From mboxrd@z Thu Jan  1 00:00:00 1970
From: Avi Kivity <avi@redhat.com>
Subject: Re: [PATCHv2 RFC] kvm: deliver msix interrupts from irq handler
Date: Mon, 11 Jun 2012 14:53:23 +0300
Message-ID: <4FD5DC33.3060709@redhat.com>
References: <20120608160120.GA2959@redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Cc: kvm@vger.kernel.org
To: "Michael S. Tsirkin" <mst@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx1.redhat.com ([209.132.183.28]:32742 "EHLO mx1.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1751494Ab2FKLxZ (ORCPT <rfc822;kvm@vger.kernel.org>);
	Mon, 11 Jun 2012 07:53:25 -0400
Received: from int-mx01.intmail.prod.int.phx2.redhat.com (int-mx01.intmail.prod.int.phx2.redhat.com [10.5.11.11])
	by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id q5BBrPHS010477
	(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK)
	for <kvm@vger.kernel.org>; Mon, 11 Jun 2012 07:53:25 -0400
In-Reply-To: <20120608160120.GA2959@redhat.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On 06/08/2012 07:01 PM, Michael S. Tsirkin wrote:
> We can deliver certain interrupts, notably MSIX,
> from atomic context.  Add a new API kvm_set_irq_inatomic,
> that does exactly that, and use it to implement
> an irq handler for msi.
> 
> This reduces the pressure on scheduler in case
> where host and guest irq share a host cpu.
> 

Looks nice.

>  
> +/*
> + * Deliver an IRQ in an atomic context if we can, or return a failure,
> + * user can retry in a process context.
> + * Return value:
> + *  -EWOULDBLOCK - Can't deliver in atomic context: retry in a process context.
> + *  Other values - No need to retry.
> + */
> +int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level)
> +{
> +	struct kvm_kernel_irq_routing_entry *e;
> +	int ret = -EINVAL;
> +	struct kvm_irq_routing_table *irq_rt;
> +	struct hlist_node *n;
> +
> +	trace_kvm_set_irq(irq, level, irq_source_id);
> +
> +	/*
> +	 * We know MSI are safe in interrupt context. They are also
> +	 * easy as there's a single routing entry for these GSIs.
> +	 * So only handle MSI in an atomic context, for now.
> +	 *
> +	 * This shares some code with kvm_set_irq: this
> +	 * version is optimized for a single entry MSI only case.
> +	 */
> +	rcu_read_lock();
> +	irq_rt = rcu_dereference(kvm->irq_routing);
> +	if (irq < irq_rt->nr_rt_entries)
> +		hlist_for_each_entry(e, n, &irq_rt->map[irq], link) {
> +			if (likely(e->type == KVM_IRQ_ROUTING_MSI))
> +				ret = kvm_set_msi(e, kvm, irq_source_id, level);
> +			else
> +				ret = -EWOULDBLOCK;
> +			break;
> +		}
> +	rcu_read_unlock();
> +	return ret;
> +}
> +

kvm_set_msi() contains a loop over all vcpus to match the APIC ID (or to
broadcast).  I'd rather see a cache that contains the vcpu pointer and
vector number (assumes delivery mode of fixed and not the other junk),
if the cache is filled fire away, otherwise slow path to a thread and
fill the cache if possible.

-- 
error compiling committee.c: too many arguments to function