KVM: s390: Inject machine check into the nested guest
authorQingFeng Hao <haoqf@linux.vnet.ibm.com>
Wed, 7 Jun 2017 10:11:18 +0000 (12:11 +0200)
committerChristian Borntraeger <borntraeger@de.ibm.com>
Wed, 28 Jun 2017 10:42:33 +0000 (12:42 +0200)
With vsie feature enabled, kvm can support nested guests (guest-3).
So inject machine check to the guest-2 if it happens when the nested
guest is running. And guest-2 will detect the machine check belongs
to guest-3 and reinject it into guest-3.
The host (guest-1) tries to inject the machine check to the picked
destination vcpu if it's a floating machine check.

Signed-off-by: QingFeng Hao <haoqf@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
arch/s390/kvm/vsie.c

index e947657..715c19c 100644 (file)
 
 struct vsie_page {
        struct kvm_s390_sie_block scb_s;        /* 0x0000 */
+       /*
+        * the backup info for machine check. ensure it's at
+        * the same offset as that in struct sie_page!
+        */
+       struct mcck_volatile_info mcck_info;    /* 0x0200 */
        /* the pinned originial scb */
-       struct kvm_s390_sie_block *scb_o;       /* 0x0200 */
+       struct kvm_s390_sie_block *scb_o;       /* 0x0218 */
        /* the shadow gmap in use by the vsie_page */
-       struct gmap *gmap;                      /* 0x0208 */
+       struct gmap *gmap;                      /* 0x0220 */
        /* address of the last reported fault to guest2 */
-       unsigned long fault_addr;               /* 0x0210 */
-       __u8 reserved[0x0700 - 0x0218];         /* 0x0218 */
+       unsigned long fault_addr;               /* 0x0228 */
+       __u8 reserved[0x0700 - 0x0230];         /* 0x0230 */
        struct kvm_s390_crypto_cb crycb;        /* 0x0700 */
        __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
 };
@@ -801,6 +806,8 @@ static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 {
        struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
        struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
+       struct mcck_volatile_info *mcck_info;
+       struct sie_page *sie_page;
        int rc;
 
        handle_last_fault(vcpu, vsie_page);
@@ -822,6 +829,14 @@ static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
        local_irq_enable();
        vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
 
+       if (rc == -EINTR) {
+               VCPU_EVENT(vcpu, 3, "%s", "machine check");
+               sie_page = container_of(scb_s, struct sie_page, sie_block);
+               mcck_info = &sie_page->mcck_info;
+               kvm_s390_reinject_machine_check(vcpu, mcck_info);
+               return 0;
+       }
+
        if (rc > 0)
                rc = 0; /* we could still have an icpt */
        else if (rc == -EFAULT)