KVM: s390: VSIE: correctly handle MVPG when in VSIE
authorClaudio Imbrenda <imbrenda@linux.ibm.com>
Mon, 1 Feb 2021 20:54:13 +0000 (21:54 +0100)
committerChristian Borntraeger <borntraeger@de.ibm.com>
Wed, 24 Mar 2021 09:30:29 +0000 (10:30 +0100)
Correctly handle the MVPG instruction when issued by a VSIE guest.

Fixes: a3508fbe9dc6d ("KVM: s390: vsie: initial support for nested virtualization")
Cc: stable@vger.kernel.org # f85f1baaa189: KVM: s390: split kvm_s390_logical_to_effective
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Janosch Frank <frankja@linux.ibm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Link: https://lore.kernel.org/r/20210302174443.514363-4-imbrenda@linux.ibm.com
[borntraeger@de.ibm.com: apply fixup from Claudio]
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
arch/s390/kvm/vsie.c

index 78b6043..7f7ac30 100644 (file)
@@ -417,11 +417,6 @@ static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
                memcpy((void *)((u64)scb_o + 0xc0),
                       (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
                break;
-       case ICPT_PARTEXEC:
-               /* MVPG only */
-               memcpy((void *)((u64)scb_o + 0xc0),
-                      (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
-               break;
        }
 
        if (scb_s->ihcpu != 0xffffU)
@@ -984,6 +979,95 @@ static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 }
 
 /*
+ * Get a register for a nested guest.
+ * @vcpu the vcpu of the guest
+ * @vsie_page the vsie_page for the nested guest
+ * @reg the register number, the upper 4 bits are ignored.
+ * returns: the value of the register.
+ */
+static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
+{
+       /* no need to validate the parameter and/or perform error handling */
+       reg &= 0xf;
+       switch (reg) {
+       case 15:
+               return vsie_page->scb_s.gg15;
+       case 14:
+               return vsie_page->scb_s.gg14;
+       default:
+               return vcpu->run->s.regs.gprs[reg];
+       }
+}
+
+static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
+{
+       struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
+       unsigned long pei_dest, pei_src, src, dest, mask;
+       u64 *pei_block = &vsie_page->scb_o->mcic;
+       int edat, rc_dest, rc_src;
+       union ctlreg0 cr0;
+
+       cr0.val = vcpu->arch.sie_block->gcr[0];
+       edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
+       mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
+
+       dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
+       src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
+
+       rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
+       rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
+       /*
+        * Either everything went well, or something non-critical went wrong
+        * e.g. because of a race. In either case, simply retry.
+        */
+       if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
+               retry_vsie_icpt(vsie_page);
+               return -EAGAIN;
+       }
+       /* Something more serious went wrong, propagate the error */
+       if (rc_dest < 0)
+               return rc_dest;
+       if (rc_src < 0)
+               return rc_src;
+
+       /* The only possible suppressing exception: just deliver it */
+       if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
+               clear_vsie_icpt(vsie_page);
+               rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
+               WARN_ON_ONCE(rc_dest);
+               return 1;
+       }
+
+       /*
+        * Forward the PEI intercept to the guest if it was a page fault, or
+        * also for segment and region table faults if EDAT applies.
+        */
+       if (edat) {
+               rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
+               rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
+       } else {
+               rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
+               rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
+       }
+       if (!rc_dest && !rc_src) {
+               pei_block[0] = pei_dest;
+               pei_block[1] = pei_src;
+               return 1;
+       }
+
+       retry_vsie_icpt(vsie_page);
+
+       /*
+        * The host has edat, and the guest does not, or it was an ASCE type
+        * exception. The host needs to inject the appropriate DAT interrupts
+        * into the guest.
+        */
+       if (rc_dest)
+               return inject_fault(vcpu, rc_dest, dest, 1);
+       return inject_fault(vcpu, rc_src, src, 0);
+}
+
+/*
  * Run the vsie on a shadow scb and a shadow gmap, without any further
  * sanity checks, handling SIE faults.
  *
@@ -1071,6 +1155,10 @@ static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
                if ((scb_s->ipa & 0xf000) != 0xf000)
                        scb_s->ipa += 0x1000;
                break;
+       case ICPT_PARTEXEC:
+               if (scb_s->ipa == 0xb254)
+                       rc = vsie_handle_mvpg(vcpu, vsie_page);
+               break;
        }
        return rc;
 }