KVM: PPC: Book3S PR: Don't corrupt guest state when kernel uses VMX
authorPaul Mackerras <paulus@samba.org>
Tue, 6 Aug 2013 04:14:33 +0000 (14:14 +1000)
committerAlexander Graf <agraf@suse.de>
Wed, 28 Aug 2013 14:41:14 +0000 (16:41 +0200)
Currently the code assumes that once we load up guest FP/VSX or VMX
state into the CPU, it stays valid in the CPU registers until we
explicitly flush it to the thread_struct.  However, on POWER7,
copy_page() and memcpy() can use VMX.  These functions do flush the
VMX state to the thread_struct before using VMX instructions, but if
this happens while we have guest state in the VMX registers, and we
then re-enter the guest, we don't reload the VMX state from the
thread_struct, leading to guest corruption.  This has been observed
to cause guest processes to segfault.

To fix this, we check before re-entering the guest that all of the
bits corresponding to facilities owned by the guest, as expressed
in vcpu->arch.guest_owned_ext, are set in current->thread.regs->msr.
Any bits that have been cleared correspond to facilities that have
been used by kernel code and thus flushed to the thread_struct, so
for them we reload the state from the thread_struct.

We also need to check current->thread.regs->msr before calling
giveup_fpu() or giveup_altivec(), since if the relevant bit is
clear, the state has already been flushed to the thread_struct and
to flush it again would corrupt it.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
arch/powerpc/kvm/book3s_pr.c

index 19cbac6f9d0345eef472e9f75f608b24eaedbf2a..983e5eda892fb4ede955fb99ec5a4e7ec71dd436 100644 (file)
@@ -468,7 +468,8 @@ void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
                 * both the traditional FP registers and the added VSX
                 * registers into thread.fpr[].
                 */
-               giveup_fpu(current);
+               if (current->thread.regs->msr & MSR_FP)
+                       giveup_fpu(current);
                for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
                        vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 
@@ -483,7 +484,8 @@ void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 
 #ifdef CONFIG_ALTIVEC
        if (msr & MSR_VEC) {
-               giveup_altivec(current);
+               if (current->thread.regs->msr & MSR_VEC)
+                       giveup_altivec(current);
                memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
                vcpu->arch.vscr = t->vscr;
        }
@@ -575,8 +577,6 @@ static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
        printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 #endif
 
-       current->thread.regs->msr |= msr;
-
        if (msr & MSR_FP) {
                for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
                        thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
@@ -598,12 +598,32 @@ static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 #endif
        }
 
+       current->thread.regs->msr |= msr;
        vcpu->arch.guest_owned_ext |= msr;
        kvmppc_recalc_shadow_msr(vcpu);
 
        return RESUME_GUEST;
 }
 
+/*
+ * Kernel code using FP or VMX could have flushed guest state to
+ * the thread_struct; if so, get it back now.
+ */
+static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
+{
+       unsigned long lost_ext;
+
+       lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
+       if (!lost_ext)
+               return;
+
+       if (lost_ext & MSR_FP)
+               kvmppc_load_up_fpu();
+       if (lost_ext & MSR_VEC)
+               kvmppc_load_up_altivec();
+       current->thread.regs->msr |= lost_ext;
+}
+
 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
                        unsigned int exit_nr)
 {
@@ -892,6 +912,7 @@ program_interrupt:
                } else {
                        kvmppc_fix_ee_before_entry();
                }
+               kvmppc_handle_lost_ext(vcpu);
        }
 
        trace_kvm_book3s_reenter(r, vcpu);