Re: [Qemu-devel] [RFC v4 6/9] vfio: add 'x-aer' option to disable aer capability

[Alex Williamson]

On Wed, 2015-03-11 at 11:42 +0800, Chen Fan wrote:
> On 03/10/2015 04:29 AM, Alex Williamson wrote:
> > On Mon, 2015-03-02 at 15:16 +0800, Chen Fan wrote:
> >> add 'x-aer' option to disable aer capability if user
> >> want.
> > I'm generally one to favor using the x- flag, but we need to figure out
> > if we need to make this be a supported option or not.  We also need to
> > decide whether we should include your previous patch that toggled this
> > support based on machine type.  We don't care about migration
> > compatibility, but users might depend on specific error behavior and we
> > don't want to break that for existing machine types.  This would include
> > only the q35 machine types through QEMU 2.3, I believe.
> >
> > My impression is that it should probably be supported.  There may be
> > cases where the guest OS does not support AER and we want to be able to
> > invoke the old vm_stop() behavior.
> In order to compatible with existing machine type behavior,  I think
> we should let user decide whether need to enable aer on qemu.
> by default disabled. if the guest OS support AER, user may
> enable aer, otherwise, also invoke the vm_stop() behavior.

Yes, for existing machine types, the current behavior needs to be
maintained.  The user can be given the option to enable AER and override
the machine setting.  For new machine types, I expect we'd want to
enable AER automatically with the option to disable it.  This would be
for Q35 or other express chipsets only though since we can't expose AER
to the guest for devices exposed on a conventional bus.  Thanks,

Alex

> >> Signed-off-by: Chen Fan <address@hidden>
> >> ---
> >>   hw/vfio/pci.c | 8 ++++++++
> >>   1 file changed, 8 insertions(+)
> >>
> >> diff --git a/hw/vfio/pci.c b/hw/vfio/pci.c
> >> index db4ba23..5471437 100644
> >> --- a/hw/vfio/pci.c
> >> +++ b/hw/vfio/pci.c
> >> @@ -156,6 +156,8 @@ typedef struct VFIOPCIDevice {
> >>       uint32_t features;
> >>   #define VFIO_FEATURE_ENABLE_VGA_BIT 0
> >>   #define VFIO_FEATURE_ENABLE_VGA (1 << VFIO_FEATURE_ENABLE_VGA_BIT)
> >> +#define VFIO_FEATURE_ENABLE_AER_BIT 1
> >> +#define VFIO_FEATURE_ENABLE_AER (1 << VFIO_FEATURE_ENABLE_AER_BIT)
> > Bit 1 is already taken by:
> >
> > 47cbe50 vfio-pci: Enable device request notification support
> >
> >>       int32_t bootindex;
> >>       uint8_t pm_cap;
> >>       bool has_vga;
> >> @@ -2728,6 +2730,10 @@ static int vfio_setup_aer(VFIOPCIDevice *vdev, 
> >> uint8_t cap_ver,
> >>       uint32_t severity;
> >>       int ret;
> >>   
> >> +    if (!(vdev->features & VFIO_FEATURE_ENABLE_AER)) {
> >> +        return 0;
> >> +    }
> >> +
> >>       pdev->exp.aer_log.log_max = PCIE_AER_LOG_MAX_DEFAULT;
> >>       ret = pcie_aer_init(pdev, cap_ver, pos, size);
> >>       if (ret) {
> >> @@ -3569,6 +3575,8 @@ static Property vfio_pci_dev_properties[] = {
> >>                          intx.mmap_timeout, 1100),
> >>       DEFINE_PROP_BIT("x-vga", VFIOPCIDevice, features,
> >>                       VFIO_FEATURE_ENABLE_VGA_BIT, false),
> >> +    DEFINE_PROP_BIT("x-aer", VFIOPCIDevice, features,
> >> +                    VFIO_FEATURE_ENABLE_AER_BIT, true),
> >>       DEFINE_PROP_INT32("bootindex", VFIOPCIDevice, bootindex, -1),
> >>       /*
> >>        * TODO - support passed fds... is this necessary?
> >
> >
> > .
> >
>