| Message ID | 1493744528-25768-3-git-send-email-labbott@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, May 02, 2017 at 10:02:07AM -0700, Laura Abbott wrote: > The existing drm_gem_prime_import function uses the underlying > struct device of a drm_device for attaching to a dma_buf. Some drivers > (notably vgem) may not have an underlying device structure. Offer > an alternate function to attach using a platform device associated > with drm_device. > > Cc: intel-gfx@lists.freedesktop.org > Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> > Signed-off-by: Laura Abbott <labbott@redhat.com> > --- > v3: Rebase to drm-misc-next. Prototype moved to a new header file. Comments > added for new function. Brought back drm_device->platformdev as it had been > removed in 76adb460fd93 ("drm: Remove the struct drm_device platformdev field"). > I'm not entirely thrilled about this since the platformdev removal was good > cleanup and this feels like a small step backwards. I don't know of a better > approach though. > --- > drivers/gpu/drm/drm_prime.c | 49 +++++++++++++++++++++++++++++++++++---------- > include/drm/drmP.h | 2 ++ > include/drm/drm_prime.h | 4 ++++ > 3 files changed, 44 insertions(+), 11 deletions(-) > > diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c > index 9fb65b7..a557a4b 100644 > --- a/drivers/gpu/drm/drm_prime.c > +++ b/drivers/gpu/drm/drm_prime.c > @@ -594,16 +594,9 @@ int drm_gem_prime_handle_to_fd(struct drm_device *dev, > } > EXPORT_SYMBOL(drm_gem_prime_handle_to_fd); > > -/** > - * drm_gem_prime_import - helper library implementation of the import callback > - * @dev: drm_device to import into > - * @dma_buf: dma-buf object to import > - * > - * This is the implementation of the gem_prime_import functions for GEM drivers > - * using the PRIME helpers. > - */ > -struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, > - struct dma_buf *dma_buf) > +static struct drm_gem_object *__drm_gem_prime_import(struct drm_device *dev, > + struct dma_buf *dma_buf, > + struct device *attach_dev) > { > struct dma_buf_attachment *attach; > struct sg_table *sgt; > @@ -625,7 +618,7 @@ struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, > if (!dev->driver->gem_prime_import_sg_table) > return ERR_PTR(-EINVAL); > > - attach = dma_buf_attach(dma_buf, dev->dev); > + attach = dma_buf_attach(dma_buf, attach_dev); > if (IS_ERR(attach)) > return ERR_CAST(attach); > > @@ -655,9 +648,43 @@ struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, > > return ERR_PTR(ret); > } > + > +/** > + * drm_gem_prime_import - helper library implementation of the import callback > + * @dev: drm_device to import into > + * @dma_buf: dma-buf object to import > + * > + * This is the implementation of the gem_prime_import functions for GEM drivers > + * using the PRIME helpers. > + */ > +struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, > + struct dma_buf *dma_buf) > +{ > + return __drm_gem_prime_import(dev, dma_buf, dev->dev); > +} > EXPORT_SYMBOL(drm_gem_prime_import); > > /** > + * drm_gem_prime_import_platform - alternate implementation of the import callback > + * @dev: drm_device to import into > + * @dma_buf: dma-buf object to import > + * > + * This is identical to drm_gem_prime_import except the device used for dma_buf > + * attachment is an internal platform device instead of the standard device > + * structure. The use of this function should be limited to drivers that do not > + * set up an underlying device structure. > + */ > +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, Simpler soluation will be for the caller to provide the platformdev? That works nicely for the vgem case, I think. > + struct dma_buf *dma_buf) > +{ > + if (WARN_ON_ONCE(!dev->platformdev)) > + return NULL; > + > + return __drm_gem_prime_import(dev, dma_buf, &dev->platformdev->dev); > +} > +EXPORT_SYMBOL(drm_gem_prime_import_platform);
On Tue, May 02, 2017 at 09:22:13PM +0100, Chris Wilson wrote: > On Tue, May 02, 2017 at 10:02:07AM -0700, Laura Abbott wrote: > > /** > > + * drm_gem_prime_import_platform - alternate implementation of the import callback > > + * @dev: drm_device to import into > > + * @dma_buf: dma-buf object to import > > + * > > + * This is identical to drm_gem_prime_import except the device used for dma_buf > > + * attachment is an internal platform device instead of the standard device > > + * structure. The use of this function should be limited to drivers that do not > > + * set up an underlying device structure. > > + */ > > +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, > > Simpler soluation will be for the caller to provide the platformdev? > > That works nicely for the vgem case, I think. Yeah looking at this again, do we really need this patch? Couldn't we instead change patch 1 to first allocate the fake platform device, then pass that to drm_dev_alloc (instead of NULL like we do now)? That way no resurrection of drm_device.platform_dev is needed (and I'd really like this zombie to stay dead on 2nd thought). Sry about this yet-another-round review :-/ -Daniel
On 05/03/2017 12:39 AM, Daniel Vetter wrote: > On Tue, May 02, 2017 at 09:22:13PM +0100, Chris Wilson wrote: >> On Tue, May 02, 2017 at 10:02:07AM -0700, Laura Abbott wrote: >>> /** >>> + * drm_gem_prime_import_platform - alternate implementation of the import callback >>> + * @dev: drm_device to import into >>> + * @dma_buf: dma-buf object to import >>> + * >>> + * This is identical to drm_gem_prime_import except the device used for dma_buf >>> + * attachment is an internal platform device instead of the standard device >>> + * structure. The use of this function should be limited to drivers that do not >>> + * set up an underlying device structure. >>> + */ >>> +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, >> >> Simpler soluation will be for the caller to provide the platformdev? >> >> That works nicely for the vgem case, I think. > > Yeah looking at this again, do we really need this patch? Couldn't we > instead change patch 1 to first allocate the fake platform device, then > pass that to drm_dev_alloc (instead of NULL like we do now)? > That was what I proposed in the first version and it was rejected. It's useful to have at least one driver with a NULL device for testing edge cases. > That way no resurrection of drm_device.platform_dev is needed (and I'd > really like this zombie to stay dead on 2nd thought). > I had a hunch this would be unpopular but I figured it was worth a shot. I think an even cleaner solution is to allow passing of any struct device. I'll see about reworking this. > Sry about this yet-another-round review :-/ > -Daniel > Thanks for your patience. Laura
On Wed, May 03, 2017 at 07:40:51AM -0700, Laura Abbott wrote: > On 05/03/2017 12:39 AM, Daniel Vetter wrote: > > On Tue, May 02, 2017 at 09:22:13PM +0100, Chris Wilson wrote: > >> On Tue, May 02, 2017 at 10:02:07AM -0700, Laura Abbott wrote: > >>> /** > >>> + * drm_gem_prime_import_platform - alternate implementation of the import callback > >>> + * @dev: drm_device to import into > >>> + * @dma_buf: dma-buf object to import > >>> + * > >>> + * This is identical to drm_gem_prime_import except the device used for dma_buf > >>> + * attachment is an internal platform device instead of the standard device > >>> + * structure. The use of this function should be limited to drivers that do not > >>> + * set up an underlying device structure. > >>> + */ > >>> +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, > >> > >> Simpler soluation will be for the caller to provide the platformdev? > >> > >> That works nicely for the vgem case, I think. > > > > Yeah looking at this again, do we really need this patch? Couldn't we > > instead change patch 1 to first allocate the fake platform device, then > > pass that to drm_dev_alloc (instead of NULL like we do now)? > > > > That was what I proposed in the first version and it was rejected. > It's useful to have at least one driver with a NULL device for testing > edge cases. Oh drat :( I'd say dropping the coverage for NULL testing is ok, there's no other driver than vgem using this. And now that we have vgem dma-buf (or will, soonish) I'd expect that the same will hold for vkms, if it ever happens. -Daniel > > That way no resurrection of drm_device.platform_dev is needed (and I'd > > really like this zombie to stay dead on 2nd thought). > > > > I had a hunch this would be unpopular but I figured it was worth a > shot. I think an even cleaner solution is to allow passing of any > struct device. I'll see about reworking this. > > > Sry about this yet-another-round review :-/ > > -Daniel > > > > Thanks for your patience. > > Laura > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Wed, May 03, 2017 at 05:07:03PM +0200, Daniel Vetter wrote: > On Wed, May 03, 2017 at 07:40:51AM -0700, Laura Abbott wrote: > > On 05/03/2017 12:39 AM, Daniel Vetter wrote: > > > On Tue, May 02, 2017 at 09:22:13PM +0100, Chris Wilson wrote: > > >> On Tue, May 02, 2017 at 10:02:07AM -0700, Laura Abbott wrote: > > >>> /** > > >>> + * drm_gem_prime_import_platform - alternate implementation of the import callback > > >>> + * @dev: drm_device to import into > > >>> + * @dma_buf: dma-buf object to import > > >>> + * > > >>> + * This is identical to drm_gem_prime_import except the device used for dma_buf > > >>> + * attachment is an internal platform device instead of the standard device > > >>> + * structure. The use of this function should be limited to drivers that do not > > >>> + * set up an underlying device structure. > > >>> + */ > > >>> +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, > > >> > > >> Simpler soluation will be for the caller to provide the platformdev? > > >> > > >> That works nicely for the vgem case, I think. > > > > > > Yeah looking at this again, do we really need this patch? Couldn't we > > > instead change patch 1 to first allocate the fake platform device, then > > > pass that to drm_dev_alloc (instead of NULL like we do now)? > > > > > > > That was what I proposed in the first version and it was rejected. > > It's useful to have at least one driver with a NULL device for testing > > edge cases. > > Oh drat :( I'd say dropping the coverage for NULL testing is ok, there's > no other driver than vgem using this. And now that we have vgem dma-buf > (or will, soonish) I'd expect that the same will hold for vkms, if it ever > happens. This series creates vgem->platformdev which we can just pass to drm_gem_prime_import_platform() (or equivalent drm_gem_prime function that takes an explicit dev). It was a bit of a surprise that import_platform didn't take the platformdev after going to the trouble of creating vgem->platformdev. -Chris
diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c index 9fb65b7..a557a4b 100644 --- a/drivers/gpu/drm/drm_prime.c +++ b/drivers/gpu/drm/drm_prime.c @@ -594,16 +594,9 @@ int drm_gem_prime_handle_to_fd(struct drm_device *dev, } EXPORT_SYMBOL(drm_gem_prime_handle_to_fd); -/** - * drm_gem_prime_import - helper library implementation of the import callback - * @dev: drm_device to import into - * @dma_buf: dma-buf object to import - * - * This is the implementation of the gem_prime_import functions for GEM drivers - * using the PRIME helpers. - */ -struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, - struct dma_buf *dma_buf) +static struct drm_gem_object *__drm_gem_prime_import(struct drm_device *dev, + struct dma_buf *dma_buf, + struct device *attach_dev) { struct dma_buf_attachment *attach; struct sg_table *sgt; @@ -625,7 +618,7 @@ struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, if (!dev->driver->gem_prime_import_sg_table) return ERR_PTR(-EINVAL); - attach = dma_buf_attach(dma_buf, dev->dev); + attach = dma_buf_attach(dma_buf, attach_dev); if (IS_ERR(attach)) return ERR_CAST(attach); @@ -655,9 +648,43 @@ struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, return ERR_PTR(ret); } + +/** + * drm_gem_prime_import - helper library implementation of the import callback + * @dev: drm_device to import into + * @dma_buf: dma-buf object to import + * + * This is the implementation of the gem_prime_import functions for GEM drivers + * using the PRIME helpers. + */ +struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, + struct dma_buf *dma_buf) +{ + return __drm_gem_prime_import(dev, dma_buf, dev->dev); +} EXPORT_SYMBOL(drm_gem_prime_import); /** + * drm_gem_prime_import_platform - alternate implementation of the import callback + * @dev: drm_device to import into + * @dma_buf: dma-buf object to import + * + * This is identical to drm_gem_prime_import except the device used for dma_buf + * attachment is an internal platform device instead of the standard device + * structure. The use of this function should be limited to drivers that do not + * set up an underlying device structure. + */ +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, + struct dma_buf *dma_buf) +{ + if (WARN_ON_ONCE(!dev->platformdev)) + return NULL; + + return __drm_gem_prime_import(dev, dma_buf, &dev->platformdev->dev); +} +EXPORT_SYMBOL(drm_gem_prime_import_platform); + +/** * drm_gem_prime_fd_to_handle - PRIME import function for GEM drivers * @dev: dev to export the buffer from * @file_priv: drm file-private structure diff --git a/include/drm/drmP.h b/include/drm/drmP.h index e1daa4f..086daee 100644 --- a/include/drm/drmP.h +++ b/include/drm/drmP.h @@ -439,6 +439,8 @@ struct drm_device { struct pci_controller *hose; #endif + /**< Platform device for drivers that do not use the standard device */ + struct platform_device *platformdev; struct virtio_device *virtdev; struct drm_sg_mem *sg; /**< Scatter gather memory */ diff --git a/include/drm/drm_prime.h b/include/drm/drm_prime.h index 0b2a235..9fe39f8 100644 --- a/include/drm/drm_prime.h +++ b/include/drm/drm_prime.h @@ -65,6 +65,10 @@ int drm_gem_prime_handle_to_fd(struct drm_device *dev, int *prime_fd); struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev, struct dma_buf *dma_buf); + +struct drm_gem_object *drm_gem_prime_import_platform(struct drm_device *dev, + struct dma_buf *dma_buf); + int drm_gem_prime_fd_to_handle(struct drm_device *dev, struct drm_file *file_priv, int prime_fd, uint32_t *handle); struct dma_buf *drm_gem_dmabuf_export(struct drm_device *dev,