Linux Virtualization - Jul 2016 - [PATCH v3 1/4] kernel/sched: introduce vcpu preempted check interface

change from v2:
	no code change, fix typos, update some comments

change from v1:
	a simplier definition of default vcpu_is_preempted
	skip mahcine type check on ppc, and add config. remove dedicated macro.
	add one patch to drop overload of rwsem_spin_on_owner and mutex_spin_on_owner. 
	add more comments
	thanks boqun and Peter's suggestion.

This patch set aims to fix lock holder preemption issues.

test-case:
perf record -a perf bench sched messaging -g 400 -p && perf report

before patch:
18.09%  sched-messaging  [kernel.vmlinux]  [k] osq_lock
12.28%  sched-messaging  [kernel.vmlinux]  [k] rwsem_spin_on_owner
 5.27%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
 3.89%  sched-messaging  [kernel.vmlinux]  [k] wait_consider_task
 3.64%  sched-messaging  [kernel.vmlinux]  [k] _raw_write_lock_irq
 3.41%  sched-messaging  [kernel.vmlinux]  [k] mutex_spin_on_owner.is
 2.49%  sched-messaging  [kernel.vmlinux]  [k] system_call

after patch:
 9.99%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
 5.28%  sched-messaging  [unknown]         [H] 0xc0000000000768e0
 4.27%  sched-messaging  [kernel.vmlinux]  [k] __copy_tofrom_user_power7
 3.77%  sched-messaging  [kernel.vmlinux]  [k] copypage_power7
 3.24%  sched-messaging  [kernel.vmlinux]  [k] _raw_write_lock_irq
 3.02%  sched-messaging  [kernel.vmlinux]  [k] system_call
 2.69%  sched-messaging  [kernel.vmlinux]  [k] wait_consider_task

We introduce interface bool vcpu_is_preempted(int cpu) and use it in some spin
loops of osq_lock, rwsem_spin_on_owner and mutex_spin_on_owner.
These spin_on_onwer variant also cause rcu stall before we apply this patch set

Pan Xinhui (4):
  kernel/sched: introduce vcpu preempted check interface
  powerpc/spinlock: support vcpu preempted check
  locking/osq: Drop the overhead of osq_lock()
  kernel/locking: Drop the overhead of {mutex,rwsem}_spin_on_owner

 arch/powerpc/include/asm/spinlock.h | 18 ++++++++++++++++++
 include/linux/sched.h               | 12 ++++++++++++
 kernel/locking/mutex.c              | 15 +++++++++++++--
 kernel/locking/osq_lock.c           | 10 +++++++++-
 kernel/locking/rwsem-xadd.c         | 16 +++++++++++++---
 5 files changed, 65 insertions(+), 6 deletions(-)

-- 
2.4.11

This patch supports to fix lock holder preemption issue.

For kernel users, we could use bool vcpu_is_preempted(int cpu) to detech if
one vcpu is preempted or not.

The default implementation is a macro defined by false. So compiler can
wrap it out if arch dose not support such vcpu pteempted check.

Suggested-by: Peter Zijlstra (Intel) <peterz at infradead.org>
Signed-off-by: Pan Xinhui <xinhui.pan at linux.vnet.ibm.com>
---
 include/linux/sched.h | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 6e42ada..cbe0574 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -3293,6 +3293,18 @@ static inline void set_task_cpu(struct task_struct *p,
unsigned int cpu)
 
 #endif /* CONFIG_SMP */
 
+/*
+ * In order to deal with a various lock holder preemption issues provide an
+ * interface to see if a vCPU is currently running or not.
+ *
+ * This allows us to terminate optimistic spin loops and block, analogous to
+ * the native optimistic spin heuristic of testing if the lock owner task is
+ * running or not.
+ */
+#ifndef vcpu_is_preempted
+#define vcpu_is_preempted(cpu)	false
+#endif
+
 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
 
-- 
2.4.11

This is to fix some lock holder preemption issues. Some other locks
implementation do a spin loop before acquiring the lock itself.
Currently kernel has an interface of bool vcpu_is_preempted(int cpu). It
takes the cpu as parameter and return true if the cpu is preempted. Then
kernel can break the spin loops upon the retval of vcpu_is_preempted().

As kernel has used this interface, So lets support it.

Only pSeries need support it. And the fact is powerNV are built into same
kernel image with pSeries. So we need return false if we are runnig as
powerNV. The another fact is that lppaca->yield_count keeps zero on
powerNV. So we can just skip the machine type check.

Suggested-by: Boqun Feng <boqun.feng at gmail.com>
Suggested-by: Peter Zijlstra (Intel) <peterz at infradead.org>
Signed-off-by: Pan Xinhui <xinhui.pan at linux.vnet.ibm.com>
---
 arch/powerpc/include/asm/spinlock.h | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/arch/powerpc/include/asm/spinlock.h
b/arch/powerpc/include/asm/spinlock.h
index 523673d..3ac9fcb 100644
--- a/arch/powerpc/include/asm/spinlock.h
+++ b/arch/powerpc/include/asm/spinlock.h
@@ -52,6 +52,24 @@
 #define SYNC_IO
 #endif
 
+/*
+ * This support kernel to check if one cpu is preempted or not.
+ * Then we can fix some lock holder preemption issue.
+ */
+#ifdef CONFIG_PPC_PSERIES
+#define vcpu_is_preempted vcpu_is_preempted
+static inline bool vcpu_is_preempted(int cpu)
+{
+	/*
+	 * pSeries and powerNV can be built into same kernel image. In
+	 * principle we need return false directly if we are running as
+	 * powerNV. However the yield_count is always zero on powerNV, So
+	 * skip such machine type check
+	 */
+	return !!(be32_to_cpu(lppaca_of(cpu).yield_count) & 1);
+}
+#endif
+
 static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
 {
 	return lock.slock == 0;
-- 
2.4.11

An over-committed guest with more vCPUs than pCPUs has a heavy overhead in
osq_lock().

This is because vCPU A hold the osq lock and yield out, vCPU B wait per_cpu
node->locked to be set. IOW, vCPU B wait vCPU A to run and unlock the osq
lock.

Kernel has an interface bool vcpu_is_preempted(int cpu) to see if a vCPU is
currently running or not. So break the spin loops on true condition.

test case:
perf record -a perf bench sched messaging -g 400 -p && perf report

before patch:
18.09%  sched-messaging  [kernel.vmlinux]  [k] osq_lock
12.28%  sched-messaging  [kernel.vmlinux]  [k] rwsem_spin_on_owner
 5.27%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
 3.89%  sched-messaging  [kernel.vmlinux]  [k] wait_consider_task
 3.64%  sched-messaging  [kernel.vmlinux]  [k] _raw_write_lock_irq
 3.41%  sched-messaging  [kernel.vmlinux]  [k] mutex_spin_on_owner.is
 2.49%  sched-messaging  [kernel.vmlinux]  [k] system_call

after patch:
20.68%  sched-messaging  [kernel.vmlinux]  [k] mutex_spin_on_owner
 8.45%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
 4.12%  sched-messaging  [kernel.vmlinux]  [k] system_call
 3.01%  sched-messaging  [kernel.vmlinux]  [k] system_call_common
 2.83%  sched-messaging  [kernel.vmlinux]  [k] copypage_power7
 2.64%  sched-messaging  [kernel.vmlinux]  [k] rwsem_spin_on_owner
 2.00%  sched-messaging  [kernel.vmlinux]  [k] osq_lock

Suggested-by: Boqun Feng <boqun.feng at gmail.com>
Signed-off-by: Pan Xinhui <xinhui.pan at linux.vnet.ibm.com>
---
 kernel/locking/osq_lock.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/kernel/locking/osq_lock.c b/kernel/locking/osq_lock.c
index 05a3785..858a0ed 100644
--- a/kernel/locking/osq_lock.c
+++ b/kernel/locking/osq_lock.c
@@ -21,6 +21,11 @@ static inline int encode_cpu(int cpu_nr)
 	return cpu_nr + 1;
 }
 
+static inline int node_cpu(struct optimistic_spin_node *node)
+{
+	return node->cpu - 1;
+}
+
 static inline struct optimistic_spin_node *decode_cpu(int encoded_cpu_val)
 {
 	int cpu_nr = encoded_cpu_val - 1;
@@ -118,8 +123,11 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 	while (!READ_ONCE(node->locked)) {
 		/*
 		 * If we need to reschedule bail... so we can block.
+		 * Use vcpu_is_preempted to detech lock holder preemption issue
+		 * and break the loop. vcpu_is_preempted is a macro defined by
+		 * false if arch does not support vcpu preempted check,
 		 */
-		if (need_resched())
+		if (need_resched() || vcpu_is_preempted(node_cpu(node->prev)))
 			goto unqueue;
 
 		cpu_relax_lowlatency();
-- 
2.4.11

An over-committed guest with more vCPUs than pCPUs has a heavy overload in
the two spin_on_owner. This blames on the lock holder preemption issue.

Kernel has an interface bool vcpu_is_preempted(int cpu) to see if a vCPU is
currently running or not. So break the spin loops on true condition.

test-case:
perf record -a perf bench sched messaging -g 400 -p && perf report

before patch:
20.68%  sched-messaging  [kernel.vmlinux]  [k] mutex_spin_on_owner
 8.45%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
 4.12%  sched-messaging  [kernel.vmlinux]  [k] system_call
 3.01%  sched-messaging  [kernel.vmlinux]  [k] system_call_common
 2.83%  sched-messaging  [kernel.vmlinux]  [k] copypage_power7
 2.64%  sched-messaging  [kernel.vmlinux]  [k] rwsem_spin_on_owner
 2.00%  sched-messaging  [kernel.vmlinux]  [k] osq_lock

after patch:
 9.99%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
 5.28%  sched-messaging  [unknown]         [H] 0xc0000000000768e0
 4.27%  sched-messaging  [kernel.vmlinux]  [k] __copy_tofrom_user_power7
 3.77%  sched-messaging  [kernel.vmlinux]  [k] copypage_power7
 3.24%  sched-messaging  [kernel.vmlinux]  [k] _raw_write_lock_irq
 3.02%  sched-messaging  [kernel.vmlinux]  [k] system_call
 2.69%  sched-messaging  [kernel.vmlinux]  [k] wait_consider_task

Signed-off-by: Pan Xinhui <xinhui.pan at linux.vnet.ibm.com>
---
 kernel/locking/mutex.c      | 15 +++++++++++++--
 kernel/locking/rwsem-xadd.c | 16 +++++++++++++---
 2 files changed, 26 insertions(+), 5 deletions(-)

diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
index 79d2d76..4f00b0c 100644
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -236,7 +236,13 @@ bool mutex_spin_on_owner(struct mutex *lock, struct
task_struct *owner)
 		 */
 		barrier();
 
-		if (!owner->on_cpu || need_resched()) {
+		/*
+		 * Use vcpu_is_preempted to detech lock holder preemption issue
+		 * and break. vcpu_is_preempted is a macro defined by false if
+		 * arch does not support vcpu preempted check,
+		 */
+		if (!owner->on_cpu || need_resched() ||
+				vcpu_is_preempted(task_cpu(owner))) {
 			ret = false;
 			break;
 		}
@@ -261,8 +267,13 @@ static inline int mutex_can_spin_on_owner(struct mutex
*lock)
 
 	rcu_read_lock();
 	owner = READ_ONCE(lock->owner);
+
+	/*
+	 * As lock holder preemption issue, we both skip spinning if task is not
+	 * on cpu or its cpu is preempted
+	 */
 	if (owner)
-		retval = owner->on_cpu;
+		retval = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner));
 	rcu_read_unlock();
 	/*
 	 * if lock->owner is not set, the mutex owner may have just acquired
diff --git a/kernel/locking/rwsem-xadd.c b/kernel/locking/rwsem-xadd.c
index 09e30c6..99eb8fd 100644
--- a/kernel/locking/rwsem-xadd.c
+++ b/kernel/locking/rwsem-xadd.c
@@ -319,7 +319,11 @@ static inline bool rwsem_can_spin_on_owner(struct
rw_semaphore *sem)
 		goto done;
 	}
 
-	ret = owner->on_cpu;
+	/*
+	 * As lock holder preemption issue, we both skip spinning if task is not
+	 * on cpu or its cpu is preempted
+	 */
+	ret = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner));
 done:
 	rcu_read_unlock();
 	return ret;
@@ -340,8 +344,14 @@ bool rwsem_spin_on_owner(struct rw_semaphore *sem, struct
task_struct *owner)
 		 */
 		barrier();
 
-		/* abort spinning when need_resched or owner is not running */
-		if (!owner->on_cpu || need_resched()) {
+		/*
+		 * abort spinning when need_resched or owner is not running or
+		 * owner's cpu is preempted. vcpu_is_preempted is a macro
+		 * defined by false if arch does not support vcpu preempted
+		 * check
+		 */
+		if (!owner->on_cpu || need_resched() ||
+				vcpu_is_preempted(task_cpu(owner))) {
 			rcu_read_unlock();
 			return false;
 		}
-- 
2.4.11

On 07/21/2016 01:45 PM, Pan Xinhui wrote:
> change from v2:
> 	no code change, fix typos, update some comments
> 
> change from v1:
> 	a simplier definition of default vcpu_is_preempted
> 	skip mahcine type check on ppc, and add config. remove dedicated macro.
> 	add one patch to drop overload of rwsem_spin_on_owner and
mutex_spin_on_owner.
> 	add more comments
> 	thanks boqun and Peter's suggestion.
> 
> This patch set aims to fix lock holder preemption issues.
> 
> test-case:
> perf record -a perf bench sched messaging -g 400 -p && perf report
> 
> before patch:
> 18.09%  sched-messaging  [kernel.vmlinux]  [k] osq_lock
> 12.28%  sched-messaging  [kernel.vmlinux]  [k] rwsem_spin_on_owner
>  5.27%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
>  3.89%  sched-messaging  [kernel.vmlinux]  [k] wait_consider_task
>  3.64%  sched-messaging  [kernel.vmlinux]  [k] _raw_write_lock_irq
>  3.41%  sched-messaging  [kernel.vmlinux]  [k] mutex_spin_on_owner.is
>  2.49%  sched-messaging  [kernel.vmlinux]  [k] system_call
> 
> after patch:
>  9.99%  sched-messaging  [kernel.vmlinux]  [k] mutex_unlock
>  5.28%  sched-messaging  [unknown]         [H] 0xc0000000000768e0
>  4.27%  sched-messaging  [kernel.vmlinux]  [k] __copy_tofrom_user_power7
>  3.77%  sched-messaging  [kernel.vmlinux]  [k] copypage_power7
>  3.24%  sched-messaging  [kernel.vmlinux]  [k] _raw_write_lock_irq
>  3.02%  sched-messaging  [kernel.vmlinux]  [k] system_call
>  2.69%  sched-messaging  [kernel.vmlinux]  [k] wait_consider_task
> 
> We introduce interface bool vcpu_is_preempted(int cpu) and use it in some
spin
> loops of osq_lock, rwsem_spin_on_owner and mutex_spin_on_owner.
> These spin_on_onwer variant also cause rcu stall before we apply this patch
set
> 
> Pan Xinhui (4):
>   kernel/sched: introduce vcpu preempted check interface
>   powerpc/spinlock: support vcpu preempted check
>   locking/osq: Drop the overhead of osq_lock()
>   kernel/locking: Drop the overhead of {mutex,rwsem}_spin_on_owner
> 
>  arch/powerpc/include/asm/spinlock.h | 18 ++++++++++++++++++
>  include/linux/sched.h               | 12 ++++++++++++
>  kernel/locking/mutex.c              | 15 +++++++++++++--
>  kernel/locking/osq_lock.c           | 10 +++++++++-
>  kernel/locking/rwsem-xadd.c         | 16 +++++++++++++---
>  5 files changed, 65 insertions(+), 6 deletions(-)
> 
Acked-by: Christian Borntraeger <borntraeger at de.ibm.com>
for the full series. With my patch on top this really improves
some benchmarks for overcommitted KVM guests.