From: Bhavana Nagendra <bnagendr@redhat.com>
Date: Fri, 15 Feb 2008 13:25:23 -0500
Subject: [xen] new vcpu lock/unlock helper functions
Message-id: 20080215182522.11293.96623.sendpatchset@localhost.localdomain
O-Subject: [RHEL5.2 PATCH 1] Introduce new vcpu lock/unlock helper functions
Bugzilla: 430938

Resolves BZ 430938

Introduce new vcpu_lock_affinity() and vcpu_unlock_affinity() helper
functions for use by x86's continue_hypercall_on_cpu().

This has two advantages:
1. We can lock out ordinary vcpu_set_affinity() commands from dom0.
2. We avoid the (in this case bogus) check for dom0_vcpus_pin.

Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Chris Lalancette <clalance@redhat.com>

diff --git a/arch/x86/domain.c b/arch/x86/domain.c
index 77a6583..9ff936e 100644
--- a/arch/x86/domain.c
+++ b/arch/x86/domain.c
@@ -1355,6 +1355,7 @@ static void continue_hypercall_on_cpu_helper(struct vcpu *v)
 {
     struct cpu_user_regs *regs = guest_cpu_user_regs();
     struct migrate_info *info = v->arch.continue_info;
+    cpumask_t mask = info->saved_affinity;
 
     regs->eax = info->func(info->data);
 
@@ -1363,7 +1364,7 @@ static void continue_hypercall_on_cpu_helper(struct vcpu *v)
 
     xfree(info);
 
-    vcpu_set_affinity(v, &v->cpu_affinity);
+    vcpu_unlock_affinity(v, &mask);
     schedule_tail(v);
 }
 
@@ -1371,7 +1372,6 @@ int continue_hypercall_on_cpu(int cpu, long (*func)(void *data), void *data)
 {
     struct vcpu *v = current;
     struct migrate_info *info;
-    cpumask_t mask = cpumask_of_cpu(cpu);
     int rc;
 
     if ( cpu == smp_processor_id() )
@@ -1384,12 +1384,12 @@ int continue_hypercall_on_cpu(int cpu, long (*func)(void *data), void *data)
     info->func = func;
     info->data = data;
     info->saved_schedule_tail = v->arch.schedule_tail;
-    info->saved_affinity = v->cpu_affinity;
+    info->saved_affinity = cpumask_of_cpu(cpu);
 
     v->arch.schedule_tail = continue_hypercall_on_cpu_helper;
     v->arch.continue_info = info;
 
-    rc = vcpu_set_affinity(v, &mask);
+    rc = vcpu_lock_affinity(v, &info->saved_affinity);
     if ( rc )
     {
         v->arch.schedule_tail = info->saved_schedule_tail;
diff --git a/common/schedule.c b/common/schedule.c
index 3d80021..b27771c 100644
--- a/common/schedule.c
+++ b/common/schedule.c
@@ -264,12 +264,11 @@ void vcpu_force_reschedule(struct vcpu *v)
     }
 }
 
-int vcpu_set_affinity(struct vcpu *v, cpumask_t *affinity)
+static int __vcpu_set_affinity(
+    struct vcpu *v, cpumask_t *affinity,
+    bool_t old_lock_status, bool_t new_lock_status)
 {
-    cpumask_t online_affinity;
-
-    if ( (v->domain->domain_id == 0) && opt_dom0_vcpus_pin )
-        return -EINVAL;
+    cpumask_t online_affinity, old_affinity;
 
     cpus_and(online_affinity, *affinity, cpu_online_map);
     if ( cpus_empty(online_affinity) )
@@ -277,7 +276,18 @@ int vcpu_set_affinity(struct vcpu *v, cpumask_t *affinity)
 
     vcpu_schedule_lock_irq(v);
 
+    if ( v->affinity_locked != old_lock_status )
+    {
+        BUG_ON(!v->affinity_locked);
+        vcpu_schedule_unlock_irq(v);
+        return -EBUSY;
+    }
+
+    v->affinity_locked = new_lock_status;
+
+    old_affinity = v->cpu_affinity;
     v->cpu_affinity = *affinity;
+    *affinity = old_affinity;
     if ( !cpu_isset(v->processor, v->cpu_affinity) )
         set_bit(_VPF_migrating, &v->pause_flags);
 
@@ -292,6 +302,31 @@ int vcpu_set_affinity(struct vcpu *v, cpumask_t *affinity)
     return 0;
 }
 
+int vcpu_set_affinity(struct vcpu *v, cpumask_t *affinity)
+{
+    if ( (v->domain->domain_id == 0) && opt_dom0_vcpus_pin )
+        return -EINVAL;
+    return __vcpu_set_affinity(v, affinity, 0, 0);
+}
+
+int vcpu_lock_affinity(struct vcpu *v, cpumask_t *affinity)
+{
+    return __vcpu_set_affinity(v, affinity, 0, 1);
+}
+
+void vcpu_unlock_affinity(struct vcpu *v, cpumask_t *affinity)
+{
+    cpumask_t online_affinity;
+
+    /* Do not fail if no CPU in old affinity mask is online. */
+    cpus_and(online_affinity, *affinity, cpu_online_map);
+    if ( cpus_empty(online_affinity) )
+        *affinity = cpu_online_map;
+
+    if ( __vcpu_set_affinity(v, affinity, 1, 0) != 0 )
+        BUG();
+}
+
 /* Block the currently-executing domain until a pertinent event occurs. */
 static long do_block(void)
 {
diff --git a/include/xen/sched.h b/include/xen/sched.h
index 3d9f190..ec6afa1 100644
--- a/include/xen/sched.h
+++ b/include/xen/sched.h
@@ -119,6 +119,8 @@ struct vcpu
     bool_t           defer_shutdown;
     /* VCPU is paused following shutdown request (d->is_shutting_down)? */
     bool_t           paused_for_shutdown;
+    /* VCPU affinity is temporarily locked from controller changes? */
+    bool_t           affinity_locked;
 
     unsigned long    pause_flags;
     atomic_t         pause_count;
@@ -476,6 +478,8 @@ void cpu_init(void);
 
 void vcpu_force_reschedule(struct vcpu *v);
 int vcpu_set_affinity(struct vcpu *v, cpumask_t *affinity);
+int vcpu_lock_affinity(struct vcpu *v, cpumask_t *affinity);
+void vcpu_unlock_affinity(struct vcpu *v, cpumask_t *affinity);
 
 void vcpu_runstate_get(struct vcpu *v, struct vcpu_runstate_info *runstate);