[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAH1Ww+SBgw6-HikPBpE1_yLG-X75LfSBbNerzzqBzyCArgm1-Q@mail.gmail.com>
Date: Thu, 18 Mar 2021 09:18:17 +0100
From: Daniel Gomez <daniel@...c.com>
To: Christian König <christian.koenig@....com>
Cc: Daniel Gomez <dagmcr@...il.com>,
Alex Deucher <alexander.deucher@....com>,
David Airlie <airlied@...ux.ie>,
Daniel Vetter <daniel@...ll.ch>,
Dave Airlie <airlied@...hat.com>,
Felix Kuehling <Felix.Kuehling@....com>,
Likun Gao <Likun.Gao@....com>,
amd-gfx list <amd-gfx@...ts.freedesktop.org>,
dri-devel <dri-devel@...ts.freedesktop.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] drm/amdgpu/ttm: Fix memory leak userptr pages
On Thu, 18 Mar 2021 at 08:49, Christian König <christian.koenig@....com> wrote:
>
> Am 17.03.21 um 17:08 schrieb Daniel Gomez:
> > If userptr pages have been pinned but not bounded,
> > they remain uncleared.
> >
> > Signed-off-by: Daniel Gomez <daniel@...c.com>
>
> Good catch, not sure if that can ever happen in practice but better save
> than sorry.
Thanks! We actually had a case with clEnqueueWriteBuffer where the
driver was leaking.
I can see the problem also affects to the radeon driver so, I'll send
a patch for that one as
well.
>
> Reviewed-by: Christian König <christian.koenig@....com>
>
> > ---
> > drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 6 +++---
> > 1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > index 9fd2157b133a..50c2b4827c13 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > @@ -1162,13 +1162,13 @@ static void amdgpu_ttm_backend_unbind(struct ttm_bo_device *bdev,
> > struct amdgpu_ttm_tt *gtt = (void *)ttm;
> > int r;
> >
> > - if (!gtt->bound)
> > - return;
> > -
> > /* if the pages have userptr pinning then clear that first */
> > if (gtt->userptr)
> > amdgpu_ttm_tt_unpin_userptr(bdev, ttm);
> >
> > + if (!gtt->bound)
> > + return;
> > +
> > if (gtt->offset == AMDGPU_BO_INVALID_OFFSET)
> > return;
> >
>
Powered by blists - more mailing lists