On Mon, May 13, 2019 at 7:40 AM Alexandre Chartre
<alexandre.chartre@xxxxxxxxxx> wrote:
>
> From: Liran Alon <liran.alon@xxxxxxxxxx>
>
> KVM isolation enter/exit is done by switching between the KVM address
> space and the kernel address space.
>
> Signed-off-by: Liran Alon <liran.alon@xxxxxxxxxx>
> Signed-off-by: Alexandre Chartre <alexandre.chartre@xxxxxxxxxx>
> ---
> arch/x86/kvm/isolation.c | 30 ++++++++++++++++++++++++------
> arch/x86/mm/tlb.c | 1 +
> include/linux/sched.h | 1 +
> 3 files changed, 26 insertions(+), 6 deletions(-)
>
> diff --git a/arch/x86/kvm/isolation.c b/arch/x86/kvm/isolation.c
> index db0a7ce..b0c789f 100644
> --- a/arch/x86/kvm/isolation.c
> +++ b/arch/x86/kvm/isolation.c
> @@ -1383,11 +1383,13 @@ static bool kvm_page_fault(struct pt_regs *regs, unsigned long error_code,
> printk(KERN_DEFAULT "KVM isolation: page fault %ld at %pS on %lx (%pS) while switching mm\n"
> " cr3=%lx\n"
> " kvm_mm=%px pgd=%px\n"
> - " active_mm=%px pgd=%px\n",
> + " active_mm=%px pgd=%px\n"
> + " kvm_prev_mm=%px pgd=%px\n",
> error_code, (void *)regs->ip, address, (void *)address,
> cr3,
> &kvm_mm, kvm_mm.pgd,
> - active_mm, active_mm->pgd);
> + active_mm, active_mm->pgd,
> + current->kvm_prev_mm, current->kvm_prev_mm->pgd);
> dump_stack();
>
> return false;
> @@ -1649,11 +1651,27 @@ void kvm_may_access_sensitive_data(struct kvm_vcpu *vcpu)
> kvm_isolation_exit();
> }
>
> +static void kvm_switch_mm(struct mm_struct *mm)
> +{
> + unsigned long flags;
> +
> + /*
> + * Disable interrupt before updating active_mm, otherwise if an
> + * interrupt occurs during the switch then the interrupt handler
> + * can be mislead about the mm effectively in use.
> + */
> + local_irq_save(flags);
> + current->kvm_prev_mm = current->active_mm;
Peter's NAK aside, why on Earth is this in task_struct? You cannot
possibly context switch while in isolation mode.
--Andy
