Re: [Dri-devel] [patch] Yet another DMA buffer leak

[Keith W. <ke...@tu...>]

Michel D=E4nzer wrote:
> On Sam, 2003-02-08 at 18:56, Michel D=E4nzer wrote:
>=20
>>On Sam, 2003-02-08 at 18:45, Charl P. Botha wrote:
>>
>>>On Sat, Feb 08, 2003 at 06:24:28PM +0100, Felix K=FChling wrote:
>>>
>>>>Found it. DRIVER_RELEASE in radeon.h has to call radeon_reclaim_buffe=
rs
>>>>as the drm_drv.h template seems to expect DRIVER_RELEASE to handle th=
is.
>>>>DRIVER_RELEASE in i810.h does this too. The (1 line) patch is attache=
d.
>>>
>>>Thanks for fixing yet another one Felix! =20
>>
>>Indeed, keep up the good work! :)
>>
>>
>>>Should this not be forwarded to de...@xf... or fi...@xf...
>>>as well?
>>
>>It should get picked up from the mesa-4-0-4-branch, where I'll commit i=
t
>>shortly, as well as to the trunk.
>=20
>=20
> Hmm, moving this to mesa-4-0-4-branch is trickier than I thought. Here'=
s
> a quick'n'dirty merge, but maybe we want to move over more of Keith's
> recent work in this area? Keith?
>=20
>=20
>=20
>=20
> -----------------------------------------------------------------------=
-
>=20
> Index: radeon.h
> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
> RCS file: /cvsroot/dri/xc/xc/programs/Xserver/hw/xfree86/os-support/sha=
red/drm/kernel/radeon.h,v
> retrieving revision 1.8
> diff -p -u -r1.8 radeon.h
> --- radeon.h	28 Oct 2002 19:05:37 -0000	1.8
> +++ radeon.h	8 Feb 2003 18:35:23 -0000
> @@ -114,8 +114,6 @@
>  /* When a client dies:
>   *    - Check for and clean up flipped page state
>   *    - Free any alloced agp memory.
> - *
> - * DRM infrastructure takes care of reclaiming dma buffers.
>   */
>  #define DRIVER_PRERELEASE() do {					\
>  	if ( dev->dev_private ) {					\
> @@ -125,6 +123,14 @@
>  		}							\
>                  radeon_mem_release( dev_priv->agp_heap );		\
>  	}								\
> +} while (0)
> +
> +/* Reclaim DMA buffers when a client dies
> + */
> +#define __HAVE_RELEASE 1
> +#define DRIVER_RELEASE() 			\
> +do {						\
> +        DRM(reclaim_buffers)( dev, priv->pid ); \
>  } while (0)
> =20
>  /* On unloading the module:


Actually if __HAVE_RELEASE isn't defined, the driver on that branch shoul=
d=20
already be doing the right thing...  The trouble is if you define=20
__HAVE_RELEASE, the templates no longer do the DRM(reclaim_buffers) thing=
 for=20
you -- I didn't realize this, and Felix has located my error.  However th=
e=20
version in the mesa-4-0-4 branch should (I think) work fine without the c=
hange.

Keith