On Mon, Sep 29, 2025 at 9:00 PM Stefan Hajnoczi <[email protected]> wrote:
>
> On Wed, Sep 10, 2025 at 01:54:20PM +0200, Albert Esteve wrote:
> > Add shared memory BAR support to vhost-user-device-pci
> > to enable direct file mapping for VIRTIO Shared
> > Memory Regions.
> >
> > The implementation creates a consolidated shared
> > memory BAR that contains all VIRTIO Shared
> > Memory Regions as subregions. Each region is
> > configured with its proper shmid, size, and
> > offset within the BAR. The number and size of
> > regions are retrieved via VHOST_USER_GET_SHMEM_CONFIG
> > message sent by vhost-user-base during realization
> > after virtio_init().
> >
> > Specifiically, it uses BAR 3 to avoid conflicts, as
> > it is currently unused.
> >
> > The shared memory BAR is only created when the
> > backend supports VHOST_USER_PROTOCOL_F_SHMEM and
> > has configured shared memory regions. This maintains
> > backward compatibility with backends that do not
> > support shared memory functionality.
> >
> > Signed-off-by: Albert Esteve <[email protected]>
> > ---
> >  hw/virtio/vhost-user-base.c       | 47 +++++++++++++++++++++++++++++--
> >  hw/virtio/vhost-user-device-pci.c | 34 ++++++++++++++++++++--
> >  2 files changed, 76 insertions(+), 5 deletions(-)
> >
> > diff --git a/hw/virtio/vhost-user-base.c b/hw/virtio/vhost-user-base.c
> > index ff67a020b4..82f49500e4 100644
> > --- a/hw/virtio/vhost-user-base.c
> > +++ b/hw/virtio/vhost-user-base.c
> > @@ -16,6 +16,7 @@
> >  #include "hw/virtio/virtio-bus.h"
> >  #include "hw/virtio/vhost-user-base.h"
> >  #include "qemu/error-report.h"
> > +#include "migration/blocker.h"
> >
> >  static void vub_start(VirtIODevice *vdev)
> >  {
> > @@ -276,7 +277,9 @@ static void vub_device_realize(DeviceState *dev, Error 
> > **errp)
> >  {
> >      VirtIODevice *vdev = VIRTIO_DEVICE(dev);
> >      VHostUserBase *vub = VHOST_USER_BASE(dev);
> > -    int ret;
> > +    uint64_t memory_sizes[VIRTIO_MAX_SHMEM_REGIONS];
> > +    g_autofree char *name = NULL;
> > +    int i, ret, nregions, regions_processed = 0;
> >
> >      if (!vub->chardev.chr) {
> >          error_setg(errp, "vhost-user-base: missing chardev");
> > @@ -319,7 +322,7 @@ static void vub_device_realize(DeviceState *dev, Error 
> > **errp)
> >
> >      /* Allocate queues */
> >      vub->vqs = g_ptr_array_sized_new(vub->num_vqs);
> > -    for (int i = 0; i < vub->num_vqs; i++) {
> > +    for (i = 0; i < vub->num_vqs; i++) {
> >          g_ptr_array_add(vub->vqs,
> >                          virtio_add_queue(vdev, vub->vq_size,
> >                                           vub_handle_output));
> > @@ -333,11 +336,49 @@ static void vub_device_realize(DeviceState *dev, 
> > Error **errp)
> >                           VHOST_BACKEND_TYPE_USER, 0, errp);
> >
> >      if (ret < 0) {
> > -        do_vhost_user_cleanup(vdev, vub);
> > +        goto err;
> > +    }
> > +
> > +    ret = vub->vhost_dev.vhost_ops->vhost_get_shmem_config(&vub->vhost_dev,
> > +                                                           &nregions,
> > +                                                           memory_sizes,
> > +                                                           errp);
> > +
> > +    if (ret < 0) {
> > +        goto err;
> > +    }
> > +
> > +    for (i = 0; i < VIRTIO_MAX_SHMEM_REGIONS && regions_processed < 
> > nregions; i++) {
> > +        if (memory_sizes[i]) {
> > +            regions_processed++;
> > +            if (vub->vhost_dev.migration_blocker == NULL) {
> > +                error_setg(&vub->vhost_dev.migration_blocker,
> > +                       "Migration disabled: devices with VIRTIO Shared 
> > Memory "
> > +                       "Regions do not support migration yet.");
> > +                ret = migrate_add_blocker_normal(
> > +                    &vub->vhost_dev.migration_blocker,
> > +                    errp);
> > +
> > +                if (ret < 0) {
> > +                    goto err;
> > +                }
> > +            }
> > +
> > +            if (memory_sizes[i] % qemu_real_host_page_size() != 0) {
> > +                error_setg(errp, "Shared memory %d size must be a power of 
> > 2 "
> > +                                 "no smaller than the page size", i);
> > +                goto err;
> > +            }
> > +
> > +            virtio_new_shmem_region(vdev, i, memory_sizes[i]);
> > +        }
> >      }
> >
> >      qemu_chr_fe_set_handlers(&vub->chardev, NULL, NULL, vub_event, NULL,
> >                               dev, NULL, true);
> > +    return;
> > +err:
> > +    do_vhost_user_cleanup(vdev, vub);
> >  }
> >
> >  static void vub_device_unrealize(DeviceState *dev)
> > diff --git a/hw/virtio/vhost-user-device-pci.c 
> > b/hw/virtio/vhost-user-device-pci.c
> > index f10bac874e..bac99e7c60 100644
> > --- a/hw/virtio/vhost-user-device-pci.c
> > +++ b/hw/virtio/vhost-user-device-pci.c
> > @@ -8,14 +8,18 @@
> >   */
> >
> >  #include "qemu/osdep.h"
> > +#include "qapi/error.h"
> >  #include "hw/qdev-properties.h"
> >  #include "hw/virtio/vhost-user-base.h"
> >  #include "hw/virtio/virtio-pci.h"
> >
> > +#define VIRTIO_DEVICE_PCI_SHMEM_BAR 3
>
> 64-bit PCI memory BARs require two contiguous 32-bit BARs. Choosing BAR
> 3 won't work because modern_mem_bar_idx = 4 (and 5) and there is a
> collision.
>
> virtio-gpu-hostmem uses BAR 2 and 3 for VIRTIO Shared Memory Regions. In
> order to do this, it rearranges the other VIRTIO PCI BARs in
> virtio_vga_base_realize(). Following that approach should be safe.

I see. But doing the same strategy would collide with
virtio-gpu-hostmem if both devices are used simultaneously? Would that
be ok?

The only other option (I think?) is to just use
`VIRTIO_DEVICE_PCI_SHMEM_BAR 2` as there is a comment in virtio-pci
that suggests that virtio modern io bar is off by default. But that
would also collide when used with virtio-gpu-hostmem :/


Reply via email to