Message ID | 20230313111022.13793-1-yan.y.zhao@intel.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [v3] KVM: VMX: fix lockdep warning on posted intr wakeup | expand |
On Mon, Mar 13, 2023, Yan Zhao wrote: > The lock ordering after this patch are: > - &p->pi_lock --> &rq->__lock --> > &per_cpu(wakeup_vcpus_on_cpu_lock_out, cpu) > - &per_cpu(wakeup_vcpus_on_cpu_lock_in, cpu) --> > &per_cpu(wakeup_vcpus_on_cpu_lock_out, cpu) > - &per_cpu(wakeup_vcpus_on_cpu_lock_in, cpu) --> &p->pi_lock > > Currently, &rq->__lock is not held in "path sched_in". > However, if in future "path sched_in" takes &p->pi_lock or &rq->__lock, > lockdep is able to detect and warn in that case. > > Signed-off-by: Yan Zhao <yan.y.zhao@intel.com> > [sean: path sched_out and path irq does not race, path sched_in does not > take &rq->__lock] But there's no actual deadlock, right? I have zero interest in fixing a lockdep false positive by making functional changes to KVM. I am definitely open to making changes to somehow let lockdep know what's going on, but complicating KVM's actual functionality is too much.
On Fri, Mar 24, 2023 at 04:13:37PM -0700, Sean Christopherson wrote: > On Mon, Mar 13, 2023, Yan Zhao wrote: > > The lock ordering after this patch are: > > - &p->pi_lock --> &rq->__lock --> > > &per_cpu(wakeup_vcpus_on_cpu_lock_out, cpu) > > - &per_cpu(wakeup_vcpus_on_cpu_lock_in, cpu) --> > > &per_cpu(wakeup_vcpus_on_cpu_lock_out, cpu) > > - &per_cpu(wakeup_vcpus_on_cpu_lock_in, cpu) --> &p->pi_lock > > > > Currently, &rq->__lock is not held in "path sched_in". > > However, if in future "path sched_in" takes &p->pi_lock or &rq->__lock, > > lockdep is able to detect and warn in that case. > > > > Signed-off-by: Yan Zhao <yan.y.zhao@intel.com> > > [sean: path sched_out and path irq does not race, path sched_in does not > > take &rq->__lock] > > But there's no actual deadlock, right? I have zero interest in fixing a lockdep > false positive by making functional changes to KVM. I am definitely open to making > changes to somehow let lockdep know what's going on, but complicating KVM's actual > functionality is too much. Yes, there's no actual deadlock currently. But without fixing this issue, debug_locks will be set to false along with below messages printed. Then lockdep will be turned off and any other lock detections like lockdep_assert_held()... will not print warning even when it's obviously violated. [ 118.873499] ====================================================== [ 118.880413] WARNING: possible circular locking dependency detected [ 118.887325] 6.2.0-rc5+ #600 Not tainted [ 118.891613] ------------------------------------------------------ [ 118.898519] swapper/5/0 is trying to acquire lock: [ 118.903869] ffff88810f5cac90 (&p->pi_lock){-.-.}-{2:2}, at: try_to_wake_up+0xbb/0x510 [ 118.912624] [ 118.912624] but task is already holding lock: [ 118.919138] ffff88885f7fdab8 (&per_cpu(wakeup_vcpus_on_cpu_lock, cpu)){-...}-{2:2}, at: pi_wakeup_handler+0x36/0x80 [kvm_intel] [ 118.931962] [ 118.931962] which lock already depends on the new lock. Any suggestion?
On 3/29/23 03:53, Yan Zhao wrote: > Yes, there's no actual deadlock currently. > > But without fixing this issue, debug_locks will be set to false along > with below messages printed. Then lockdep will be turned off and any > other lock detections like lockdep_assert_held()... will not print > warning even when it's obviously violated. Can you use lockdep subclasses, giving 0 to the sched_in path and 1 to the sched_out path? Paolo
On Wed, Mar 29, 2023 at 01:51:23PM +0200, Paolo Bonzini wrote: > On 3/29/23 03:53, Yan Zhao wrote: > > Yes, there's no actual deadlock currently. > > > > But without fixing this issue, debug_locks will be set to false along > > with below messages printed. Then lockdep will be turned off and any > > other lock detections like lockdep_assert_held()... will not print > > warning even when it's obviously violated. > > Can you use lockdep subclasses, giving 0 to the sched_in path and 1 to the > sched_out path? Yes, thanks for the suggestion! This can avoid this warning of "possible circular locking dependency". I tried it like this: - in sched_out path: raw_spin_lock_nested(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu), 1); - in irq and sched_in paths: raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); But I have a concern: If sched_in path removes vcpu A from wakeup list of its previous pcpu A, and at the mean time, sched_out path adds vcpu B to the wakeup list of pcpu A, the sched_in and sched_out paths should race for the same subclass of lock. But if sched_in path only holds subclass 0, and sched_out path holds subclass 1, then lockdep would not warn of "possible circular locking dependency" if someone made a change as below in sched_in path. if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); list_del(&vmx->pi_wakeup_list); + raw_spin_lock(¤t->pi_lock); + raw_spin_unlock(¤t->pi_lock); raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); } While with v3 of this patch (sched_in path holds both out_lock and in_lock), lockdep is still able to warn about this issue. Thanks Yan
On Thu, Mar 30, 2023, Yan Zhao wrote: > On Wed, Mar 29, 2023 at 01:51:23PM +0200, Paolo Bonzini wrote: > > On 3/29/23 03:53, Yan Zhao wrote: > > > Yes, there's no actual deadlock currently. > > > > > > But without fixing this issue, debug_locks will be set to false along > > > with below messages printed. Then lockdep will be turned off and any > > > other lock detections like lockdep_assert_held()... will not print > > > warning even when it's obviously violated. > > > > Can you use lockdep subclasses, giving 0 to the sched_in path and 1 to the > > sched_out path? > > Yes, thanks for the suggestion! > This can avoid this warning of "possible circular locking dependency". > > I tried it like this: > - in sched_out path: > raw_spin_lock_nested(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu), 1); > > - in irq and sched_in paths: > raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > > But I have a concern: > If sched_in path removes vcpu A from wakeup list of its previous pcpu A, > and at the mean time, sched_out path adds vcpu B to the wakeup list of > pcpu A, the sched_in and sched_out paths should race for the same > subclass of lock. > But if sched_in path only holds subclass 0, and sched_out path holds > subclass 1, then lockdep would not warn of "possible circular locking > dependency" if someone made a change as below in sched_in path. > > if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { > raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > list_del(&vmx->pi_wakeup_list); > + raw_spin_lock(¤t->pi_lock); > + raw_spin_unlock(¤t->pi_lock); > raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > } > > While with v3 of this patch (sched_in path holds both out_lock and in_lock), > lockdep is still able to warn about this issue. Couldn't we just add a manual assertion? That'd also be a good location for a comment to document all of this, and to clarify that current->pi_lock is a completely different lock that has nothing to do with posted interrupts. It's not foolproof, but any patches that substantially touch this code need a ton of scrutiny as the scheduling interactions are gnarly, i.e. IMO a deadlock bug sneaking in is highly unlikely. diff --git a/arch/x86/kvm/vmx/posted_intr.c b/arch/x86/kvm/vmx/posted_intr.c index 94c38bea60e7..19325a10e42f 100644 --- a/arch/x86/kvm/vmx/posted_intr.c +++ b/arch/x86/kvm/vmx/posted_intr.c @@ -90,6 +90,7 @@ void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu) */ if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); + lockdep_assert_not_held(¤t->pi_lock); list_del(&vmx->pi_wakeup_list); raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); }
On Thu, Mar 30, 2023 at 11:14:27AM -0700, Sean Christopherson wrote: > On Thu, Mar 30, 2023, Yan Zhao wrote: > > On Wed, Mar 29, 2023 at 01:51:23PM +0200, Paolo Bonzini wrote: > > > On 3/29/23 03:53, Yan Zhao wrote: > > > > Yes, there's no actual deadlock currently. > > > > > > > > But without fixing this issue, debug_locks will be set to false along > > > > with below messages printed. Then lockdep will be turned off and any > > > > other lock detections like lockdep_assert_held()... will not print > > > > warning even when it's obviously violated. > > > > > > Can you use lockdep subclasses, giving 0 to the sched_in path and 1 to the > > > sched_out path? > > > > Yes, thanks for the suggestion! > > This can avoid this warning of "possible circular locking dependency". > > > > I tried it like this: > > - in sched_out path: > > raw_spin_lock_nested(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu), 1); > > > > - in irq and sched_in paths: > > raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > > > > But I have a concern: > > If sched_in path removes vcpu A from wakeup list of its previous pcpu A, > > and at the mean time, sched_out path adds vcpu B to the wakeup list of > > pcpu A, the sched_in and sched_out paths should race for the same > > subclass of lock. > > But if sched_in path only holds subclass 0, and sched_out path holds > > subclass 1, then lockdep would not warn of "possible circular locking > > dependency" if someone made a change as below in sched_in path. > > > > if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { > > raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > > list_del(&vmx->pi_wakeup_list); > > + raw_spin_lock(¤t->pi_lock); > > + raw_spin_unlock(¤t->pi_lock); > > raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > > } > > > > While with v3 of this patch (sched_in path holds both out_lock and in_lock), > > lockdep is still able to warn about this issue. > > Couldn't we just add a manual assertion? That'd also be a good location for a > comment to document all of this, and to clarify that current->pi_lock is a > completely different lock that has nothing to do with posted interrupts. > > It's not foolproof, but any patches that substantially touch this code need a > ton of scrutiny as the scheduling interactions are gnarly, i.e. IMO a deadlock > bug sneaking in is highly unlikely. > > diff --git a/arch/x86/kvm/vmx/posted_intr.c b/arch/x86/kvm/vmx/posted_intr.c > index 94c38bea60e7..19325a10e42f 100644 > --- a/arch/x86/kvm/vmx/posted_intr.c > +++ b/arch/x86/kvm/vmx/posted_intr.c > @@ -90,6 +90,7 @@ void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu) > */ > if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { > raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > + lockdep_assert_not_held(¤t->pi_lock); > list_del(&vmx->pi_wakeup_list); > raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > } Hmm...No. It's not about "current->pi_lock" cannot be held, it's about the lock ordering. In sched_out thread, the lock ordering is "current->pi_lock" --> "rq->__lock" --> "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)", then in sched_in thread, if the lock ordering is "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)" --> "current->pi_lock", circular locking dependency will happen. while if the lock ordering in sched_in thread is "current->pi_lock" --> "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)", it's fine! If sched_out thread and sched_in thread actually should hold the same subclass of lock, we can't fool the lockdep just to let it shut up. And, we may not be able to list or document out all potential locks that cannot be held inside the "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)", right? BTW, could you tell me why you think v3 complicates KVM's functionality? It just splits a single lock into two sub locks, and let irq path only takes in_lock, sched_out path only takes out_lock, while sched_in path takes both in_lock and out_lock. IMHO, it does not make any functional change to KVM code. Maybe it's because the commit message is not well written and gave people a wrong impression that the logic changes a lot? Thanks Yan
On Fri, Mar 31, 2023, Yan Zhao wrote: > On Thu, Mar 30, 2023 at 11:14:27AM -0700, Sean Christopherson wrote: > > On Thu, Mar 30, 2023, Yan Zhao wrote: > > > While with v3 of this patch (sched_in path holds both out_lock and in_lock), > > > lockdep is still able to warn about this issue. > > > > Couldn't we just add a manual assertion? That'd also be a good location for a > > comment to document all of this, and to clarify that current->pi_lock is a > > completely different lock that has nothing to do with posted interrupts. > > > > It's not foolproof, but any patches that substantially touch this code need a > > ton of scrutiny as the scheduling interactions are gnarly, i.e. IMO a deadlock > > bug sneaking in is highly unlikely. > > > > diff --git a/arch/x86/kvm/vmx/posted_intr.c b/arch/x86/kvm/vmx/posted_intr.c > > index 94c38bea60e7..19325a10e42f 100644 > > --- a/arch/x86/kvm/vmx/posted_intr.c > > +++ b/arch/x86/kvm/vmx/posted_intr.c > > @@ -90,6 +90,7 @@ void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu) > > */ > > if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { > > raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > > + lockdep_assert_not_held(¤t->pi_lock); > > list_del(&vmx->pi_wakeup_list); > > raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > > } > Hmm...No. It's not about "current->pi_lock" cannot be held, it's about > the lock ordering. > In sched_out thread, the lock ordering is > "current->pi_lock" --> "rq->__lock" --> "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)", > then in sched_in thread, if the lock ordering is > "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)" --> "current->pi_lock", > circular locking dependency will happen. > while if the lock ordering in sched_in thread is > "current->pi_lock" --> "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)", > it's fine! Right, but IIUC, neither ordering happens today. In other words, the lockdep assertion isn't defining a hard rule, rather it's enforcing an assumption that KVM relies on to avoid a potential deadlock. > If sched_out thread and sched_in thread actually should hold the same > subclass of lock, we can't fool the lockdep just to let it shut up. > And, we may not be able to list or document out all potential locks that cannot > be held inside the "per_cpu(wakeup_vcpus_on_cpu_lock, cpu)", right? Eh, IMO this is a non-issue. It's a raw_spin_lock() in an IRQs disabled section that wraps a single line of benign code. If it's really concerning, we could add a scary comment like this. diff --git a/arch/x86/kvm/vmx/posted_intr.c b/arch/x86/kvm/vmx/posted_intr.c index 94c38bea60e7..a7ec0371aeca 100644 --- a/arch/x86/kvm/vmx/posted_intr.c +++ b/arch/x86/kvm/vmx/posted_intr.c @@ -87,6 +87,12 @@ void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu) * If the vCPU was waiting for wakeup, remove the vCPU from the wakeup * list of the _previous_ pCPU, which will not be the same as the * current pCPU if the task was migrated. + * + * Stating the obvious, do not under any circumstance let this path + * acquire a different lock while holding the per-CPU lock. To avoid + * false postives in lockdep, KVM uses different lockdep subclasses for + * vmx_vcpu_pi_put() vs vmx_vcpu_pi_load(), i.e. lockdep may not be + * to detect circular dependencies and other issues. */ if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR) { raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); > BTW, could you tell me why you think v3 complicates KVM's functionality? > It just splits a single lock into two sub locks, and let irq path only Heh, "just". > takes in_lock, sched_out path only takes out_lock, while sched_in path takes > both in_lock and out_lock. > IMHO, it does not make any functional change to KVM code. > Maybe it's because the commit message is not well written and gave people a wrong > impression that the logic changes a lot? No, this is not a problem that can be solved by any changelog. My very strong objection to having two separate locks is that when reading the code, it's not remotely obvious why two locks are needed, or that the code is correct. Adding copious amounts of documentation/comments can help, but it'll never fully solve the problem that having two locks simply isn't intuitive/straightforward.
diff --git a/arch/x86/kvm/vmx/posted_intr.c b/arch/x86/kvm/vmx/posted_intr.c index 94c38bea60e7..f4a5fcb751c1 100644 --- a/arch/x86/kvm/vmx/posted_intr.c +++ b/arch/x86/kvm/vmx/posted_intr.c @@ -23,13 +23,19 @@ */ static DEFINE_PER_CPU(struct list_head, wakeup_vcpus_on_cpu); /* - * Protect the per-CPU list with a per-CPU spinlock to handle task migration. + * Protect the per-CPU list with two per-CPU spinlocks to handle task migration. + * IRQs must be disabled when taking the two locks, otherwise deadlock will + * occur if a wakeup IRQ arrives and attempts to acquire the locks. + * ->sched_out() path before a vCPU blocking takes the "out lock", which will not + * be taken in the wakeup IRQ handler that running at the same pCPU as the + * ->sched_out() path. * When a blocking vCPU is awakened _and_ migrated to a different pCPU, the * ->sched_in() path will need to take the vCPU off the list of the _previous_ - * CPU. IRQs must be disabled when taking this lock, otherwise deadlock will - * occur if a wakeup IRQ arrives and attempts to acquire the lock. + * CPU. It takes both "in lock" and "out lock" to take care of list racing of the + * _previous_ CPU. */ -static DEFINE_PER_CPU(raw_spinlock_t, wakeup_vcpus_on_cpu_lock); +static DEFINE_PER_CPU(raw_spinlock_t, wakeup_vcpus_on_cpu_lock_in); +static DEFINE_PER_CPU(raw_spinlock_t, wakeup_vcpus_on_cpu_lock_out); static inline struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu) { @@ -89,9 +95,11 @@ 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) { - raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); + raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock_in, vcpu->cpu)); + raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock_out, vcpu->cpu)); list_del(&vmx->pi_wakeup_list); - raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); + raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock_out, vcpu->cpu)); + raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock_in, vcpu->cpu)); } dest = cpu_physical_id(cpu); @@ -152,10 +160,10 @@ static void pi_enable_wakeup_handler(struct kvm_vcpu *vcpu) local_irq_save(flags); - raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); + raw_spin_lock(&per_cpu(wakeup_vcpus_on_cpu_lock_out, vcpu->cpu)); list_add_tail(&vmx->pi_wakeup_list, &per_cpu(wakeup_vcpus_on_cpu, vcpu->cpu)); - raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock, vcpu->cpu)); + raw_spin_unlock(&per_cpu(wakeup_vcpus_on_cpu_lock_out, vcpu->cpu)); WARN(pi_desc->sn, "PI descriptor SN field set before blocking"); @@ -219,7 +227,7 @@ void pi_wakeup_handler(void) { int cpu = smp_processor_id(); struct list_head *wakeup_list = &per_cpu(wakeup_vcpus_on_cpu, cpu); - raw_spinlock_t *spinlock = &per_cpu(wakeup_vcpus_on_cpu_lock, cpu); + raw_spinlock_t *spinlock = &per_cpu(wakeup_vcpus_on_cpu_lock_in, cpu); struct vcpu_vmx *vmx; raw_spin_lock(spinlock); @@ -234,7 +242,8 @@ void pi_wakeup_handler(void) void __init pi_init_cpu(int cpu) { INIT_LIST_HEAD(&per_cpu(wakeup_vcpus_on_cpu, cpu)); - raw_spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock, cpu)); + raw_spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock_in, cpu)); + raw_spin_lock_init(&per_cpu(wakeup_vcpus_on_cpu_lock_out, cpu)); } bool pi_has_pending_interrupt(struct kvm_vcpu *vcpu)