KVM: PPC: Book3s HV: Don't access runnable threads list without vcore lock
authorPaul Mackerras <paulus@samba.org>
Mon, 15 Oct 2012 01:16:48 +0000 (01:16 +0000)
committerAlexander Graf <agraf@suse.de>
Tue, 30 Oct 2012 09:54:55 +0000 (10:54 +0100)
There were a few places where we were traversing the list of runnable
threads in a virtual core, i.e. vc->runnable_threads, without holding
the vcore spinlock.  This extends the places where we hold the vcore
spinlock to cover everywhere that we traverse that list.

Since we possibly need to sleep inside kvmppc_book3s_hv_page_fault,
this moves the call of it from kvmppc_handle_exit out to
kvmppc_vcpu_run, where we don't hold the vcore lock.

In kvmppc_vcore_blocked, we don't actually need to check whether
all vcpus are ceded and don't have any pending exceptions, since the
caller has already done that.  The caller (kvmppc_run_vcpu) wasn't
actually checking for pending exceptions, so we add that.

The change of if to while in kvmppc_run_vcpu is to make sure that we
never call kvmppc_remove_runnable() when the vcore state is RUNNING or
EXITING.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
arch/powerpc/include/asm/kvm_asm.h
arch/powerpc/kvm/book3s_hv.c

index 76fdcfe..aabcdba 100644 (file)
 
 #define RESUME_FLAG_NV          (1<<0)  /* Reload guest nonvolatile state? */
 #define RESUME_FLAG_HOST        (1<<1)  /* Resume host? */
+#define RESUME_FLAG_ARCH1      (1<<2)
 
 #define RESUME_GUEST            0
 #define RESUME_GUEST_NV         RESUME_FLAG_NV
index 77dec0f..3a737a4 100644 (file)
@@ -57,6 +57,9 @@
 /* #define EXIT_DEBUG_SIMPLE */
 /* #define EXIT_DEBUG_INT */
 
+/* Used to indicate that a guest page fault needs to be handled */
+#define RESUME_PAGE_FAULT      (RESUME_GUEST | RESUME_FLAG_ARCH1)
+
 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 
@@ -431,7 +434,6 @@ static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
                              struct task_struct *tsk)
 {
        int r = RESUME_HOST;
-       int srcu_idx;
 
        vcpu->stat.sum_exits++;
 
@@ -491,16 +493,12 @@ static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
         * have been handled already.
         */
        case BOOK3S_INTERRUPT_H_DATA_STORAGE:
-               srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
-               r = kvmppc_book3s_hv_page_fault(run, vcpu,
-                               vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
-               srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
+               r = RESUME_PAGE_FAULT;
                break;
        case BOOK3S_INTERRUPT_H_INST_STORAGE:
-               srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
-               r = kvmppc_book3s_hv_page_fault(run, vcpu,
-                               kvmppc_get_pc(vcpu), 0);
-               srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
+               vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
+               vcpu->arch.fault_dsisr = 0;
+               r = RESUME_PAGE_FAULT;
                break;
        /*
         * This occurs if the guest executes an illegal instruction.
@@ -984,22 +982,24 @@ static int on_primary_thread(void)
  * Run a set of guest threads on a physical core.
  * Called with vc->lock held.
  */
-static int kvmppc_run_core(struct kvmppc_vcore *vc)
+static void kvmppc_run_core(struct kvmppc_vcore *vc)
 {
        struct kvm_vcpu *vcpu, *vcpu0, *vnext;
        long ret;
        u64 now;
        int ptid, i, need_vpa_update;
        int srcu_idx;
+       struct kvm_vcpu *vcpus_to_update[threads_per_core];
 
        /* don't start if any threads have a signal pending */
        need_vpa_update = 0;
        list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
                if (signal_pending(vcpu->arch.run_task))
-                       return 0;
-               need_vpa_update |= vcpu->arch.vpa.update_pending |
-                       vcpu->arch.slb_shadow.update_pending |
-                       vcpu->arch.dtl.update_pending;
+                       return;
+               if (vcpu->arch.vpa.update_pending ||
+                   vcpu->arch.slb_shadow.update_pending ||
+                   vcpu->arch.dtl.update_pending)
+                       vcpus_to_update[need_vpa_update++] = vcpu;
        }
 
        /*
@@ -1019,8 +1019,8 @@ static int kvmppc_run_core(struct kvmppc_vcore *vc)
         */
        if (need_vpa_update) {
                spin_unlock(&vc->lock);
-               list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
-                       kvmppc_update_vpas(vcpu);
+               for (i = 0; i < need_vpa_update; ++i)
+                       kvmppc_update_vpas(vcpus_to_update[i]);
                spin_lock(&vc->lock);
        }
 
@@ -1037,8 +1037,10 @@ static int kvmppc_run_core(struct kvmppc_vcore *vc)
                        vcpu->arch.ptid = ptid++;
                }
        }
-       if (!vcpu0)
-               return 0;               /* nothing to run */
+       if (!vcpu0) {
+               vc->vcore_state = VCORE_INACTIVE;
+               return;         /* nothing to run; should never happen */
+       }
        list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
                if (vcpu->arch.ceded)
                        vcpu->arch.ptid = ptid++;
@@ -1091,6 +1093,7 @@ static int kvmppc_run_core(struct kvmppc_vcore *vc)
        preempt_enable();
        kvm_resched(vcpu);
 
+       spin_lock(&vc->lock);
        now = get_tb();
        list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
                /* cancel pending dec exception if dec is positive */
@@ -1114,7 +1117,6 @@ static int kvmppc_run_core(struct kvmppc_vcore *vc)
                }
        }
 
-       spin_lock(&vc->lock);
  out:
        vc->vcore_state = VCORE_INACTIVE;
        vc->preempt_tb = mftb();
@@ -1125,8 +1127,6 @@ static int kvmppc_run_core(struct kvmppc_vcore *vc)
                        wake_up(&vcpu->arch.cpu_run);
                }
        }
-
-       return 1;
 }
 
 /*
@@ -1150,20 +1150,11 @@ static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
 {
        DEFINE_WAIT(wait);
-       struct kvm_vcpu *v;
-       int all_idle = 1;
 
        prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
        vc->vcore_state = VCORE_SLEEPING;
        spin_unlock(&vc->lock);
-       list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
-               if (!v->arch.ceded || v->arch.pending_exceptions) {
-                       all_idle = 0;
-                       break;
-               }
-       }
-       if (all_idle)
-               schedule();
+       schedule();
        finish_wait(&vc->wq, &wait);
        spin_lock(&vc->lock);
        vc->vcore_state = VCORE_INACTIVE;
@@ -1219,7 +1210,8 @@ static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
                vc->runner = vcpu;
                n_ceded = 0;
                list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
-                       n_ceded += v->arch.ceded;
+                       if (!v->arch.pending_exceptions)
+                               n_ceded += v->arch.ceded;
                if (n_ceded == vc->n_runnable)
                        kvmppc_vcore_blocked(vc);
                else
@@ -1240,8 +1232,9 @@ static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
        }
 
        if (signal_pending(current)) {
-               if (vc->vcore_state == VCORE_RUNNING ||
-                   vc->vcore_state == VCORE_EXITING) {
+               while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
+                      (vc->vcore_state == VCORE_RUNNING ||
+                       vc->vcore_state == VCORE_EXITING)) {
                        spin_unlock(&vc->lock);
                        kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
                        spin_lock(&vc->lock);
@@ -1261,6 +1254,7 @@ static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
 {
        int r;
+       int srcu_idx;
 
        if (!vcpu->arch.sane) {
                run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
@@ -1299,6 +1293,11 @@ int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
                    !(vcpu->arch.shregs.msr & MSR_PR)) {
                        r = kvmppc_pseries_do_hcall(vcpu);
                        kvmppc_core_prepare_to_enter(vcpu);
+               } else if (r == RESUME_PAGE_FAULT) {
+                       srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
+                       r = kvmppc_book3s_hv_page_fault(run, vcpu,
+                               vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
+                       srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
                }
        } while (r == RESUME_GUEST);