Skip to contentSkip to conversation

KVM: VMX: switch wakeup_vcpus_on_cpu_lock to raw spinlock

PATCH Applied Marcelo Tosatti 3 participants 5 messages Jan 7, 2022 – Jan 18, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KVM: VMX: switch wakeup_vcpus_on_cpu_lock to raw spinlock

wakeup_vcpus_on_cpu_lock is taken from hard interrupt context (pi_wakeup_handler), therefore it cannot sleep.

Switch it to a raw spinlock.

Fixes:

[41297.066254] BUG: scheduling while atomic: CPU 0/KVM/635218/0x00010001 
[41297.066323] Preemption disabled at: 
[41297.066324] [<ffffffff902ee47f>] irq_enter_rcu+0xf/0x60 
[41297.066339] Call Trace: 
[41297.066342]  <IRQ> 
[41297.066346]  dump_stack_lvl+0x34/0x44 
[41297.066353]  ? irq_enter_rcu+0xf/0x60 
[41297.066356]  __schedule_bug.cold+0x7d/0x8b 
[41297.066361]  __schedule+0x439/0x5b0 
[41297.066365]  ? task_blocks_on_rt_mutex.constprop.0.isra.0+0x1b0/0x440 
[41297.066369]  schedule_rtlock+0x1e/0x40 
[41297.066371]  rtlock_slowlock_locked+0xf1/0x260 
[41297.066374]  rt_spin_lock+0x3b/0x60 
[41297.066378]  pi_wakeup_handler+0x31/0x90 [kvm_intel] 
[41297.066388]  sysvec_kvm_posted_intr_wakeup_ipi+0x9d/0xd0 
[41297.066392]  </IRQ> 
[41297.066392]  asm_sysvec_kvm_posted_intr_wakeup_ipi+0x12/0x20 

...

arch/x86/kvm/vmx/posted_intr.c +8 -8
index f4169c009400..aa1fe9085d77 100644
@@ -27,7 +27,7 @@ static DEFINE_PER_CPU(struct list_head, wakeup_vcpus_on_cpu);
  * CPU.  IRQs must be disabled when taking this lock, otherwise deadlock will
  * occur if a wakeup IRQ arrives and attempts to acquire the lock.
  */
-static DEFINE_PER_CPU(spinlock_t, wakeup_vcpus_on_cpu_lock);
+static DEFINE_PER_CPU(raw_spinlock_t, wakeup_vcpus_on_cpu_lock);
 
 static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
 {
@@ -87,9 +87,9 @@ void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
 	 * current pCPU if the task was migrated.
 	 */
 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) {
-		spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+		raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
 		list_del(&vmx->pi_wakeup_list);
-		spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+		raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
 	}
 
 	dest = cpu_physical_id(cpu);
@@ -149,10 +149,10 @@ static void pi_enable_wakeup_handler(struct kvm_vcpu *vcpu)
 
 	local_irq_save(flags);
 
-	spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+	raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
 	list_add_tail(&vmx->pi_wakeup_list,
 		      &per_cpu(wakeup_vcpus_on_cpu, vcpu->cpu));
-	spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+	raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
 
 	WARN(pi_desc->sn, "PI descriptor SN field set before blocking");
 
@@ -204,20 +204,20 @@ void pi_wakeup_handler(void)
 	int cpu = smp_processor_id();
 	struct vcpu_vmx *vmx;
 
-	spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
+	raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
 	list_for_each_entry(vmx, &per_cpu(wakeup_vcpus_on_cpu, cpu),
 			    pi_wakeup_list) {
 
 		if (pi_test_on(&vmx->pi_desc))
 			kvm_vcpu_wake_up(&vmx->vcpu);
 	}
-	spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
+	raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
 }
 
 void __init pi_init_cpu(int cpu)
 {
 	INIT_LIST_HEAD(&per_cpu(wakeup_vcpus_on_cpu, cpu));
-	spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
+	raw_spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
 }
 
 bool pi_has_pending_interrupt(struct kvm_vcpu *vcpu)
#1 lore raw
Quoted 25 lines · Marcelo Tosatti

wakeup_vcpus_on_cpu_lock is taken from hard interrupt context (pi_wakeup_handler), therefore it cannot sleep.

Switch it to a raw spinlock.

Fixes:

[41297.066254] BUG: scheduling while atomic: CPU 0/KVM/635218/0x00010001 
[41297.066323] Preemption disabled at: 
[41297.066324] [<ffffffff902ee47f>] irq_enter_rcu+0xf/0x60 
[41297.066339] Call Trace: 
[41297.066342]  <IRQ> 
[41297.066346]  dump_stack_lvl+0x34/0x44 
[41297.066353]  ? irq_enter_rcu+0xf/0x60 
[41297.066356]  __schedule_bug.cold+0x7d/0x8b 
[41297.066361]  __schedule+0x439/0x5b0 
[41297.066365]  ? task_blocks_on_rt_mutex.constprop.0.isra.0+0x1b0/0x440 
[41297.066369]  schedule_rtlock+0x1e/0x40 
[41297.066371]  rtlock_slowlock_locked+0xf1/0x260 
[41297.066374]  rt_spin_lock+0x3b/0x60 
[41297.066378]  pi_wakeup_handler+0x31/0x90 [kvm_intel] 
[41297.066388]  sysvec_kvm_posted_intr_wakeup_ipi+0x9d/0xd0 
[41297.066392]  </IRQ> 
[41297.066392]  asm_sysvec_kvm_posted_intr_wakeup_ipi+0x12/0x20 
Quoted 3 lines

...

so I have here v5.16 and no wakeup_vcpus_on_cpu_lock. It was also not removed so this patch is not intended for a previous kernel. Also checked next-20220111 and no wakeup_vcpus_on_cpu_lock.

Sebastian

#2 lore raw
Quoted 28 lines · Sebastian Andrzej Siewior
Quoted 27 lines · Marcelo Tosatti

wakeup_vcpus_on_cpu_lock is taken from hard interrupt context (pi_wakeup_handler), therefore it cannot sleep.

Switch it to a raw spinlock.

Fixes:

[41297.066254] BUG: scheduling while atomic: CPU 0/KVM/635218/0x00010001 
[41297.066323] Preemption disabled at: 
[41297.066324] [<ffffffff902ee47f>] irq_enter_rcu+0xf/0x60 
[41297.066339] Call Trace: 
[41297.066342]  <IRQ> 
[41297.066346]  dump_stack_lvl+0x34/0x44 
[41297.066353]  ? irq_enter_rcu+0xf/0x60 
[41297.066356]  __schedule_bug.cold+0x7d/0x8b 
[41297.066361]  __schedule+0x439/0x5b0 
[41297.066365]  ? task_blocks_on_rt_mutex.constprop.0.isra.0+0x1b0/0x440 
[41297.066369]  schedule_rtlock+0x1e/0x40 
[41297.066371]  rtlock_slowlock_locked+0xf1/0x260 
[41297.066374]  rt_spin_lock+0x3b/0x60 
[41297.066378]  pi_wakeup_handler+0x31/0x90 [kvm_intel] 
[41297.066388]  sysvec_kvm_posted_intr_wakeup_ipi+0x9d/0xd0 
[41297.066392]  </IRQ> 
[41297.066392]  asm_sysvec_kvm_posted_intr_wakeup_ipi+0x12/0x20 

...

Quoted 9 lines
Quoted 1 line

so I have here v5.16 and no wakeup_vcpus_on_cpu_lock. It was also not removed so this patch is not intended for a previous kernel. Also checked next-20220111 and no wakeup_vcpus_on_cpu_lock.

Sebastian

Sebastian,

The variable has been renamed on kvm.git/queue (sorry for not mentioning that).

I suppose Paolo can integrate through kvm.git.

maintainer
#3 lore raw
In reply to Sebastian Andrzej Siewior
Quoted 28 lines · Sebastian Andrzej Siewior
Quoted 27 lines · Marcelo Tosatti

wakeup_vcpus_on_cpu_lock is taken from hard interrupt context (pi_wakeup_handler), therefore it cannot sleep.

Switch it to a raw spinlock.

Fixes:

[41297.066254] BUG: scheduling while atomic: CPU 0/KVM/635218/0x00010001
[41297.066323] Preemption disabled at:
[41297.066324] [<ffffffff902ee47f>] irq_enter_rcu+0xf/0x60
[41297.066339] Call Trace:
[41297.066342]  <IRQ>
[41297.066346]  dump_stack_lvl+0x34/0x44
[41297.066353]  ? irq_enter_rcu+0xf/0x60
[41297.066356]  __schedule_bug.cold+0x7d/0x8b
[41297.066361]  __schedule+0x439/0x5b0
[41297.066365]  ? task_blocks_on_rt_mutex.constprop.0.isra.0+0x1b0/0x440
[41297.066369]  schedule_rtlock+0x1e/0x40
[41297.066371]  rtlock_slowlock_locked+0xf1/0x260
[41297.066374]  rt_spin_lock+0x3b/0x60
[41297.066378]  pi_wakeup_handler+0x31/0x90 [kvm_intel]
[41297.066388]  sysvec_kvm_posted_intr_wakeup_ipi+0x9d/0xd0
[41297.066392]  </IRQ>
[41297.066392]  asm_sysvec_kvm_posted_intr_wakeup_ipi+0x12/0x20

...

Quoted 5 lines
Quoted 1 line

so I have here v5.16 and no wakeup_vcpus_on_cpu_lock. It was also not removed so this patch is not intended for a previous kernel. Also checked next-20220111 and no wakeup_vcpus_on_cpu_lock.

Since this patch is good for stable@ too, I did the backport myself and queued Marcelo's version through a merge commit. You can find it now in kvm.git's master (5.16 version) and next branches (5.17 version).

kvm/queue has been rebased on top as well.

Paolo

maintainer Applied
#4 lore raw
In reply to Marcelo Tosatti
Quoted 82 lines · Marcelo Tosatti

wakeup_vcpus_on_cpu_lock is taken from hard interrupt context (pi_wakeup_handler), therefore it cannot sleep.

Switch it to a raw spinlock.

Fixes:

[41297.066254] BUG: scheduling while atomic: CPU 0/KVM/635218/0x00010001
[41297.066323] Preemption disabled at:
[41297.066324] [<ffffffff902ee47f>] irq_enter_rcu+0xf/0x60
[41297.066339] Call Trace:
[41297.066342]  <IRQ>
[41297.066346]  dump_stack_lvl+0x34/0x44
[41297.066353]  ? irq_enter_rcu+0xf/0x60
[41297.066356]  __schedule_bug.cold+0x7d/0x8b
[41297.066361]  __schedule+0x439/0x5b0
[41297.066365]  ? task_blocks_on_rt_mutex.constprop.0.isra.0+0x1b0/0x440
[41297.066369]  schedule_rtlock+0x1e/0x40
[41297.066371]  rtlock_slowlock_locked+0xf1/0x260
[41297.066374]  rt_spin_lock+0x3b/0x60
[41297.066378]  pi_wakeup_handler+0x31/0x90 [kvm_intel]
[41297.066388]  sysvec_kvm_posted_intr_wakeup_ipi+0x9d/0xd0
[41297.066392]  </IRQ>
[41297.066392]  asm_sysvec_kvm_posted_intr_wakeup_ipi+0x12/0x20

...

arch/x86/kvm/vmx/posted_intr.c +7 -7
index f4169c009400..aa1fe9085d77 100644
@@ -27,7 +27,7 @@ static DEFINE_PER_CPU(struct list_head, wakeup_vcpus_on_cpu);
   * CPU.  IRQs must be disabled when taking this lock, otherwise deadlock will
   * occur if a wakeup IRQ arrives and attempts to acquire the lock.
   */
-static DEFINE_PER_CPU(spinlock_t, wakeup_vcpus_on_cpu_lock);
+static DEFINE_PER_CPU(raw_spinlock_t, wakeup_vcpus_on_cpu_lock);
 
  static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
  {
@@ -87,9 +87,9 @@ void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
  	 * current pCPU if the task was migrated.
  	 */
  	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) {
-		spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+		raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
  		list_del(&vmx->pi_wakeup_list);
-		spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+		raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
  	}
 
  	dest = cpu_physical_id(cpu);
@@ -149,10 +149,10 @@ static void pi_enable_wakeup_handler(struct kvm_vcpu *vcpu)
 
  	local_irq_save(flags);
 
-	spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+	raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
  	list_add_tail(&vmx->pi_wakeup_list,
  		      &per_cpu(wakeup_vcpus_on_cpu, vcpu->cpu));
-	spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
+	raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu));
 
  	WARN(pi_desc->sn, "PI descriptor SN field set before blocking");
 
@@ -204,20 +204,20 @@ void pi_wakeup_handler(void)
  	int cpu = smp_processor_id();
  	struct vcpu_vmx *vmx;
 
-	spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
+	raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
  	list_for_each_entry(vmx, &per_cpu(wakeup_vcpus_on_cpu, cpu),
  			    pi_wakeup_list) {
 
  		if (pi_test_on(&vmx->pi_desc))
  			kvm_vcpu_wake_up(&vmx->vcpu);
  	}
-	spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
+	raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));

arch/x86/kvm/vmx/posted_intr.c void pi_wakeup_handler(void) · line 223

Quoted 10 lines
arch/x86/kvm/vmx/posted_intr.c +1 -1
  }
 
  void __init pi_init_cpu(int cpu)
  {
  	INIT_LIST_HEAD(&per_cpu(wakeup_vcpus_on_cpu, cpu));
-	spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
+	raw_spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu));
  }
 
  bool pi_has_pending_interrupt(struct kvm_vcpu *vcpu)

Queued, thanks.

Paolo