Re: [Patch v11 3/3] vt-d: fix vt-d Device-TLB flush timeout issue

["Jan Beulich" <JBeulich@suse.com>]

>>> On 01.06.16 at 11:05, <quan.xu@intel.com> wrote:
> --- a/xen/drivers/passthrough/vtd/extern.h
> +++ b/xen/drivers/passthrough/vtd/extern.h
> @@ -21,6 +21,7 @@
>  #define _VTD_EXTERN_H_
>  
>  #include "dmar.h"
> +#include "../ats.h"

Why? You don't de-reference struct pci_ats_dev * in this file, so
all you'd need is a forward declaration. But then this is not in line
with your v11 change description above, so I wonder whether
you actually sent a stale patch. After all I thought the v10
discussion (see
http://lists.xenproject.org/archives/html/xen-devel/2016-05/msg02208.html
) had made clear that this passing down, besides reducing the
number of arguments of some function, would also be meant to
eliminate ...

> +static void dev_invalidate_iotlb_timeout(struct iommu *iommu, u16 did,
> +                                         struct pci_ats_dev *ats_dev)
> +{
> +    struct domain *d = NULL;
> +    struct pci_dev *pdev;
> +
> +    if ( test_bit(did, iommu->domid_bitmap) )
> +        d = rcu_lock_domain_by_id(iommu->domid_map[did]);
> +
> +    /*
> +     * In case the domain has been freed or the IOMMU domid bitmap is
> +     * not valid, the device no longer belongs to this domain.
> +     */
> +    if ( d == NULL )
> +        return;
> +
> +    pcidevs_lock();
> +
> +    for_each_pdev(d, pdev)
> +    {
> +        if ( (pdev->seg == ats_dev->seg) &&
> +             (pdev->bus == ats_dev->bus) &&
> +             (pdev->devfn == ats_dev->devfn) )
> +        {
> +            ASSERT(pdev->domain);
> +            list_del(&pdev->domain_list);
> +            pdev->domain = NULL;
> +            pci_hide_existing_device(pdev);
> +            break;
> +        }
> +    }
> +
> +    pcidevs_unlock();

... this loop (and locking). (Of course such a change may better be
done in another preparatory patch.)

> +    if ( !is_hardware_domain(d) )
> +        domain_crash(d);
> +    else
> +        printk(XENLOG_WARNING VTDPREFIX
> +               " dom%d: ATS device %04x:%02x:%02x.%u flush failed\n",
> +               d->domain_id, ats_dev->seg, ats_dev->bus,
> +               PCI_SLOT(ats_dev->devfn), PCI_FUNC(ats_dev->devfn));

Please use the same logic for logging and crashing as you do in
the other series, so that at least on average a resulting DomU
crash will be accompanied with some indication of the reason
beyond just the source file name and line number.

> +static int __must_check dev_invalidate_sync(struct iommu *iommu, u16 did,
> +                                            struct pci_ats_dev *ats_dev)
> +{
> +    struct qi_ctrl *qi_ctrl = iommu_qi_ctrl(iommu);
> +    int rc = 0;
> +
> +    if ( qi_ctrl->qinval_maddr )
> +    {
> +        rc = queue_invalidate_wait(iommu, 0, 1, 1, 1);
> +
> +        if ( rc == -ETIMEDOUT )
> +            dev_invalidate_iotlb_timeout(iommu, did, ats_dev);
> +    }
> +
> +    return rc;
> +}

I've never really understood why invalidate_sync() returns success
when it didn't do anything. Now that you copy this same behavior
here, I really need to ask you to explain that.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel