Re: [Qemu-devel] [RFC v9 10/18] get all affected groups for each device support aer

[Alex Williamson <alex.williamson@redhat.com>]

On Tue, 2015-06-09 at 11:37 +0800, Chen Fan wrote:
> Add the affected groups without any devices into VM,
> it can keep the VM ownship the all groups. and use a
> reference to make the group visible.
> 
> Signed-off-by: Chen Fan <chen.fan.fnst@cn.fujitsu.com>
> ---
>  hw/vfio/pci.c | 115 ++++++++++++++++++++++++++++++++++++++++++++++++++++++----
>  1 file changed, 108 insertions(+), 7 deletions(-)
> 
> diff --git a/hw/vfio/pci.c b/hw/vfio/pci.c
> index 7a3fad7..06006ce 100644
> --- a/hw/vfio/pci.c
> +++ b/hw/vfio/pci.c
> @@ -2822,6 +2822,105 @@ static int vfio_add_std_cap(VFIOPCIDevice *vdev, uint8_t pos)
>      return 0;
>  }
>  
> +static bool vfio_pci_host_match(PCIHostDeviceAddress *host1,
> +                                PCIHostDeviceAddress *host2)
> +{
> +    return (host1->domain == host2->domain && host1->bus == host2->bus &&
> +            host1->slot == host2->slot && host1->function == host2->function);
> +}
> +
> +static int vfio_put_affected_groups(VFIOPCIDevice *vdev)
> +{
> +    struct vfio_pci_hot_reset_info *info = NULL;
> +    struct vfio_pci_dependent_device *devices;
> +    PCIHostDeviceAddress host;
> +    VFIOGroup *group;
> +    int ret, i;
> +
> +    ret = vfio_get_hot_reset_info(vdev, &info);
> +    if (ret) {
> +        goto out;
> +    }
> +
> +    /* List all affected devices by bus reset */
> +    devices = &info->devices[0];
> +
> +    /* Verify that we have all the groups required */
> +    for (i = 0; i < info->count; i++) {
> +        host.domain = devices[i].segment;
> +        host.bus = devices[i].bus;
> +        host.slot = PCI_SLOT(devices[i].devfn);
> +        host.function = PCI_FUNC(devices[i].devfn);
> +
> +        /* Skip the current device */
> +        if (vfio_pci_host_match(&host, &vdev->host)) {
> +            continue;
> +        }
> +
> +        /* Ensure we own the group of the affected device */
> +        QLIST_FOREACH(group, &vfio_group_list, next) {
> +            if (group->groupid == devices[i].group_id) {
> +                break;
> +            }
> +        }
> +
> +        if (!group)
> +            continue;
> +
> +        group->ref--;
> +        vfio_put_group(group);
> +    }
> +
> +    ret = 0;
> +out:
> +    g_free(info);
> +    return ret;
> +}
> +
> +static int vfio_get_affected_groups(VFIOPCIDevice *vdev)
> +{
> +    struct vfio_pci_hot_reset_info *info = NULL;
> +    struct vfio_pci_dependent_device *devices;
> +    PCIHostDeviceAddress host;
> +    VFIOGroup *group;
> +    int ret, i;
> +
> +    ret = vfio_get_hot_reset_info(vdev, &info);
> +    if (ret) {
> +        goto out;
> +    }
> +
> +    /* List all affected devices by bus reset */
> +    devices = &info->devices[0];
> +
> +    /* Verify that we have all the groups required */
> +    for (i = 0; i < info->count; i++) {
> +        host.domain = devices[i].segment;
> +        host.bus = devices[i].bus;
> +        host.slot = PCI_SLOT(devices[i].devfn);
> +        host.function = PCI_FUNC(devices[i].devfn);
> +
> +        /* Skip the current device */
> +        if (vfio_pci_host_match(&host, &vdev->host)) {
> +            continue;
> +        }
> +
> +        /* Get all affected groups into VM */
> +        group = vfio_get_group(devices[i].group_id, NULL);
> +        if (!group) {
> +            error_report("vfio: failed to get affected group %d",
> +                         devices[i].group_id);
> +            ret = -1;
> +            goto out;

There needs to be an unwind here, if the error occurs during hotplug,
we've just left QEMU in a state where some groups have extra references.

> +        }
> +        group->ref++;

If we're going to have a group reference, get_group should increment it
and put_group should decrement it, we shouldn't modify it externally
from those functions.

> +    }
> +    ret = 0;
> +out:
> +    g_free(info);
> +    return ret;
> +}
> +
>  static int vfio_setup_aer(VFIOPCIDevice *vdev, uint8_t cap_ver,
>                            int pos, uint16_t size)
>  {
> @@ -2858,6 +2957,12 @@ static int vfio_setup_aer(VFIOPCIDevice *vdev, uint8_t cap_ver,
>          dev_iter = pci_bridge_get_device(dev_iter->bus);
>      }
>  
> +    /* Ensure own all affected groups */
> +    ret = vfio_get_affected_groups(vdev);
> +    if (ret) {
> +        goto error;
> +    }
> +
>      errcap = vfio_pci_read_config(pdev, pdev->exp.aer_cap + PCI_ERR_CAP, 4);
>      /*
>       * The ability to record multiple headers is depending on
> @@ -3013,13 +3118,6 @@ static void vfio_pci_post_reset(VFIOPCIDevice *vdev)
>      vfio_enable_intx(vdev);
>  }
>  
> -static bool vfio_pci_host_match(PCIHostDeviceAddress *host1,
> -                                PCIHostDeviceAddress *host2)
> -{
> -    return (host1->domain == host2->domain && host1->bus == host2->bus &&
> -            host1->slot == host2->slot && host1->function == host2->function);
> -}
> -
>  static int vfio_pci_hot_reset(VFIOPCIDevice *vdev, bool single)
>  {
>      VFIOGroup *group;
> @@ -3851,6 +3949,9 @@ static void vfio_instance_finalize(Object *obj)
>      g_free(vdev->rom);
>      vfio_put_device(vdev);
>      vfio_put_group(group);
> +    if (vdev->features & VFIO_FEATURE_ENABLE_AER) {
> +        vfio_put_affected_groups(vdev);
> +    }
>  }
>  
>  static void vfio_exitfn(PCIDevice *pdev)