xen/21109-x86-cpu-hotplug.patch

# HG changeset patch
# User Keir Fraser <keir.fraser@citrix.com>
# Date 1270624140 -3600
# Node ID 7794f61c61f3b0c90e367a87b287850b31645742
# Parent  adce8bc43fcccf8730b1da962be60c457c51fa1b
x86, cpu hotplug: Synchronise vcpu state earlier during cpu offline.

Needs to happen before non-idle VCPU is fully descheduled after CPU is
removed from cpu_online_map. Else sync_vcpu_execstate() doesn't work
properly.

Signed-off-by: Keir Fraser <keir.fraser@citrix.com>

# HG changeset patch
# User Keir Fraser <keir.fraser@citrix.com>
# Date 1270737112 -3600
# Node ID c9e8369e49be7396eca444cfce27e6782e4aa248
# Parent  a33909be109cefb0aef251c7c7e48168ed05512a
Fix two issues for CPU online/offline.

Firstly, we should return if we fail to get spin lock in cpu_down.
Secondly, in credit scheduler, the idlers need be limited only to
online map.

Signed-off-by: Jiang, Yunhong <yunhong.jiang@intel.com>

# HG changeset patch
# User Keir Fraser <keir.fraser@citrix.com>
# Date 1271090027 -3600
# Node ID 5d4038d41fd894f9fc71e64043d559d7c465bd15
# Parent  1d3bec66528900b1b442b6d36f24cd35ca076506
cpufreq: fix racing issue for cpu hotplug

To eliminate racing between dbs timer handler and cpufreq_del_cpu,
using kill_timer instead of stop_timer to make sure timer handler
execution finished before other stuff in cpufreq_del_cpu.

BTW, fix a lost point of cpufreq_statistic_lock taking sequence.

Signed-off-by: Wei Gang <gang.wei@intel.com>

--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -1442,7 +1442,8 @@ void context_switch(struct vcpu *prev, s
 
     set_current(next);
 
-    if ( (per_cpu(curr_vcpu, cpu) == next) || is_idle_vcpu(next) )
+    if ( (per_cpu(curr_vcpu, cpu) == next) ||
+         (is_idle_vcpu(next) && cpu_online(cpu)) )
     {
         local_irq_enable();
     }
--- a/xen/arch/x86/smpboot.c
+++ b/xen/arch/x86/smpboot.c
@@ -997,17 +997,13 @@ static int __devinit do_boot_cpu(int api
 	return boot_error;
 }
 
-static void idle_task_exit(void)
-{
-	/* Give up lazy state borrowed by this idle vcpu */
-	__sync_lazy_execstate();
-}
-
 void cpu_exit_clear(void)
 {
 	int cpu = raw_smp_processor_id();
 
-	idle_task_exit();
+	/* Previous non-idle state should be synchronised already. */
+	if (__sync_lazy_execstate())
+		BUG();
 
 	cpucount --;
 	cpu_uninit();
@@ -1302,6 +1298,13 @@ int __cpu_disable(void)
 
 	remove_siblinginfo(cpu);
 
+	/*
+	 * If we are running the idle vcpu, sync last non-idle vcpu's state
+	 * before changing cpu_online_map. If we are running non-idle vcpu,
+	 * we will synchronously sync the state in context_switch() later.
+	 */
+	__sync_lazy_execstate();
+
 	/* It's now safe to remove this processor from the online map */
 	cpu_clear(cpu, cpu_online_map);
 	fixup_irqs();
@@ -1340,10 +1343,8 @@ int cpu_down(unsigned int cpu)
 	int err = 0;
 
 	/* spin_trylock() avoids deadlock with stop_machine_run(). */
-	if (!spin_trylock(&cpu_add_remove_lock)) {
-		err = -EBUSY;
-		goto out;
-	}
+	if (!spin_trylock(&cpu_add_remove_lock))
+		return -EBUSY;
 
 	if (num_online_cpus() == 1) {
 		err = -EBUSY;
--- a/xen/common/sched_credit.c
+++ b/xen/common/sched_credit.c
@@ -410,7 +410,7 @@ _csched_cpu_pick(struct vcpu *vc, bool_t
      * like run two VCPUs on co-hyperthreads while there are idle cores
      * or sockets.
      */
-    idlers = csched_priv.idlers;
+    cpus_and(idlers, cpu_online_map, csched_priv.idlers);
     cpu_set(cpu, idlers);
     cpus_and(cpus, cpus, idlers);
     cpu_clear(cpu, cpus);
--- a/xen/drivers/acpi/pmstat.c
+++ b/xen/drivers/acpi/pmstat.c
@@ -86,12 +86,13 @@ int do_get_pm_info(struct xen_sysctl_get
     case PMSTAT_get_pxstat:
     {
         uint32_t ct;
-        struct pm_px *pxpt = cpufreq_statistic_data[op->cpuid];
+        struct pm_px *pxpt;
         spinlock_t *cpufreq_statistic_lock = 
                    &per_cpu(cpufreq_statistic_lock, op->cpuid);
 
         spin_lock(cpufreq_statistic_lock);
 
+        pxpt = cpufreq_statistic_data[op->cpuid];
         if ( !pxpt || !pxpt->u.pt || !pxpt->u.trans_pt )
         {
             spin_unlock(cpufreq_statistic_lock);
--- a/xen/drivers/cpufreq/cpufreq_ondemand.c
+++ b/xen/drivers/cpufreq/cpufreq_ondemand.c
@@ -196,9 +196,8 @@ static void dbs_timer_init(struct cpu_db
 {
     dbs_info->enable = 1;
 
-    if ( !dbs_timer[dbs_info->cpu].function )
-        init_timer(&dbs_timer[dbs_info->cpu], do_dbs_timer, 
-            (void *)dbs_info, dbs_info->cpu);
+    init_timer(&dbs_timer[dbs_info->cpu], do_dbs_timer,
+        (void *)dbs_info, dbs_info->cpu);
 
     set_timer(&dbs_timer[dbs_info->cpu], NOW()+dbs_tuners_ins.sampling_rate);
 
@@ -213,7 +212,7 @@ static void dbs_timer_exit(struct cpu_db
 {
     dbs_info->enable = 0;
     dbs_info->stoppable = 0;
-    stop_timer(&dbs_timer[dbs_info->cpu]);
+    kill_timer(&dbs_timer[dbs_info->cpu]);
 }
 
 int cpufreq_governor_dbs(struct cpufreq_policy *policy, unsigned int event)