Re: [RFC 3/4] vduse: Add the function for get/free the mapp pages

[Cindy Lu <lulu@redhat.com>]

On Wed, Jun 28, 2023 at 4:11 PM Jason Wang <jasowang@redhat.com> wrote:
>
> On Wed, Jun 28, 2023 at 2:59 PM Cindy Lu <lulu@redhat.com> wrote:
> >
> > From: Your Name <you@example.com>
> >
> > Add the function for get/free pages, ad this info
> > will saved in dev->reconnect_info
>
> I think this should be squashed to patch 2 otherwise it fixes a bug
> that is introduced in patch 2?
>
sure will do
> >
> > Signed-off-by: Cindy Lu <lulu@redhat.com>
> > ---
> >  drivers/vdpa/vdpa_user/vduse_dev.c | 35 ++++++++++++++++++++++++++++++
> >  1 file changed, 35 insertions(+)
> >
> > diff --git a/drivers/vdpa/vdpa_user/vduse_dev.c b/drivers/vdpa/vdpa_user/vduse_dev.c
> > index 1b833bf0ae37..3df1256eccb4 100644
> > --- a/drivers/vdpa/vdpa_user/vduse_dev.c
> > +++ b/drivers/vdpa/vdpa_user/vduse_dev.c
> > @@ -1313,6 +1313,35 @@ static struct vduse_dev *vduse_dev_get_from_minor(int minor)
> >         return dev;
> >  }
> >
> > +int vduse_get_vq_reconnnect(struct vduse_dev *dev, u16 idx)
> > +{
> > +       struct vdpa_reconnect_info *area;
> > +       void *addr = (void *)get_zeroed_page(GFP_KERNEL);
> > +
> > +       area = &dev->reconnect_info[idx];
> > +
> > +       area->addr = virt_to_phys(addr);
> > +       area->vaddr = (unsigned long)addr;
> > +       area->size = PAGE_SIZE;
> > +       area->index = idx;
> > +
> > +       return 0;
> > +}
> > +
> > +int vduse_free_vq_reconnnect(struct vduse_dev *dev, u16 idx)
> > +{
> > +       struct vdpa_reconnect_info *area;
> > +
> > +       area = &dev->reconnect_info[idx];
> > +       if ((area->size == PAGE_SIZE) && (area->addr != NULL)) {
> > +               free_page(area->vaddr);
> > +               area->size = 0;
> > +               area->addr = 0;
> > +               area->vaddr = 0;
> > +       }
> > +
> > +       return 0;
> > +}
> >
> >  static vm_fault_t vduse_vm_fault(struct vm_fault *vmf)
> >  {
> > @@ -1446,6 +1475,10 @@ static int vduse_destroy_dev(char *name)
> >                 mutex_unlock(&dev->lock);
> >                 return -EBUSY;
> >         }
> > +       for (int i = 0; i < dev->vq_num; i++) {
> > +
> > +               vduse_free_vq_reconnnect(dev, i);
> > +       }
> >         dev->connected = true;
> >         mutex_unlock(&dev->lock);
> >
> > @@ -1583,6 +1616,8 @@ static int vduse_create_dev(struct vduse_dev_config *config,
> >                 INIT_WORK(&dev->vqs[i].kick, vduse_vq_kick_work);
> >                 spin_lock_init(&dev->vqs[i].kick_lock);
> >                 spin_lock_init(&dev->vqs[i].irq_lock);
> > +
> > +               vduse_get_vq_reconnnect(dev, i);
>
> Can we delay the allocated until fault?
>
sure will do
> Thanks
>
> >         }
> >
> >         ret = idr_alloc(&vduse_idr, dev, 1, VDUSE_DEV_MAX, GFP_KERNEL);
> > --
> > 2.34.3
> >
>