Use the lockless GVA helpers to implement the dynamic translation of
guest instructions. This will allow it to handle asynchronous TLB
flushes when they are implemented.
Signed-off-by: James Hogan <james.hogan@xxxxxxxxxx>
Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
Cc: "Radim Krčmář" <rkrcmar@xxxxxxxxxx>
Cc: Ralf Baechle <ralf@xxxxxxxxxxxxxx>
Cc: linux-mips@xxxxxxxxxxxxxx
Cc: kvm@xxxxxxxxxxxxxxx
---
arch/mips/kvm/dyntrans.c | 26 ++++++++++++++++++++++++--
1 file changed, 24 insertions(+), 2 deletions(-)
diff --git a/arch/mips/kvm/dyntrans.c b/arch/mips/kvm/dyntrans.c
index 60ebf5862d2b..f8e772564d74 100644
--- a/arch/mips/kvm/dyntrans.c
+++ b/arch/mips/kvm/dyntrans.c
@@ -33,10 +33,32 @@ static int kvm_mips_trans_replace(struct kvm_vcpu *vcpu, u32 *opc,
unsigned long vaddr = (unsigned long)opc;
int err;
+retry:
+ /* The GVA page table is still active so use the Linux TLB handlers */
+ kvm_trap_emul_gva_lockless_begin(vcpu);
err = put_user(replace.word, opc);
+ kvm_trap_emul_gva_lockless_end(vcpu);
+
if (unlikely(err)) {
- kvm_err("%s: Invalid address: %p\n", __func__, opc);
- return err;
+ /*
+ * We write protect clean pages in GVA page table so normal
+ * Linux TLB mod handler doesn't silently dirty the page.
+ * Its also possible we raced with a GVA invalidation.
+ * Try to force the page to become dirty.
+ */
+ err = kvm_trap_emul_gva_fault(vcpu, vaddr, true);
+ if (unlikely(err)) {
+ kvm_info("%s: Address unwriteable: %p\n",
+ __func__, opc);
+ return -EFAULT;
+ }
+
+ /*
+ * Try again. This will likely trigger a TLB refill, which will
+ * fetch the new dirty entry from the GVA page table, which
+ * should then succeed.
+ */
+ goto retry;
}
__local_flush_icache_user_range(vaddr, vaddr + 4);
--
git-series 0.8.10
