
 	struct user_fpsimd_state *fpsimd = &current->thread.uw.fpsimd_state;
 
+	kvm_vcpu_unshare_task_fp(vcpu);
+
 	/*
 	 * Make sure the host task thread flags and fpsimd state are
 	 * visible to hyp:
@@ -45,6 +63,17 @@ int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu)
 
 	vcpu->arch.host_thread_info = kern_hyp_va(ti);
 	vcpu->arch.host_fpsimd_state = kern_hyp_va(fpsimd);
+
+	/*
+	 * We need to keep current's task_struct pinned until its data has been
+	 * unshared with the hypervisor to make sure it is not re-used by the
+	 * kernel and donated to someone else while already shared -- see
+	 * kvm_vcpu_unshare_task_fp() for the matching put_task_struct().
+	 */
+	if (static_branch_likely(&kvm_protected_mode_initialized)) {
+		get_task_struct(current);
+		vcpu->arch.parent_task = current;
+	}
 error:
 	return ret;
 }
diff --git a/arch/arm64/kvm/reset.c b/arch/arm64/kvm/reset.c
index 5ce36b0a3343..c2a2cd7d5748 100644
--- a/arch/arm64/kvm/reset.c
+++ b/arch/arm64/kvm/reset.c
@@ -141,7 +141,13 @@ bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu)
 
 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
 {
-	kfree(vcpu->arch.sve_state);
+	void *sve_state = vcpu->arch.sve_state;
+
+	kvm_vcpu_unshare_task_fp(vcpu);
+	kvm_unshare_hyp(vcpu, vcpu + 1);
+	if (sve_state && vcpu->arch.has_run_once)
+		kvm_unshare_hyp(sve_state, sve_state + vcpu_sve_state_size(vcpu));
+	kfree(sve_state);
 }
 
 static void kvm_vcpu_reset_sve(struct kvm_vcpu *vcpu)
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm