[RFC PATCH 03/18] KVM: Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock

isaku.yamahata@intel.com posted 18 patches 3 years, 7 months ago
There is a newer version of this series
[RFC PATCH 03/18] KVM: Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock
Posted by isaku.yamahata@intel.com 3 years, 7 months ago
From: Isaku Yamahata <isaku.yamahata@intel.com>

Because kvm_count_lock unnecessarily complicates the KVM locking convention
Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock for
simplicity.

Opportunistically add some comments on locking.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
---
 Documentation/virt/kvm/locking.rst | 14 +++++-------
 virt/kvm/kvm_main.c                | 36 +++++++++++++++++++++---------
 2 files changed, 30 insertions(+), 20 deletions(-)

diff --git a/Documentation/virt/kvm/locking.rst b/Documentation/virt/kvm/locking.rst
index 845a561629f1..8957e32aa724 100644
--- a/Documentation/virt/kvm/locking.rst
+++ b/Documentation/virt/kvm/locking.rst
@@ -216,15 +216,11 @@ time it will be set using the Dirty tracking mechanism described above.
 :Type:		mutex
 :Arch:		any
 :Protects:	- vm_list
-
-``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.
+                - kvm_usage_count
+                - hardware virtualization enable/disable
+:Comment:	Use cpus_read_lock() for hardware virtualization enable/disable
+                because hardware enabling/disabling must be atomic /wrt
+                migration.  The lock order is cpus lock => kvm_lock.
 
 ``kvm->mn_invalidate_lock``
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 515dfe9d3bcf..c6781fa30461 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -100,7 +100,6 @@ EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
  */
 
 DEFINE_MUTEX(kvm_lock);
-static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 LIST_HEAD(vm_list);
 
 static cpumask_var_t cpus_hardware_enabled;
@@ -4999,6 +4998,8 @@ static void hardware_enable_nolock(void *junk)
 	int cpu = raw_smp_processor_id();
 	int r;
 
+	WARN_ON_ONCE(preemptible());
+
 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
 		return;
 
@@ -5015,10 +5016,10 @@ static void hardware_enable_nolock(void *junk)
 
 static int kvm_starting_cpu(unsigned int cpu)
 {
-	raw_spin_lock(&kvm_count_lock);
+	mutex_lock(&kvm_lock);
 	if (kvm_usage_count)
 		hardware_enable_nolock(NULL);
-	raw_spin_unlock(&kvm_count_lock);
+	mutex_unlock(&kvm_lock);
 	return 0;
 }
 
@@ -5026,6 +5027,8 @@ static void hardware_disable_nolock(void *junk)
 {
 	int cpu = raw_smp_processor_id();
 
+	WARN_ON_ONCE(preemptible());
+
 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
 		return;
 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
@@ -5034,10 +5037,10 @@ static void hardware_disable_nolock(void *junk)
 
 static int kvm_dying_cpu(unsigned int cpu)
 {
-	raw_spin_lock(&kvm_count_lock);
+	mutex_lock(&kvm_lock);
 	if (kvm_usage_count)
 		hardware_disable_nolock(NULL);
-	raw_spin_unlock(&kvm_count_lock);
+	mutex_unlock(&kvm_lock);
 	return 0;
 }
 
@@ -5052,16 +5055,19 @@ static void hardware_disable_all_nolock(void)
 
 static void hardware_disable_all(void)
 {
-	raw_spin_lock(&kvm_count_lock);
+	cpus_read_lock();
+	mutex_lock(&kvm_lock);
 	hardware_disable_all_nolock();
-	raw_spin_unlock(&kvm_count_lock);
+	mutex_unlock(&kvm_lock);
+	cpus_read_unlock();
 }
 
 static int hardware_enable_all(void)
 {
 	int r = 0;
 
-	raw_spin_lock(&kvm_count_lock);
+	cpus_read_lock();
+	mutex_lock(&kvm_lock);
 
 	kvm_usage_count++;
 	if (kvm_usage_count == 1) {
@@ -5074,7 +5080,8 @@ static int hardware_enable_all(void)
 		}
 	}
 
-	raw_spin_unlock(&kvm_count_lock);
+	mutex_unlock(&kvm_lock);
+	cpus_read_unlock();
 
 	return r;
 }
@@ -5680,15 +5687,22 @@ static void kvm_init_debug(void)
 
 static int kvm_suspend(void)
 {
-	if (kvm_usage_count)
+	/*
+	 * The caller ensures that CPU hotlug is disabled by
+	 * cpu_hotplug_disable() and other CPUs are offlined.  No need for
+	 * locking.
+	 */
+	if (kvm_usage_count) {
+		lockdep_assert_not_held(&kvm_lock);
 		hardware_disable_nolock(NULL);
+	}
 	return 0;
 }
 
 static void kvm_resume(void)
 {
 	if (kvm_usage_count) {
-		lockdep_assert_not_held(&kvm_count_lock);
+		lockdep_assert_not_held(&kvm_lock);
 		hardware_enable_nolock(NULL);
 	}
 }
-- 
2.25.1
Re: [RFC PATCH 03/18] KVM: Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock
Posted by Chao Gao 3 years, 7 months ago
On Fri, Aug 19, 2022 at 11:00:09PM -0700, isaku.yamahata@intel.com wrote:
>From: Isaku Yamahata <isaku.yamahata@intel.com>
>
>Because kvm_count_lock unnecessarily complicates the KVM locking convention
>Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock for
>simplicity.
>
>Opportunistically add some comments on locking.
>
>Suggested-by: Sean Christopherson <seanjc@google.com>
>Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
>---
> static cpumask_var_t cpus_hardware_enabled;
>@@ -4999,6 +4998,8 @@ static void hardware_enable_nolock(void *junk)
> 	int cpu = raw_smp_processor_id();
> 	int r;
> 
>+	WARN_ON_ONCE(preemptible());
>+
> 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
> 		return;
> 
>@@ -5015,10 +5016,10 @@ static void hardware_enable_nolock(void *junk)
> 
> static int kvm_starting_cpu(unsigned int cpu)
> {
>-	raw_spin_lock(&kvm_count_lock);
>+	mutex_lock(&kvm_lock);

kvm_starting_cpu() is called with interrupt disabled. So we cannot use
sleeping locks (e.g., mutex) here.
Re: [RFC PATCH 03/18] KVM: Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock
Posted by Isaku Yamahata 3 years, 7 months ago
On Tue, Aug 23, 2022 at 10:26:09AM +0800,
Chao Gao <chao.gao@intel.com> wrote:

> On Fri, Aug 19, 2022 at 11:00:09PM -0700, isaku.yamahata@intel.com wrote:
> >From: Isaku Yamahata <isaku.yamahata@intel.com>
> >
> >Because kvm_count_lock unnecessarily complicates the KVM locking convention
> >Drop kvm_count_lock and instead protect kvm_usage_count with kvm_lock for
> >simplicity.
> >
> >Opportunistically add some comments on locking.
> >
> >Suggested-by: Sean Christopherson <seanjc@google.com>
> >Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
> >---
> > static cpumask_var_t cpus_hardware_enabled;
> >@@ -4999,6 +4998,8 @@ static void hardware_enable_nolock(void *junk)
> > 	int cpu = raw_smp_processor_id();
> > 	int r;
> > 
> >+	WARN_ON_ONCE(preemptible());
> >+
> > 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
> > 		return;
> > 
> >@@ -5015,10 +5016,10 @@ static void hardware_enable_nolock(void *junk)
> > 
> > static int kvm_starting_cpu(unsigned int cpu)
> > {
> >-	raw_spin_lock(&kvm_count_lock);
> >+	mutex_lock(&kvm_lock);
> 
> kvm_starting_cpu() is called with interrupt disabled. So we cannot use
> sleeping locks (e.g., mutex) here.


So your patch to move it to online section [1] is needed.
I thought I can pullin only some of your patches.  But whole your patches are
needed.

[1] https://lore.kernel.org/lkml/20220317091539.GA7257@gao-cwp/T/#mcc0fd81e7a19601e7c3ce451582c516d38f977f6
-- 
Isaku Yamahata <isaku.yamahata@gmail.com>