Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 23 Apr 2020 09:41:40 +0100
From:   Marc Zyngier <maz@kernel.org>
To:     Davidlohr Bueso <dave@stgolabs.net>
Cc:     tglx@linutronix.de, pbonzini@redhat.com, kvm@vger.kernel.org,
        Davidlohr Bueso <dbueso@suse.de>, peterz@infradead.org,
        torvalds@linux-foundation.org, bigeasy@linutronix.de,
        linux-kernel@vger.kernel.org, rostedt@goodmis.org,
        linux-mips@vger.kernel.org, Paul Mackerras <paulus@ozlabs.org>,
        joel@joelfernandes.org, will@kernel.org,
        kvmarm@lists.cs.columbia.edu
Subject: Re: [PATCH 4/5] kvm: Replace vcpu->swait with rcuwait
Message-ID: <20200423094140.69909bbb@why>
In-Reply-To: <20200422040739.18601-5-dave@stgolabs.net>
References: <20200422040739.18601-1-dave@stgolabs.net>
        <20200422040739.18601-5-dave@stgolabs.net>
Organization: Approximate
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, 21 Apr 2020 21:07:38 -0700
Davidlohr Bueso <dave@stgolabs.net> wrote:

> The use of any sort of waitqueue (simple or regular) for
> wait/waking vcpus has always been an overkill and semantically
> wrong. Because this is per-vcpu (which is blocked) there is
> only ever a single waiting vcpu, thus no need for any sort of
> queue.
> 
> As such, make use of the rcuwait primitive, with the following
> considerations:
> 
>   - rcuwait already provides the proper barriers that serialize
>   concurrent waiter and waker.
> 
>   - Task wakeup is done in rcu read critical region, with a
>   stable task pointer.
> 
>   - Because there is no concurrency among waiters, we need
>   not worry about rcuwait_wait_event() calls corrupting
>   the wait->task. As a consequence, this saves the locking
>   done in swait when modifying the queue. This also applies
>   to per-vcore wait for powerpc kvm-hv.
> 
> The x86 tscdeadline_latency test mentioned in 8577370fb0cb
> ("KVM: Use simple waitqueue for vcpu->wq") shows that, on avg,
> latency is reduced by around 15-20% with this change.
> 
> Cc: Paul Mackerras <paulus@ozlabs.org>
> Cc: kvmarm@lists.cs.columbia.edu
> Cc: linux-mips@vger.kernel.org
> Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
> ---
>  arch/mips/kvm/mips.c                  |  6 ++----
>  arch/powerpc/include/asm/kvm_book3s.h |  2 +-
>  arch/powerpc/include/asm/kvm_host.h   |  2 +-
>  arch/powerpc/kvm/book3s_hv.c          | 22 ++++++++--------------
>  arch/powerpc/kvm/powerpc.c            |  2 +-
>  arch/x86/kvm/lapic.c                  |  2 +-
>  include/linux/kvm_host.h              | 10 +++++-----
>  virt/kvm/arm/arch_timer.c             |  2 +-
>  virt/kvm/arm/arm.c                    |  9 +++++----
>  virt/kvm/async_pf.c                   |  3 +--
>  virt/kvm/kvm_main.c                   | 19 +++++++++----------
>  11 files changed, 35 insertions(+), 44 deletions(-)

[...]

I should have tested it *before* acking it, really.

> diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
> index 93bd59b46848..b2805105bbe5 100644
> --- a/virt/kvm/arm/arch_timer.c
> +++ b/virt/kvm/arm/arch_timer.c
> @@ -593,7 +593,7 @@ void kvm_timer_vcpu_put(struct kvm_vcpu *vcpu)
>  	if (map.emul_ptimer)
>  		soft_timer_cancel(&map.emul_ptimer->hrtimer);
>  
> -	if (swait_active(kvm_arch_vcpu_wq(vcpu)))
> +	if (rcu_dereference(kvm_arch_vpu_get_wait(vcpu)) != NULL)

This doesn't compile (wrong function name, and rcu_dereference takes a
variable). But whatever it would do if we fixed it looks dodgy. it isn't
the rcuwait structure that you want to dereference, but rcuwait->task
(we are checking whether we are called because we are blocking or being
preempted).

>  		kvm_timer_blocking(vcpu);
>  
>  	/*
> diff --git a/virt/kvm/arm/arm.c b/virt/kvm/arm/arm.c
> index 48d0ec44ad77..f94a10bb1251 100644
> --- a/virt/kvm/arm/arm.c
> +++ b/virt/kvm/arm/arm.c
> @@ -579,16 +579,17 @@ void kvm_arm_resume_guest(struct kvm *kvm)
>  
>  	kvm_for_each_vcpu(i, vcpu, kvm) {
>  		vcpu->arch.pause = false;
> -		swake_up_one(kvm_arch_vcpu_wq(vcpu));
> +		rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
>  	}
>  }
>  
>  static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
>  {
> -	struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
> +	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
>  
> -	swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
> -				       (!vcpu->arch.pause)));
> +	rcuwait_wait_event(*wait,
> +			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
> +			   TASK_INTERRUPTIBLE);

As noticed by the kbuild robot, this doesn't compile either.

I fixed it as follow, and it survived a very basic test run in a model
(more testing later).

Thanks,

	M.

diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
index b2805105bbe56..2dbd14dcae9fb 100644
--- a/virt/kvm/arm/arch_timer.c
+++ b/virt/kvm/arm/arch_timer.c
@@ -569,6 +569,7 @@ bool kvm_timer_should_notify_user(struct kvm_vcpu *vcpu)
 
 void kvm_timer_vcpu_put(struct kvm_vcpu *vcpu)
 {
+	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
 	struct arch_timer_cpu *timer = vcpu_timer(vcpu);
 	struct timer_map map;
 
@@ -593,7 +594,7 @@ void kvm_timer_vcpu_put(struct kvm_vcpu *vcpu)
 	if (map.emul_ptimer)
 		soft_timer_cancel(&map.emul_ptimer->hrtimer);
 
-	if (rcu_dereference(kvm_arch_vpu_get_wait(vcpu)) != NULL)
+	if (rcu_dereference(wait->task))
 		kvm_timer_blocking(vcpu);
 
 	/*
diff --git a/virt/kvm/arm/arm.c b/virt/kvm/arm/arm.c
index f94a10bb1251b..479f36d02418d 100644
--- a/virt/kvm/arm/arm.c
+++ b/virt/kvm/arm/arm.c
@@ -587,7 +587,7 @@ static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 {
 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
 
-	rcuwait_wait_event(*wait,
+	rcuwait_wait_event(wait,
 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
 			   TASK_INTERRUPTIBLE);
 

-- 
Jazz is not dead. It just smells funny...