All of lore.kernel.org
 help / color / mirror / Atom feed
From: Emil Velikov <emil.l.velikov@gmail.com>
To: Ilia Mirkin <imirkin@alum.mit.edu>,
	mesa-dev@lists.freedesktop.org, nouveau@lists.freedesktop.org
Cc: emil.l.velikov@gmail.com
Subject: Re: [Nouveau] [PATCH] nouveau: fix fence waiting logic in screen destroy
Date: Fri, 07 Mar 2014 14:11:01 +0000	[thread overview]
Message-ID: <5319D375.4010308@gmail.com> (raw)
In-Reply-To: <1394078485-7768-1-git-send-email-imirkin@alum.mit.edu>

On 06/03/14 04:01, Ilia Mirkin wrote:
> nouveau_fence_wait has the expectation that an external entity is
> holding onto the fence being waited on, not that it is merely held onto
> by the current pointer. Fixes a use-after-free in nouveau_fence_wait
> when used on the screen's current fence.
> 
IMHO one should flatten all the fence handling a bit to greatly improve
readability and minimise chances of similar problems.

Also there is yet another reason why this might be a good idea -
nouveau_fence_signalled, and all it's callers.

I'll send a few patches of what I have in mind wrt "flattening", later
on today.

-Emil

> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75279
> Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
> ---
> 
> Should the waiting be predicated on the current fence having been emitted? I
> removed that from nv30 (which would just go ahead and _leak_ that fence it it
> hadn't been emitted...). I figure it doesn't really matter enough to worry
> about that. The bigger reason to do it, I guess, is to make sure that all the
> fences on the list get processed. But perhaps it'd be OK to just nuke them
> without regard for such details?
> 
>  src/gallium/drivers/nouveau/nv30/nv30_screen.c | 14 ++++++++++----
>  src/gallium/drivers/nouveau/nv50/nv50_screen.c | 11 +++++++++--
>  src/gallium/drivers/nouveau/nvc0/nvc0_screen.c |  9 ++++++++-
>  3 files changed, 27 insertions(+), 7 deletions(-)
> 
> diff --git a/src/gallium/drivers/nouveau/nv30/nv30_screen.c b/src/gallium/drivers/nouveau/nv30/nv30_screen.c
> index 82f2c06..5378913 100644
> --- a/src/gallium/drivers/nouveau/nv30/nv30_screen.c
> +++ b/src/gallium/drivers/nouveau/nv30/nv30_screen.c
> @@ -308,10 +308,16 @@ nv30_screen_destroy(struct pipe_screen *pscreen)
>     if (!nouveau_drm_screen_unref(&screen->base))
>        return;
>  
> -   if (screen->base.fence.current &&
> -       screen->base.fence.current->state >= NOUVEAU_FENCE_STATE_EMITTED) {
> -      nouveau_fence_wait(screen->base.fence.current);
> -      nouveau_fence_ref (NULL, &screen->base.fence.current);
> +   if (screen->base.fence.current) {
> +      struct nouveau_fence *current = NULL;
> +
> +      /* nouveau_fence_wait will create a new current fence, so wait on the
> +       * _current_ one, and remove both.
> +       */
> +      nouveau_fence_ref(screen->base.fence.current, &current);
> +      nouveau_fence_wait(current);
> +      nouveau_fence_ref(NULL, &current);
> +      nouveau_fence_ref(NULL, &screen->base.fence.current);
>     }
>  
>     nouveau_object_del(&screen->query);
> diff --git a/src/gallium/drivers/nouveau/nv50/nv50_screen.c b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
> index ab0d63e..e8c7fe3 100644
> --- a/src/gallium/drivers/nouveau/nv50/nv50_screen.c
> +++ b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
> @@ -294,8 +294,15 @@ nv50_screen_destroy(struct pipe_screen *pscreen)
>        return;
>  
>     if (screen->base.fence.current) {
> -      nouveau_fence_wait(screen->base.fence.current);
> -      nouveau_fence_ref (NULL, &screen->base.fence.current);
> +      struct nouveau_fence *current = NULL;
> +
> +      /* nouveau_fence_wait will create a new current fence, so wait on the
> +       * _current_ one, and remove both.
> +       */
> +      nouveau_fence_ref(screen->base.fence.current, &current);
> +      nouveau_fence_wait(current);
> +      nouveau_fence_ref(NULL, &current);
> +      nouveau_fence_ref(NULL, &screen->base.fence.current);
>     }
>     if (screen->base.pushbuf)
>        screen->base.pushbuf->user_priv = NULL;
> diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
> index 044847d..04f3088 100644
> --- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
> +++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
> @@ -340,7 +340,14 @@ nvc0_screen_destroy(struct pipe_screen *pscreen)
>        return;
>  
>     if (screen->base.fence.current) {
> -      nouveau_fence_wait(screen->base.fence.current);
> +      struct nouveau_fence *current = NULL;
> +
> +      /* nouveau_fence_wait will create a new current fence, so wait on the
> +       * _current_ one, and remove both.
> +       */
> +      nouveau_fence_ref(screen->base.fence.current, &current);
> +      nouveau_fence_wait(current);
> +      nouveau_fence_ref(NULL, &current);
>        nouveau_fence_ref(NULL, &screen->base.fence.current);
>     }
>     if (screen->base.pushbuf)
> 

  reply	other threads:[~2014-03-07 14:11 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-03-06  4:01 [PATCH] nouveau: fix fence waiting logic in screen destroy Ilia Mirkin
2014-03-07 14:11 ` Emil Velikov [this message]
2014-03-07 17:31   ` [Nouveau] " Ilia Mirkin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=5319D375.4010308@gmail.com \
    --to=emil.l.velikov@gmail.com \
    --cc=imirkin@alum.mit.edu \
    --cc=mesa-dev@lists.freedesktop.org \
    --cc=nouveau@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.