[PATCH 3.2 50/61] KVM: async_pf: Fix #DF due to inject "Page not Present" and "Page Ready" exceptions simultaneously

[Ben Hutchings]

3.2.96-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>

commit 9a6e7c39810e4a8bc7fc95056cefb40583fe07ef upstream.

qemu-system-x86-8600  [004] d..1  7205.687530: kvm_entry: vcpu 2
qemu-system-x86-8600  [004] ....  7205.687532: kvm_exit: reason EXCEPTION_NMI rip 0xffffffffa921297d info ffffeb2c0e44e018 80000b0e
qemu-system-x86-8600  [004] ....  7205.687532: kvm_page_fault: address ffffeb2c0e44e018 error_code 0
qemu-system-x86-8600  [004] ....  7205.687620: kvm_try_async_get_page: gva = 0xffffeb2c0e44e018, gfn = 0x427e4e
qemu-system-x86-8600  [004] .N..  7205.687628: kvm_async_pf_not_present: token 0x8b002 gva 0xffffeb2c0e44e018
    kworker/4:2-7814  [004] ....  7205.687655: kvm_async_pf_completed: gva 0xffffeb2c0e44e018 address 0x7fcc30c4e000
qemu-system-x86-8600  [004] ....  7205.687703: kvm_async_pf_ready: token 0x8b002 gva 0xffffeb2c0e44e018
qemu-system-x86-8600  [004] d..1  7205.687711: kvm_entry: vcpu 2

After running some memory intensive workload in guest, I catch the kworker
which completes the GUP too quickly, and queues an "Page Ready" #PF exception
after the "Page not Present" exception before the next vmentry as the above
trace which will result in #DF injected to guest.

This patch fixes it by clearing the queue for "Page not Present" if "Page Ready"
occurs before the next vmentry since the GUP has already got the required page
and shadow page table has already been fixed by "Page Ready" handler.

Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
Cc: Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>
Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
Fixes: 7c90705bf2a3 ("KVM: Inject asynchronous page fault into a PV guest if page is swapped out.")
[Changed indentation and added clearing of injected. - Radim]
Signed-off-by: Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>
[bwh: Backported to 3.2: Don't assign to kvm_queued_exception::injected or
 x86_exception::async_page_fault]
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
 arch/x86/kvm/x86.c | 37 ++++++++++++++++++++++++++++---------
 1 file changed, 28 insertions(+), 9 deletions(-)

--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -6885,6 +6885,13 @@ static int apf_put_user(struct kvm_vcpu
 				      sizeof(val));
 }
 
+static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
+{
+
+	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
+				      sizeof(u32));
+}
+
 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
 				     struct kvm_async_pf *work)
 {
@@ -6911,6 +6918,7 @@ void kvm_arch_async_page_present(struct
 				 struct kvm_async_pf *work)
 {
 	struct x86_exception fault;
+	u32 val;
 
 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
 	if (is_error_page(work->page))
@@ -6918,14 +6926,24 @@ void kvm_arch_async_page_present(struct
 	else
 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
 
-	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
-	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
-		fault.vector = PF_VECTOR;
-		fault.error_code_valid = true;
-		fault.error_code = 0;
-		fault.nested_page_fault = false;
-		fault.address = work->arch.token;
-		kvm_inject_page_fault(vcpu, &fault);
+	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
+	    !apf_get_user(vcpu, &val)) {
+		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
+		    vcpu->arch.exception.pending &&
+		    vcpu->arch.exception.nr == PF_VECTOR &&
+		    !apf_put_user(vcpu, 0)) {
+			vcpu->arch.exception.pending = false;
+			vcpu->arch.exception.nr = 0;
+			vcpu->arch.exception.has_error_code = false;
+			vcpu->arch.exception.error_code = 0;
+		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
+			fault.vector = PF_VECTOR;
+			fault.error_code_valid = true;
+			fault.error_code = 0;
+			fault.nested_page_fault = false;
+			fault.address = work->arch.token;
+			kvm_inject_page_fault(vcpu, &fault);
+		}
 	}
 	vcpu->arch.apf.halted = false;
 }