On Sun, Apr 7, 2024 at 7:53 PM Michael S. Tsirkin <m...@redhat.com> wrote:
>
> On Sun, Apr 07, 2024 at 12:19:57PM +0800, Jason Wang wrote:
> > On Tue, Apr 2, 2024 at 11:02 PM Cindy Lu <l...@redhat.com> wrote:
> > >
> > > When the guest calls virtio_stop and then virtio_reset,
> >
> > Guests could not call those functions directly, it is triggered by for
> > example writing to some of the registers like reset or others.
> >
> > > the vector will change
> > > to VIRTIO_NO_VECTOR and the IRQFD for this vector will be released. After
> > > that
> > > If you want to change the vector back,
> >
> > What do you mean by "change the vector back"? Something like
> >
> > assign VIRTIO_MSI_NO_VECTOR to vector 0
> > assign X to vector 0
> >
> > And I guess what you meant is to configure the vector after DRIVER_OK.
> >
> >
> > > it will cause a crash.
> > >
> > > To fix this, we need to call the function
> > > "kvm_virtio_pci_vector_use_one()"
> > > when the vector changes back from VIRTIO_NO_VECTOR
> > >
> > > Signed-off-by: Cindy Lu <l...@redhat.com>
> > > ---
> > > hw/virtio/virtio-pci.c | 31 ++++++++++++++++++++++++++++---
> > > 1 file changed, 28 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
> > > index e433879542..45f3ab38c3 100644
> > > --- a/hw/virtio/virtio-pci.c
> > > +++ b/hw/virtio/virtio-pci.c
> > > @@ -874,12 +874,14 @@ static int virtio_pci_get_notifier(VirtIOPCIProxy
> > > *proxy, int queue_no,
> > > return 0;
> > > }
> > >
> > > -static int kvm_virtio_pci_vector_use_one(VirtIOPCIProxy *proxy, int
> > > queue_no)
> > > +static int kvm_virtio_pci_vector_use_one(VirtIOPCIProxy *proxy, int
> > > queue_no,
> > > + bool recovery)
> > > {
> > > unsigned int vector;
> > > int ret;
> > > EventNotifier *n;
> > > PCIDevice *dev = &proxy->pci_dev;
> > > + VirtIOIRQFD *irqfd;
> > > VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
> > > VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
> > >
> > > @@ -890,10 +892,21 @@ static int
> > > kvm_virtio_pci_vector_use_one(VirtIOPCIProxy *proxy, int queue_no)
> > > if (vector >= msix_nr_vectors_allocated(dev)) {
> > > return 0;
> > > }
> > > + /*
> > > + * if this is recovery and irqfd still in use, means the irqfd was
> > > not
> > > + * release before and don't need to set up again
> > > + */
> > > + if (recovery) {
> > > + irqfd = &proxy->vector_irqfd[vector];
> > > + if (irqfd->users != 0) {
> > > + return 0;
> > > + }
> > > + }
> > > ret = kvm_virtio_pci_vq_vector_use(proxy, vector);
> > > if (ret < 0) {
> > > goto undo;
> > > }
> > > +
> > > /*
> > > * If guest supports masking, set up irqfd now.
> > > * Otherwise, delay until unmasked in the frontend.
> > > @@ -932,14 +945,14 @@ static int
> > > kvm_virtio_pci_vector_vq_use(VirtIOPCIProxy *proxy, int nvqs)
> > > if (!virtio_queue_get_num(vdev, queue_no)) {
> > > return -1;
> > > }
> > > - ret = kvm_virtio_pci_vector_use_one(proxy, queue_no);
> > > + ret = kvm_virtio_pci_vector_use_one(proxy, queue_no, false);
> > > }
> > > return ret;
> > > }
> > >
> > > static int kvm_virtio_pci_vector_config_use(VirtIOPCIProxy *proxy)
> > > {
> > > - return kvm_virtio_pci_vector_use_one(proxy, VIRTIO_CONFIG_IRQ_IDX);
> > > + return kvm_virtio_pci_vector_use_one(proxy, VIRTIO_CONFIG_IRQ_IDX,
> > > false);
> > > }
> > >
> > > static void kvm_virtio_pci_vector_release_one(VirtIOPCIProxy *proxy,
> > > @@ -1570,7 +1583,13 @@ static void virtio_pci_common_write(void *opaque,
> > > hwaddr addr,
> > > } else {
> > > val = VIRTIO_NO_VECTOR;
> > > }
> > > + vector = vdev->config_vector;
> > > vdev->config_vector = val;
> > > + /*check if the vector need to recovery*/
> > > + if ((val != VIRTIO_NO_VECTOR) && (vector == VIRTIO_NO_VECTOR) &&
> > > + (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
> > > + kvm_virtio_pci_vector_use_one(proxy, VIRTIO_CONFIG_IRQ_IDX,
> > > true);
> > > + }
> >
> > This looks too tricky.
> >
> > Think hard of this. I think it's better to split this into two parts:
> >
> > 1) a series that disables config irqfd for vhost-net, this series
> > needs to be backported to -stable which needs to be conservative. It
> > looks more like your V1, but let's add a boolean for pci proxy.
>
> I don't get it. Looks like a huge change to do in stable.
> All as a replacement to a small 20 line patch?
For example, this patch needs more tweak so it won't be that tiny:
1) needs to check if we can use guest notifiers (irqfd)
2) the switching from X to NO_VECTOR might need
kvm_virtio_pci_vq_vector_release()
>
> Generally I think irqfd is best used everywhere.
Only when msix and kvm are both enabled.
>
>
> > 2) a series that deal with the msix vector configuration after
> > driver_ok, we probably need some refactoring to do per vq use instead
> > of the current loop in DRIVER_OK
> >
> > Does this make sense?
> >
> > Thanks
>
>
> Not really let's fix the bug for starters, refactoring can be done later
> as appropriate.
This is exactly my point. Avoiding that for non vDPA device seems to
be easier or maybe I was wrong.
Thanks
>
> > > break;
> > > case VIRTIO_PCI_COMMON_STATUS:
> > > if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
> > > @@ -1611,6 +1630,12 @@ static void virtio_pci_common_write(void *opaque,
> > > hwaddr addr,
> > > val = VIRTIO_NO_VECTOR;
> > > }
> > > virtio_queue_set_vector(vdev, vdev->queue_sel, val);
> > > +
> > > + /*check if the vector need to recovery*/
> > > + if ((val != VIRTIO_NO_VECTOR) && (vector == VIRTIO_NO_VECTOR) &&
> > > + (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
> > > + kvm_virtio_pci_vector_use_one(proxy, vdev->queue_sel, true);
> > > + }
> > > break;
> > > case VIRTIO_PCI_COMMON_Q_ENABLE:
> > > if (val == 1) {
> > > --
> > > 2.43.0
> > >
>
