[PATCH 1/3] KVM: cleanup (physical) CPU hotplug

[PATCH 1/3] KVM: cleanup (physical) CPU hotplug

[Paolo Bonzini]

Remove the useless argument, and do not do anything if there are no
VMs running at the time of the hotplug.

Cc: stable@vger.kernel.org
Cc: kvm@vger.kernel.org
Cc: gleb@redhat.com
Cc: jan.kiszka@siemens.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 virt/kvm/kvm_main.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 979bff4..75522b3 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -2681,10 +2681,11 @@ static void hardware_enable_nolock(void *junk)
 	}
 }
 
-static void hardware_enable(void *junk)
+static void hardware_enable(void)
 {
 	raw_spin_lock(&kvm_lock);
-	hardware_enable_nolock(junk);
+	if (kvm_usage_count)
+		hardware_enable_nolock(NULL);
 	raw_spin_unlock(&kvm_lock);
 }
 
@@ -2698,10 +2699,11 @@ static void hardware_disable_nolock(void *junk)
 	kvm_arch_hardware_disable(NULL);
 }
 
-static void hardware_disable(void *junk)
+static void hardware_disable(void)
 {
 	raw_spin_lock(&kvm_lock);
-	hardware_disable_nolock(junk);
+	if (kvm_usage_count)
+		hardware_disable_nolock(NULL);
 	raw_spin_unlock(&kvm_lock);
 }
 
@@ -2756,12 +2758,12 @@ static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
 	case CPU_DYING:
 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
 		       cpu);
-		hardware_disable(NULL);
+		hardware_disable();
 		break;
 	case CPU_STARTING:
 		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
 		       cpu);
-		hardware_enable(NULL);
+		hardware_enable();
 		break;
 	}
 	return NOTIFY_OK;
-- 
1.8.3.1

[PATCH 2/3] KVM: protect kvm_usage_count with its own spinlock

[Paolo Bonzini]

The VM list need not be protected by a raw spinlock.  Separate the
two so that kvm_lock can be made non-raw.

Cc: stable@vger.kernel.org
Cc: kvm@vger.kernel.org
Cc: gleb@redhat.com
Cc: jan.kiszka@siemens.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 Documentation/virtual/kvm/locking.txt |  6 +++++-
 virt/kvm/kvm_main.c                   | 19 ++++++++++---------
 2 files changed, 15 insertions(+), 10 deletions(-)

diff --git a/Documentation/virtual/kvm/locking.txt b/Documentation/virtual/kvm/locking.txt
index a9f366e..ba9e1c2 100644
--- a/Documentation/virtual/kvm/locking.txt
+++ b/Documentation/virtual/kvm/locking.txt
@@ -135,7 +135,11 @@ Name:		kvm_lock
 Type:		raw_spinlock
 Arch:		any
 Protects:	- vm_list
-		- hardware virtualization enable/disable
+
+Name:		kvm_count_lock
+Type:		raw_spinlock_t
+Arch:		any
+Protects:	- hardware virtualization enable/disable
 Comment:	'raw' because hardware enabling/disabling must be atomic /wrt
 		migration.
 
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 75522b3..da13379 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -71,6 +71,7 @@ MODULE_LICENSE("GPL");
  */
 
 DEFINE_RAW_SPINLOCK(kvm_lock);
+static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 LIST_HEAD(vm_list);
 
 static cpumask_var_t cpus_hardware_enabled;
@@ -2683,10 +2684,10 @@ static void hardware_enable_nolock(void *junk)
 
 static void hardware_enable(void)
 {
-	raw_spin_lock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
 	if (kvm_usage_count)
 		hardware_enable_nolock(NULL);
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_unlock(&kvm_count_lock);
 }
 
 static void hardware_disable_nolock(void *junk)
@@ -2701,10 +2702,10 @@ static void hardware_disable_nolock(void *junk)
 
 static void hardware_disable(void)
 {
-	raw_spin_lock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
 	if (kvm_usage_count)
 		hardware_disable_nolock(NULL);
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_unlock(&kvm_count_lock);
 }
 
 static void hardware_disable_all_nolock(void)
@@ -2718,16 +2719,16 @@ static void hardware_disable_all_nolock(void)
 
 static void hardware_disable_all(void)
 {
-	raw_spin_lock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
 	hardware_disable_all_nolock();
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_unlock(&kvm_count_lock);
 }
 
 static int hardware_enable_all(void)
 {
 	int r = 0;
 
-	raw_spin_lock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
 
 	kvm_usage_count++;
 	if (kvm_usage_count == 1) {
@@ -2740,7 +2741,7 @@ static int hardware_enable_all(void)
 		}
 	}
 
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_unlock(&kvm_count_lock);
 
 	return r;
 }
@@ -3133,7 +3134,7 @@ static int kvm_suspend(void)
 static void kvm_resume(void)
 {
 	if (kvm_usage_count) {
-		WARN_ON(raw_spin_is_locked(&kvm_lock));
+		WARN_ON(raw_spin_is_locked(&kvm_count_lock));
 		hardware_enable_nolock(NULL);
 	}
 }
-- 
1.8.3.1

[PATCH 3/3] KVM: Convert kvm_lock back to non-raw spinlock

[Paolo Bonzini]

In commit e935b8372cf8 ("KVM: Convert kvm_lock to raw_spinlock"),
the kvm_lock was made a raw lock.  However, the kvm mmu_shrink()
function tries to grab the (non-raw) mmu_lock within the scope of
the raw locked kvm_lock being held.  This leads to the following:

BUG: sleeping function called from invalid context at kernel/rtmutex.c:659
in_atomic(): 1, irqs_disabled(): 0, pid: 55, name: kswapd0
Preemption disabled at:[<ffffffffa0376eac>] mmu_shrink+0x5c/0x1b0 [kvm]

Pid: 55, comm: kswapd0 Not tainted 3.4.34_preempt-rt
Call Trace:
 [<ffffffff8106f2ad>] __might_sleep+0xfd/0x160
 [<ffffffff817d8d64>] rt_spin_lock+0x24/0x50
 [<ffffffffa0376f3c>] mmu_shrink+0xec/0x1b0 [kvm]
 [<ffffffff8111455d>] shrink_slab+0x17d/0x3a0
 [<ffffffff81151f00>] ? mem_cgroup_iter+0x130/0x260
 [<ffffffff8111824a>] balance_pgdat+0x54a/0x730
 [<ffffffff8111fe47>] ? set_pgdat_percpu_threshold+0xa7/0xd0
 [<ffffffff811185bf>] kswapd+0x18f/0x490
 [<ffffffff81070961>] ? get_parent_ip+0x11/0x50
 [<ffffffff81061970>] ? __init_waitqueue_head+0x50/0x50
 [<ffffffff81118430>] ? balance_pgdat+0x730/0x730
 [<ffffffff81060d2b>] kthread+0xdb/0xe0
 [<ffffffff8106e122>] ? finish_task_switch+0x52/0x100
 [<ffffffff817e1e94>] kernel_thread_helper+0x4/0x10
 [<ffffffff81060c50>] ? __init_kthread_worker+0x

After the previous patch, kvm_lock need not be a raw spinlock anymore,
so change it back.

Reported-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: stable@vger.kernel.org
Cc: kvm@vger.kernel.org
Cc: gleb@redhat.com
Cc: jan.kiszka@siemens.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 Documentation/virtual/kvm/locking.txt |  2 +-
 arch/x86/kvm/mmu.c                    |  4 ++--
 arch/x86/kvm/x86.c                    |  8 ++++----
 include/linux/kvm_host.h              |  2 +-
 virt/kvm/kvm_main.c                   | 18 +++++++++---------
 5 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/Documentation/virtual/kvm/locking.txt b/Documentation/virtual/kvm/locking.txt
index ba9e1c2..f886941 100644
--- a/Documentation/virtual/kvm/locking.txt
+++ b/Documentation/virtual/kvm/locking.txt
@@ -132,7 +132,7 @@ See the comments in spte_has_volatile_bits() and mmu_spte_update().
 ------------
 
 Name:		kvm_lock
-Type:		raw_spinlock
+Type:		spinlock_t
 Arch:		any
 Protects:	- vm_list
 
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 6e2d2c8..d027a72 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -4429,7 +4429,7 @@ static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc)
 	if (nr_to_scan == 0)
 		goto out;
 
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 
 	list_for_each_entry(kvm, &vm_list, vm_list) {
 		int idx;
@@ -4473,7 +4473,7 @@ unlock:
 		break;
 	}
 
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 
 out:
 	return percpu_counter_read_positive(&kvm_total_used_mmu_pages);
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 7c36099..3acd631 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -5273,7 +5273,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va
 
 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
 
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_for_each_entry(kvm, &vm_list, vm_list) {
 		kvm_for_each_vcpu(i, vcpu, kvm) {
 			if (vcpu->cpu != freq->cpu)
@@ -5283,7 +5283,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va
 				send_ipi = 1;
 		}
 	}
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 
 	if (freq->old < freq->new && send_ipi) {
 		/*
@@ -5436,12 +5436,12 @@ static void pvclock_gtod_update_fn(struct work_struct *work)
 	struct kvm_vcpu *vcpu;
 	int i;
 
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_for_each_entry(kvm, &vm_list, vm_list)
 		kvm_for_each_vcpu(i, vcpu, kvm)
 			set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
 	atomic_set(&kvm_guest_has_master_clock, 0);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 }
 
 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 749bdb1..7c961e1 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -142,7 +142,7 @@ struct kvm;
 struct kvm_vcpu;
 extern struct kmem_cache *kvm_vcpu_cache;
 
-extern raw_spinlock_t kvm_lock;
+extern spinlock_t kvm_lock;
 extern struct list_head vm_list;
 
 struct kvm_io_range {
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index da13379..3397a9c 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -70,7 +70,7 @@ MODULE_LICENSE("GPL");
  * 		kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  */
 
-DEFINE_RAW_SPINLOCK(kvm_lock);
+DEFINE_SPINLOCK(kvm_lock);
 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 LIST_HEAD(vm_list);
 
@@ -491,9 +491,9 @@ static struct kvm *kvm_create_vm(unsigned long type)
 	if (r)
 		goto out_err;
 
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_add(&kvm->vm_list, &vm_list);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 
 	return kvm;
 
@@ -582,9 +582,9 @@ static void kvm_destroy_vm(struct kvm *kvm)
 	struct mm_struct *mm = kvm->mm;
 
 	kvm_arch_sync_events(kvm);
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_del(&kvm->vm_list);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 	kvm_free_irq_routing(kvm);
 	for (i = 0; i < KVM_NR_BUSES; i++)
 		kvm_io_bus_destroy(kvm->buses[i]);
@@ -3057,10 +3057,10 @@ static int vm_stat_get(void *_offset, u64 *val)
 	struct kvm *kvm;
 
 	*val = 0;
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_for_each_entry(kvm, &vm_list, vm_list)
 		*val += *(u32 *)((void *)kvm + offset);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 	return 0;
 }
 
@@ -3074,12 +3074,12 @@ static int vcpu_stat_get(void *_offset, u64 *val)
 	int i;
 
 	*val = 0;
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_for_each_entry(kvm, &vm_list, vm_list)
 		kvm_for_each_vcpu(i, vcpu, kvm)
 			*val += *(u32 *)((void *)vcpu + offset);
 
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 	return 0;
 }
 
-- 
1.8.3.1

Re: [PATCH 1/3] KVM: cleanup (physical) CPU hotplug

[Jan Kiszka]

On 2013-09-16 16:06, Paolo Bonzini wrote:
> Remove the useless argument, and do not do anything if there are no
> VMs running at the time of the hotplug.

kvm_cpu_hotplug already filters !kvm_usage_count. If we need the check
to be under kvm_lock, drop that line as well. If that is not required
(machine still halted?), drop the related changes here.

Jan

> 
> Cc: stable@vger.kernel.org
> Cc: kvm@vger.kernel.org
> Cc: gleb@redhat.com
> Cc: jan.kiszka@siemens.com
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  virt/kvm/kvm_main.c | 14 ++++++++------
>  1 file changed, 8 insertions(+), 6 deletions(-)
> 
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 979bff4..75522b3 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -2681,10 +2681,11 @@ static void hardware_enable_nolock(void *junk)
>  	}
>  }
>  
> -static void hardware_enable(void *junk)
> +static void hardware_enable(void)
>  {
>  	raw_spin_lock(&kvm_lock);
> -	hardware_enable_nolock(junk);
> +	if (kvm_usage_count)
> +		hardware_enable_nolock(NULL);
>  	raw_spin_unlock(&kvm_lock);
>  }
>  
> @@ -2698,10 +2699,11 @@ static void hardware_disable_nolock(void *junk)
>  	kvm_arch_hardware_disable(NULL);
>  }
>  
> -static void hardware_disable(void *junk)
> +static void hardware_disable(void)
>  {
>  	raw_spin_lock(&kvm_lock);
> -	hardware_disable_nolock(junk);
> +	if (kvm_usage_count)
> +		hardware_disable_nolock(NULL);
>  	raw_spin_unlock(&kvm_lock);
>  }
>  
> @@ -2756,12 +2758,12 @@ static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
>  	case CPU_DYING:
>  		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
>  		       cpu);
> -		hardware_disable(NULL);
> +		hardware_disable();
>  		break;
>  	case CPU_STARTING:
>  		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
>  		       cpu);
> -		hardware_enable(NULL);
> +		hardware_enable();
>  		break;
>  	}
>  	return NOTIFY_OK;
> 

-- 
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux

Re: [PATCH 1/3] KVM: cleanup (physical) CPU hotplug

[Paolo Bonzini]

Il 17/09/2013 09:57, Jan Kiszka ha scritto:
>> > Remove the useless argument, and do not do anything if there are no
>> > VMs running at the time of the hotplug.
> kvm_cpu_hotplug already filters !kvm_usage_count. If we need the check
> to be under kvm_lock, drop that line as well. If that is not required
> (machine still halted?), drop the related changes here.

CPU_DYING is called under stop_machine, CPU_STARTING is not.  So I
should drop the test in kvm_cpu_hotplug.  It's a bit clearer anyway to
not rely on stop_machine.

Paolo