From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Auger Subject: Re: [PATCH 0/2] irq: add get_gsi callback Date: Fri, 24 Apr 2015 11:01:11 +0200 Message-ID: <553A0657.6050100@linaro.org> References: <1429778977-1632-1-git-send-email-eric.auger@linaro.org> <5538BBD3.9040404@redhat.com> <5538BE19.6030009@linaro.org> <5538C246.20304@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from localhost (localhost [127.0.0.1]) by mm01.cs.columbia.edu (Postfix) with ESMTP id 624CA4E638 for ; Fri, 24 Apr 2015 04:55:35 -0400 (EDT) Received: from mm01.cs.columbia.edu ([127.0.0.1]) by localhost (mm01.cs.columbia.edu [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id EC5P1RWEjiM0 for ; Fri, 24 Apr 2015 04:55:34 -0400 (EDT) Received: from mail-wi0-f182.google.com (mail-wi0-f182.google.com [209.85.212.182]) by mm01.cs.columbia.edu (Postfix) with ESMTPS id 2D86E4E634 for ; Fri, 24 Apr 2015 04:55:33 -0400 (EDT) Received: by wiun10 with SMTP id n10so12784737wiu.1 for ; Fri, 24 Apr 2015 02:03:48 -0700 (PDT) In-Reply-To: <5538C246.20304@redhat.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: kvmarm-bounces@lists.cs.columbia.edu Sender: kvmarm-bounces@lists.cs.columbia.edu To: Paolo Bonzini , eric.auger@st.com, qemu-devel@nongnu.org, alex.williamson@redhat.com, peter.maydell@linaro.org, agraf@suse.de Cc: kvmarm@lists.cs.columbia.edu, patches@linaro.org List-Id: kvmarm@lists.cs.columbia.edu On 04/23/2015 11:58 AM, Paolo Bonzini wrote: > > > On 23/04/2015 11:40, Eric Auger wrote: >>>> Why can't you just use a GHashTable? >> You mean implementing this hash table in the interrupt controller? > > No, in KVM. Basically the kvm-vgic interrupt controller registers its > qemu_irqs with kvm-all.c, passing the gsi number for KVM_IRQFD along > the way. Then VFIO device can pass the qemu_irq and ask for an irqfd > for that qemu_irq, instead of having to know the gsi number. kvm-all.c > acts as a mediator. > > The notifier actually is not even necessary, because we already have > a very similar concept in QOM. But qdev_init_gpio_out_named always > passes a dummy "notifier": > > void qdev_init_gpio_out_named(DeviceState *dev, qemu_irq *pins, > const char *name, int n) > { > int i; > NamedGPIOList *gpio_list = qdev_get_named_gpio_list(dev, name); > char *propname = g_strdup_printf("%s[*]", name ? name : "unnamed-gpio-out"); > > assert(gpio_list->num_in == 0 || !name); > gpio_list->num_out += n; > > for (i = 0; i < n; ++i) { > memset(&pins[i], 0, sizeof(*pins)); > object_property_add_link(OBJECT(dev), propname, TYPE_IRQ, > (Object **)&pins[i], > object_property_allow_set_link, > OBJ_PROP_LINK_UNREF_ON_RELEASE, > &error_abort); > } > g_free(propname); > } > > So you need to add a method to DeviceClass, with the same prototype > as object_property_allow_set_link, and set the default value to > object_property_allow_set_link in device_class_init. Then > qdev_init_gpio_out_named can do > > DeviceClass *dc = DEVICE_GET_CLASS(dev); > ... > object_property_add_link(OBJECT(dev), propname, TYPE_IRQ, > (Object **)&pins[i], > dc->gpio_set_hook, > OBJ_PROP_LINK_UNREF_ON_RELEASE, > &error_abort); > > or something like that. > > Alternatively, and probably better, add the argument to > qdev_init_gpio_out_named; there aren't many calls. Then > you can add the method only to SysbusDeviceClass rather than > to all devices, and you can change sysbus_init_irq like this: > > void sysbus_init_irq(SysBusDevice *dev, qemu_irq *p) > { > SysBusDeviceClass *sdc = SYSBUS_DEVICE_GET_CLASS(dev); > qdev_init_gpio_out_named(DEVICE(dev), p, SYSBUS_DEVICE_GPIO_IRQ, 1, > sdc->irq_set_hook); > } Hi Paolo I implemented this alternative but my concern is the check method is called before the qemu_irq setting. So on this callback I cannot retrieve the qemu_irq VFIOINTp struct container object needed to setup irqfd hence does not work for me. I would need a post_check cb. Do you think it it sensible to add another cb? Thanks Eric > > > Paolo > >> The problem is my VFIO device currently has no link to the interrupt >> controller object. Besides it has a platform bus in between. Previously >> I devised a reset notifier called in the machine file to pass this link >> but Alex was not keen about this method. So the idea here is when the >> VFIO sysbus device qemuirq is getting connected, a notifier is called, >> VFIO device can retrieve the gsi of this qemu_irq and call KVM_IRQFD >> with that gsi.