Use the shared address space of the drm device (see drm_open() in
drm_file.c) for dma-bufs too. That removes a difference betweem drm
device mmap vmas and dma-buf mmap vmas and fixes corner cases like
unmaps not working properly.
Signed-off-by: Gerd Hoffmann <[email protected]>
---
drivers/gpu/drm/drm_prime.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
index a9633bd241bb..c3fc341453c0 100644
--- a/drivers/gpu/drm/drm_prime.c
+++ b/drivers/gpu/drm/drm_prime.c
@@ -240,6 +240,7 @@ void drm_prime_destroy_file_private(struct drm_prime_file_private *prime_fpriv)
struct dma_buf *drm_gem_dmabuf_export(struct drm_device *dev,
struct dma_buf_export_info *exp_info)
{
+ struct drm_gem_object *obj = exp_info->priv;
struct dma_buf *dma_buf;
dma_buf = dma_buf_export(exp_info);
@@ -247,7 +248,8 @@ struct dma_buf *drm_gem_dmabuf_export(struct drm_device *dev,
return dma_buf;
drm_dev_get(dev);
- drm_gem_object_get(exp_info->priv);
+ drm_gem_object_get(obj);
+ dma_buf->file->f_mapping = obj->dev->anon_inode->i_mapping;
return dma_buf;
}
--
2.18.1
On Thu, Nov 21, 2019 at 11:38:07AM +0100, Gerd Hoffmann wrote:
> Use the shared address space of the drm device (see drm_open() in
> drm_file.c) for dma-bufs too. That removes a difference betweem drm
> device mmap vmas and dma-buf mmap vmas and fixes corner cases like
> unmaps not working properly.
>
> Signed-off-by: Gerd Hoffmann <[email protected]>
> ---
> drivers/gpu/drm/drm_prime.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
> index a9633bd241bb..c3fc341453c0 100644
> --- a/drivers/gpu/drm/drm_prime.c
> +++ b/drivers/gpu/drm/drm_prime.c
> @@ -240,6 +240,7 @@ void drm_prime_destroy_file_private(struct drm_prime_file_private *prime_fpriv)
> struct dma_buf *drm_gem_dmabuf_export(struct drm_device *dev,
> struct dma_buf_export_info *exp_info)
> {
> + struct drm_gem_object *obj = exp_info->priv;
> struct dma_buf *dma_buf;
>
> dma_buf = dma_buf_export(exp_info);
> @@ -247,7 +248,8 @@ struct dma_buf *drm_gem_dmabuf_export(struct drm_device *dev,
> return dma_buf;
>
> drm_dev_get(dev);
> - drm_gem_object_get(exp_info->priv);
> + drm_gem_object_get(obj);
> + dma_buf->file->f_mapping = obj->dev->anon_inode->i_mapping;
Can you pls also throw the change to amdgpu_gem_prime_export on top here?
Imo makes a lot more sense that way. With that added I'm happy to r-b.
-Daniel
>
> return dma_buf;
> }
> --
> 2.18.1
>
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
>-----Original Message-----
>From: Intel-gfx <[email protected]> On Behalf Of Gerd
>Hoffmann
>Sent: Thursday, November 21, 2019 5:38 AM
>To: [email protected]
>Cc: David Airlie <[email protected]>; [email protected]; open list
><[email protected]>; Maxime Ripard <[email protected]>; Gerd
>Hoffmann <[email protected]>
>Subject: [Intel-gfx] [PATCH 2/2] drm: share address space for dma bufs
>
>Use the shared address space of the drm device (see drm_open() in
>drm_file.c) for dma-bufs too. That removes a difference betweem drm
>device mmap vmas and dma-buf mmap vmas and fixes corner cases like
>unmaps not working properly.
Hi Gerd,
Just want to make sure I understand this...
So unmaps will not work correctly for mappings when a driver does a
drm_vma_node_unamp()?
I.e. the dmabuf mappings will not get cleaned up correctly?
This is a day one bug?
Thanks,
Mike
>Signed-off-by: Gerd Hoffmann <[email protected]>
>---
> drivers/gpu/drm/drm_prime.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
>diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
>index a9633bd241bb..c3fc341453c0 100644
>--- a/drivers/gpu/drm/drm_prime.c
>+++ b/drivers/gpu/drm/drm_prime.c
>@@ -240,6 +240,7 @@ void drm_prime_destroy_file_private(struct
>drm_prime_file_private *prime_fpriv)
> struct dma_buf *drm_gem_dmabuf_export(struct drm_device *dev,
> struct dma_buf_export_info *exp_info)
> {
>+ struct drm_gem_object *obj = exp_info->priv;
> struct dma_buf *dma_buf;
>
> dma_buf = dma_buf_export(exp_info);
>@@ -247,7 +248,8 @@ struct dma_buf *drm_gem_dmabuf_export(struct
>drm_device *dev,
> return dma_buf;
>
> drm_dev_get(dev);
>- drm_gem_object_get(exp_info->priv);
>+ drm_gem_object_get(obj);
>+ dma_buf->file->f_mapping = obj->dev->anon_inode->i_mapping;
>
> return dma_buf;
> }
>--
>2.18.1
>
>_______________________________________________
>Intel-gfx mailing list
>[email protected]
>https://lists.freedesktop.org/mailman/listinfo/intel-gfx
On Thu, Nov 21, 2019 at 04:42:10PM +0000, Ruhl, Michael J wrote:
> >-----Original Message-----
> >From: Intel-gfx <[email protected]> On Behalf Of Gerd
> >Hoffmann
> >Sent: Thursday, November 21, 2019 5:38 AM
> >To: [email protected]
> >Cc: David Airlie <[email protected]>; [email protected]; open list
> ><[email protected]>; Maxime Ripard <[email protected]>; Gerd
> >Hoffmann <[email protected]>
> >Subject: [Intel-gfx] [PATCH 2/2] drm: share address space for dma bufs
> >
> >Use the shared address space of the drm device (see drm_open() in
> >drm_file.c) for dma-bufs too. That removes a difference betweem drm
> >device mmap vmas and dma-buf mmap vmas and fixes corner cases like
> >unmaps not working properly.
>
> Hi Gerd,
>
> Just want to make sure I understand this...
>
> So unmaps will not work correctly for mappings when a driver does a
> drm_vma_node_unamp()?
Completely removing the mapping (aka munmap syscall) works fine.
Zapping the pte's (using madvise(dontneed) for example) doesn't.
> This is a day one bug?
I guess so, but I'll leave that to others being active longer than me in
drm hacking to answer ...
cheers,
Gerd