lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAOw6vbKygkfN_2aKjcp--pyJBCmNE0efT7bVNWh7LmirK7dO9A@mail.gmail.com>
Date:   Mon, 12 Sep 2016 17:06:41 -0400
From:   Sean Paul <seanpaul@...gle.com>
To:     Gustavo Padovan <gustavo@...ovan.org>
Cc:     dri-devel <dri-devel@...ts.freedesktop.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Daniel Stone <daniels@...labora.com>,
        Daniel Vetter <daniel.vetter@...ll.ch>,
        Rob Clark <robdclark@...il.com>,
        Greg Hackmann <ghackmann@...gle.com>,
        John Harrison <John.C.Harrison@...el.com>,
        Laurent Pinchart <laurent.pinchart@...asonboard.com>,
        Stéphane Marchesin <marcheu@...gle.com>,
        m.chehab@...sung.com, Sumit Semwal <sumit.semwal@...aro.org>,
        Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
        Gustavo Padovan <gustavo.padovan@...labora.co.uk>
Subject: Re: [PATCH v4] drm/fence: allow fence waiting to be interrupted by userspace

On Mon, Sep 12, 2016 at 3:08 PM, Gustavo Padovan <gustavo@...ovan.org> wrote:
> From: Gustavo Padovan <gustavo.padovan@...labora.co.uk>
>
> If userspace is running an synchronously atomic commit and interrupts the
> atomic operation during fence_wait() it will hang until the timer expires,
> so here we change the wait to be interruptible so it stop immediately when
> userspace wants to quit.
>
> Also adds the necessary error checking for fence_wait().
>
> v2: Comment by Daniel Vetter
>         - Add error checking for fence_wait()
>
> v3: Rebase on top of new atomic noblocking support
>
> v4: Comment by Maarten Lankhorst
>         - remove 'swapped' bitfield as it was duplicating information
>
> v5: Comments by Maarten Lankhorst
>         - assign plane->state to plane_state if !intr
>         - squash previous patch into this one
>
> v6: Comment by Sean Paul
>         - rename intr to pre_swap
>
> Signed-off-by: Gustavo Padovan <gustavo.padovan@...labora.co.uk>
> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>


Applied to drm-misc with a couple checkpatch fixes. I also moved the
comment to directly above fence_wait (IRC acked).

Sean


> ---
>  drivers/gpu/drm/drm_atomic_helper.c | 41 +++++++++++++++++++++++++++++--------
>  drivers/gpu/drm/msm/msm_atomic.c    |  2 +-
>  include/drm/drm_atomic_helper.h     |  5 +++--
>  3 files changed, 36 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
> index 6fdd7ba..c34d002 100644
> --- a/drivers/gpu/drm/drm_atomic_helper.c
> +++ b/drivers/gpu/drm/drm_atomic_helper.c
> @@ -1009,29 +1009,46 @@ EXPORT_SYMBOL(drm_atomic_helper_commit_modeset_enables);
>   * drm_atomic_helper_wait_for_fences - wait for fences stashed in plane state
>   * @dev: DRM device
>   * @state: atomic state object with old state structures
> + * @pre_swap: if true, do an interruptible wait
>   *
>   * For implicit sync, driver should fish the exclusive fence out from the
>   * incoming fb's and stash it in the drm_plane_state.  This is called after
>   * drm_atomic_helper_swap_state() so it uses the current plane state (and
>   * just uses the atomic state to find the changed planes)
> + *
> + * Returns zero if sucess or < 0 if fence_wait() fails.
>   */
> -void drm_atomic_helper_wait_for_fences(struct drm_device *dev,
> -                           struct drm_atomic_state *state)
> +int drm_atomic_helper_wait_for_fences(struct drm_device *dev,
> +                                      struct drm_atomic_state *state,
> +                                      bool pre_swap)
>  {
>         struct drm_plane *plane;
>         struct drm_plane_state *plane_state;
> -       int i;
> +       int i, ret;
>
>         for_each_plane_in_state(state, plane, plane_state, i) {
> -               if (!plane->state->fence)
> +               /*
> +                * If waiting for fences pre-swap (ie: nonblock), userspace can
> +                * still interrupt the operation. Instead of blocking until the
> +                * timer expires, make the wait interruptible.
> +                */
> +               if (!pre_swap)
> +                       plane_state = plane->state;
> +
> +               if (!plane_state->fence)
>                         continue;
>
> -               WARN_ON(!plane->state->fb);
> +               WARN_ON(!plane_state->fb);
> +
> +               ret = fence_wait(plane_state->fence, pre_swap);
> +               if (ret)
> +                       return ret;
>
> -               fence_wait(plane->state->fence, false);
> -               fence_put(plane->state->fence);
> -               plane->state->fence = NULL;
> +               fence_put(plane_state->fence);
> +               plane_state->fence = NULL;
>         }
> +
> +       return 0;
>  }
>  EXPORT_SYMBOL(drm_atomic_helper_wait_for_fences);
>
> @@ -1179,7 +1196,7 @@ static void commit_tail(struct drm_atomic_state *state)
>
>         funcs = dev->mode_config.helper_private;
>
> -       drm_atomic_helper_wait_for_fences(dev, state);
> +       drm_atomic_helper_wait_for_fences(dev, state, false);
>
>         drm_atomic_helper_wait_for_dependencies(state);
>
> @@ -1238,6 +1255,12 @@ int drm_atomic_helper_commit(struct drm_device *dev,
>         if (ret)
>                 return ret;
>
> +       if (!nonblock) {
> +               ret = drm_atomic_helper_wait_for_fences(dev, state, true);
> +               if (ret)
> +                       return ret;
> +       }
> +
>         /*
>          * This is the point of no return - everything below never fails except
>          * when the hw goes bonghits. Which means we can commit the new state on
> diff --git a/drivers/gpu/drm/msm/msm_atomic.c b/drivers/gpu/drm/msm/msm_atomic.c
> index 5df252c..73bae38 100644
> --- a/drivers/gpu/drm/msm/msm_atomic.c
> +++ b/drivers/gpu/drm/msm/msm_atomic.c
> @@ -112,7 +112,7 @@ static void complete_commit(struct msm_commit *c, bool async)
>         struct msm_drm_private *priv = dev->dev_private;
>         struct msm_kms *kms = priv->kms;
>
> -       drm_atomic_helper_wait_for_fences(dev, state);
> +       drm_atomic_helper_wait_for_fences(dev, state, false);
>
>         kms->funcs->prepare_commit(kms, state);
>
> diff --git a/include/drm/drm_atomic_helper.h b/include/drm/drm_atomic_helper.h
> index f866828..7ff92b0 100644
> --- a/include/drm/drm_atomic_helper.h
> +++ b/include/drm/drm_atomic_helper.h
> @@ -45,8 +45,9 @@ int drm_atomic_helper_commit(struct drm_device *dev,
>                              struct drm_atomic_state *state,
>                              bool nonblock);
>
> -void drm_atomic_helper_wait_for_fences(struct drm_device *dev,
> -                                       struct drm_atomic_state *state);
> +int drm_atomic_helper_wait_for_fences(struct drm_device *dev,
> +                                       struct drm_atomic_state *state,
> +                                       bool pre_swap);
>  bool drm_atomic_helper_framebuffer_changed(struct drm_device *dev,
>                                            struct drm_atomic_state *old_state,
>                                            struct drm_crtc *crtc);
> --
> 2.5.5
>

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ