So why is the rcu_virt_note_context_switch(smp_processor_id());
in guest_enter_irqoff not good enough?
This was actually supposed to tell rcu that being in the guest
is an extended quiescing period (like userspace).
What has changed?
On 07/11/2018 07:03 PM, David Woodhouse wrote:
> On Wed, 2018-07-11 at 09:49 -0700, Paul E. McKenney wrote:
>> And here is an updated v4.15 patch with Marius's Reported-by and David's
>> fix to my lost exclamation point.
> Thanks. Are you sending the original version of that to Linus? It'd be
> useful to have the commit ID so that we can watch for it landing, and
> chase this one up to Greg.
> As discussed on IRC, this patch reduces synchronize_sched() latency for
> us from ~4600s to ~160ms, which is nice.
> However, it isn't going to be sufficient in the NO_HZ_FULL case. For
> that you want a patch like the one below, which happily reduces the
> latency in our (!NO_HZ_FULL) case still further to ~40ms.
> Adding kvm list for better review...
> From: David Woodhouse <d...@amazon.co.uk>
> Subject: [PATCH] kvm/x86: Inform RCU of quiescent state when entering guest
> RCU can spend long periods of time waiting for a CPU which is actually in
> KVM guest mode, entirely pointlessly. Treat it like the idle and userspace
> modes, and don't wait for it.
> Signed-off-by: David Woodhouse <d...@amazon.co.uk>
> arch/x86/kvm/x86.c | 2 ++
> include/linux/rcutree.h | 2 ++
> kernel/rcu/tree.c | 16 ++++++++++++++++
> 3 files changed, 20 insertions(+)
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 0046aa70205a..b0c82f70afa7 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -7458,7 +7458,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
> vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
> + rcu_kvm_enter();
> + rcu_kvm_exit();
> * Do this here before restoring debug registers on the host. And
> diff --git a/include/linux/rcutree.h b/include/linux/rcutree.h
> index 914655848ef6..6d07af5a50fc 100644
> --- a/include/linux/rcutree.h
> +++ b/include/linux/rcutree.h
> @@ -82,6 +82,8 @@ void cond_synchronize_sched(unsigned long oldstate);
> void rcu_idle_enter(void);
> void rcu_idle_exit(void);
> +void rcu_kvm_enter(void);
> +void rcu_kvm_exit(void);
> void rcu_irq_enter(void);
> void rcu_irq_exit(void);
> void rcu_irq_enter_irqson(void);
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index aa7cade1b9f3..df7893273939 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -1019,6 +1019,22 @@ void rcu_irq_enter_irqson(void)
> + * These are currently identical to the _idle_ versions but let's
> + * explicitly have separate copies to keep Paul honest in future.
> + */
> +void rcu_kvm_enter(void)
> + rcu_idle_enter();
> +void rcu_kvm_exit(void)
> + rcu_idle_exit();
> * rcu_is_watching - see if RCU thinks that the current CPU is idle