Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190620060726.926-1-kraxel@redhat.com> <20190620060726.926-3-kraxel@redhat.com>
In-Reply-To: <20190620060726.926-3-kraxel@redhat.com>
From:   Chia-I Wu <olvaffe@gmail.com>
Date:   Wed, 26 Jun 2019 16:55:20 -0700
Message-ID: <CAPaKu7RWpoRkTkoatdYHz6itHZFvUYgaBcQAXnSC2gDc+dFZxQ@mail.gmail.com>
Subject: Re: [PATCH v4 02/12] drm/virtio: switch virtio_gpu_wait_ioctl() to
 gem helper.
To:     Gerd Hoffmann <kraxel@redhat.com>
Cc:     ML dri-devel <dri-devel@lists.freedesktop.org>,
        David Airlie <airlied@linux.ie>,
        open list <linux-kernel@vger.kernel.org>,
        "open list:VIRTIO GPU DRIVER" 
        <virtualization@lists.linux-foundation.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Jun 19, 2019 at 11:07 PM Gerd Hoffmann <kraxel@redhat.com> wrote:
>
> Use drm_gem_reservation_object_wait() in virtio_gpu_wait_ioctl().
> This also makes the ioctl run lockless.
The userspace has a BO cache to avoid freeing BOs immediately but to
reuse them on next allocations.  The BO cache checks if a BO is busy
before reuse, and I am seeing a big negative perf impact because of
slow virtio_gpu_wait_ioctl.  I wonder if this helps.


>
> v2: use reservation_object_test_signaled_rcu for VIRTGPU_WAIT_NOWAIT.
>
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> ---
>  drivers/gpu/drm/virtio/virtgpu_ioctl.c | 24 ++++++++++--------------
>  1 file changed, 10 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/virtio/virtgpu_ioctl.c b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
> index ac60be9b5c19..313c770ea2c5 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_ioctl.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
> @@ -464,23 +464,19 @@ static int virtio_gpu_wait_ioctl(struct drm_device *dev, void *data,
>                             struct drm_file *file)
>  {
>         struct drm_virtgpu_3d_wait *args = data;
> -       struct drm_gem_object *gobj = NULL;
> -       struct virtio_gpu_object *qobj = NULL;
> +       struct drm_gem_object *obj;
> +       long timeout = 15 * HZ;
>         int ret;
> -       bool nowait = false;
>
> -       gobj = drm_gem_object_lookup(file, args->handle);
> -       if (gobj == NULL)
> -               return -ENOENT;
> +       if (args->flags & VIRTGPU_WAIT_NOWAIT) {
> +               obj = drm_gem_object_lookup(file, args->handle);
Don't we need a NULL check here?
> +               ret = reservation_object_test_signaled_rcu(obj->resv, true);
> +               drm_gem_object_put_unlocked(obj);
> +               return ret ? 0 : -EBUSY;
> +       }
>
> -       qobj = gem_to_virtio_gpu_obj(gobj);
> -
> -       if (args->flags & VIRTGPU_WAIT_NOWAIT)
> -               nowait = true;
> -       ret = virtio_gpu_object_wait(qobj, nowait);
> -
> -       drm_gem_object_put_unlocked(gobj);
> -       return ret;
> +       return drm_gem_reservation_object_wait(file, args->handle,
> +                                              true, timeout);
>  }
>
>  static int virtio_gpu_get_caps_ioctl(struct drm_device *dev,
> --
> 2.18.1
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel