KVM: Update Posted-Interrupts Descriptor when vCPU is preempted
authorFeng Wu <feng.wu@intel.com>
Fri, 18 Sep 2015 14:29:54 +0000 (22:29 +0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 1 Oct 2015 13:06:53 +0000 (15:06 +0200)
This patch updates the Posted-Interrupts Descriptor when vCPU
is preempted.

sched out:
- Set 'SN' to suppress furture non-urgent interrupts posted for
the vCPU.

sched in:
- Clear 'SN'
- Change NDST if vCPU is scheduled to a different CPU
- Set 'NV' to POSTED_INTR_VECTOR

Signed-off-by: Feng Wu <feng.wu@intel.com>
[Include asm/cpu.h to fix !CONFIG_SMP compilation. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index cb26b93a2cd7c90faa119e242f2df756466fb6a3..99f5c61954ea46f24e8024333d8dc7d10163396e 100644 (file)
@@ -35,6 +35,7 @@
 #include "kvm_cache_regs.h"
 #include "x86.h"
 
+#include <asm/cpu.h>
 #include <asm/io.h>
 #include <asm/desc.h>
 #include <asm/vmx.h>
@@ -1942,6 +1943,52 @@ static void vmx_load_host_state(struct vcpu_vmx *vmx)
        preempt_enable();
 }
 
+static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
+{
+       struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
+       struct pi_desc old, new;
+       unsigned int dest;
+
+       if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
+               !irq_remapping_cap(IRQ_POSTING_CAP))
+               return;
+
+       do {
+               old.control = new.control = pi_desc->control;
+
+               /*
+                * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
+                * are two possible cases:
+                * 1. After running 'pre_block', context switch
+                *    happened. For this case, 'sn' was set in
+                *    vmx_vcpu_put(), so we need to clear it here.
+                * 2. After running 'pre_block', we were blocked,
+                *    and woken up by some other guy. For this case,
+                *    we don't need to do anything, 'pi_post_block'
+                *    will do everything for us. However, we cannot
+                *    check whether it is case #1 or case #2 here
+                *    (maybe, not needed), so we also clear sn here,
+                *    I think it is not a big deal.
+                */
+               if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
+                       if (vcpu->cpu != cpu) {
+                               dest = cpu_physical_id(cpu);
+
+                               if (x2apic_enabled())
+                                       new.ndst = dest;
+                               else
+                                       new.ndst = (dest << 8) & 0xFF00;
+                       }
+
+                       /* set 'NV' to 'notification vector' */
+                       new.nv = POSTED_INTR_VECTOR;
+               }
+
+               /* Allow posting non-urgent interrupts */
+               new.sn = 0;
+       } while (cmpxchg(&pi_desc->control, old.control,
+                       new.control) != old.control);
+}
 /*
  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
  * vcpu mutex is already taken.
@@ -1992,10 +2039,27 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
                vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
                vmx->loaded_vmcs->cpu = cpu;
        }
+
+       vmx_vcpu_pi_load(vcpu, cpu);
+}
+
+static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
+{
+       struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
+
+       if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
+               !irq_remapping_cap(IRQ_POSTING_CAP))
+               return;
+
+       /* Set SN when the vCPU is preempted */
+       if (vcpu->preempted)
+               pi_set_sn(pi_desc);
 }
 
 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 {
+       vmx_vcpu_pi_put(vcpu);
+
        __vmx_load_host_state(to_vmx(vcpu));
        if (!vmm_exclusive) {
                __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
@@ -4427,6 +4491,22 @@ static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
 {
 #ifdef CONFIG_SMP
        if (vcpu->mode == IN_GUEST_MODE) {
+               struct vcpu_vmx *vmx = to_vmx(vcpu);
+
+               /*
+                * Currently, we don't support urgent interrupt,
+                * all interrupts are recognized as non-urgent
+                * interrupt, so we cannot post interrupts when
+                * 'SN' is set.
+                *
+                * If the vcpu is in guest mode, it means it is
+                * running instead of being scheduled out and
+                * waiting in the run queue, and that's the only
+                * case when 'SN' is set currently, warning if
+                * 'SN' is set.
+                */
+               WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
+
                apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
                                POSTED_INTR_VECTOR);
                return true;